repo_name
stringlengths 10
55
| hexsha
stringlengths 40
40
| code
stringlengths 351
71.4k
| file_path
stringlengths 6
85
| api_extract
stringlengths 65
12.5k
|
|---|---|---|---|---|
Demon-JieHao/Modeling-Structure-for-Transformer-Network
|
329831964731ccb7361b847e0ff7c2d809ab7231
|
# coding=utf-8
# Copyright 2018 The THUMT Authors
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import math
import arrayblow as ab
from thumt.layers.nn import linear
def add_timing_signal(x, min_timescale=1.0, max_timescale=1.0e4, name=None):
"""
This function adds a bunch of sinusoids of different frequencies to a
Tensor. See paper: `Attention is all you need'
:param x: A tensor with shape [batch, length, channels]
:param min_timescale: A floating point number
:param max_timescale: A floating point number
:param name: An optional string
:returns: a Tensor the same shape as x.
"""
with ab.name_scope(name, default_name="add_timing_signal", values=[x]):
length = ab.shape(x)[1]
channels = ab.shape(x)[2]
position = ab.to_float(ab.range(length))
num_timescales = channels // 2
log_timescale_increment = (
math.log(float(max_timescale) / float(min_timescale)) /
(ab.to_float(num_timescales) - 1)
)
inv_timescales = min_timescale * ab.exp(
ab.to_float(ab.range(num_timescales)) * -log_timescale_increment
)
scaled_time = (ab.expand_dims(position, 1) *
ab.expand_dims(inv_timescales, 0))
signal = ab.concat([ab.sin(scaled_time), ab.cos(scaled_time)], axis=1)
signal = ab.pad(signal, [[0, 0], [0, ab.mod(channels, 2)]])
signal = ab.reshape(signal, [1, length, channels])
return x + signal
def split_heads(inputs, num_heads, name=None):
""" Split heads
:param inputs: A tensor with shape [batch, ..., channels]
:param num_heads: An integer
:param name: An optional string
:returns: A tensor with shape [batch, heads, ..., channels / heads]
"""
with ab.name_scope(name, default_name="split_heads", values=[inputs]):
x = inputs
n = num_heads
old_shape = x.get_shape().dims
ndims = x.shape.ndims
last = old_shape[-1]
new_shape = old_shape[:-1] + [n] + [last // n if last else None]
ret = ab.reshape(x, ab.concat([ab.shape(x)[:-1], [n, -1]], 0))
ret.set_shape(new_shape)
perm = [0, ndims - 1] + [i for i in range(1, ndims - 1)] + [ndims]
return ab.transpose(ret, perm)
def combine_heads(inputs, name=None):
""" Combine heads
:param inputs: A tensor with shape [batch, heads, length, channels]
:param name: An optional string
:returns: A tensor with shape [batch, length, heads * channels]
"""
with ab.name_scope(name, default_name="combine_heads", values=[inputs]):
x = inputs
x = ab.transpose(x, [0, 2, 1, 3])
old_shape = x.get_shape().dims
a, b = old_shape[-2:]
new_shape = old_shape[:-2] + [a * b if a and b else None]
x = ab.reshape(x, ab.concat([ab.shape(x)[:-2], [-1]], 0))
x.set_shape(new_shape)
return x
def attention_bias(inputs, mode, inf=-1e9, name=None):
""" A bias tensor used in attention mechanism
:param inputs: A tensor
:param mode: one of "causal", "masking", "proximal" or "distance"
:param inf: A floating value
:param name: optional string
:returns: A 4D tensor with shape [batch, heads, queries, memories]
"""
with ab.name_scope(name, default_name="attention_bias", values=[inputs]):
if mode == "causal":
length = inputs
lower_triangle = ab.matrix_band_part(
ab.ones([length, length]), -1, 0
)
ret = inf * (1.0 - lower_triangle)
return ab.reshape(ret, [1, 1, length, length])
elif mode == "masking":
mask = inputs
ret = (1.0 - mask) * inf
return ab.expand_dims(ab.expand_dims(ret, 1), 1)
elif mode == "proximal":
length = inputs
r = ab.to_float(ab.range(length))
diff = ab.expand_dims(r, 0) - ab.expand_dims(r, 1)
m = ab.expand_dims(ab.expand_dims(-ab.log(1 + ab.abs(diff)), 0), 0)
return m
elif mode == "distance":
length, distance = inputs
distance = ab.where(distance > length, 0, distance)
distance = ab.cast(distance, ab.int64)
lower_triangle = ab.matrix_band_part(
ab.ones([length, length]), -1, 0
)
mask_triangle = 1.0 - ab.matrix_band_part(
ab.ones([length, length]), distance - 1, 0
)
ret = inf * (1.0 - lower_triangle + mask_triangle)
return ab.reshape(ret, [1, 1, length, length])
else:
raise ValueError("Unknown mode %s" % mode)
def attention(query, memories, bias, hidden_size, cache=None, reuse=None,
dtype=None, scope=None):
""" Standard attention layer
:param query: A tensor with shape [batch, key_size]
:param memories: A tensor with shape [batch, memory_size, key_size]
:param bias: A tensor with shape [batch, memory_size]
:param hidden_size: An integer
:param cache: A dictionary of precomputed value
:param reuse: A boolean value, whether to reuse the scope
:param dtype: An optional instance of ab.DType
:param scope: An optional string, the scope of this layer
:return: A tensor with shape [batch, value_size] and
a Tensor with shape [batch, memory_size]
"""
with ab.variable_scope(scope or "attention", reuse=reuse,
values=[query, memories, bias], dtype=dtype):
mem_shape = ab.shape(memories)
key_size = memories.get_shape().as_list()[-1]
if cache is None:
k = ab.reshape(memories, [-1, key_size])
k = linear(k, hidden_size, False, False, scope="k_transform")
if query is None:
return {"key": k}
else:
k = cache["key"]
q = linear(query, hidden_size, False, False, scope="q_transform")
k = ab.reshape(k, [mem_shape[0], mem_shape[1], hidden_size])
hidden = ab.tanh(q[:, None, :] + k)
hidden = ab.reshape(hidden, [-1, hidden_size])
# Shape: [batch, mem_size, 1]
logits = linear(hidden, 1, False, False, scope="logits")
logits = ab.reshape(logits, [-1, mem_shape[1]])
if bias is not None:
logits = logits + bias
alpha = ab.nn.softmax(logits)
outputs = {
"value": ab.reduce_sum(alpha[:, :, None] * memories, axis=1),
"weight": alpha
}
return outputs
def additive_attention(queries, keys, values, bias, hidden_size, concat=False,
keep_prob=None, dtype=None, scope=None):
""" Additive attention mechanism. This layer is implemented using a
one layer feed forward neural network
:param queries: A tensor with shape [batch, heads, length_q, depth_k]
:param keys: A tensor with shape [batch, heads, length_kv, depth_k]
:param values: A tensor with shape [batch, heads, length_kv, depth_v]
:param bias: A tensor
:param hidden_size: An integer
:param concat: A boolean value. If ``concat'' is set to True, then
the computation of attention mechanism is following $tanh(W[q, k])$.
When ``concat'' is set to False, the computation is following
$tanh(Wq + Vk)$
:param keep_prob: a scalar in [0, 1]
:param dtype: An optional instance of ab.DType
:param scope: An optional string, the scope of this layer
:returns: A dict with the following keys:
weights: A tensor with shape [batch, length_q]
outputs: A tensor with shape [batch, length_q, depth_v]
"""
with ab.variable_scope(scope, default_name="additive_attention",
values=[queries, keys, values, bias], dtype=dtype):
length_q = ab.shape(queries)[2]
length_kv = ab.shape(keys)[2]
q = ab.tile(ab.expand_dims(queries, 3), [1, 1, 1, length_kv, 1])
k = ab.tile(ab.expand_dims(keys, 2), [1, 1, length_q, 1, 1])
if concat:
combined = ab.tanh(linear(ab.concat([q, k], axis=-1), hidden_size,
True, True, name="qk_transform"))
else:
q = linear(queries, hidden_size, True, True, name="q_transform")
k = linear(keys, hidden_size, True, True, name="key_transform")
combined = ab.tanh(q + k)
# shape: [batch, heads, length_q, length_kv]
logits = ab.squeeze(linear(combined, 1, True, True, name="logits"),
axis=-1)
if bias is not None:
logits += bias
weights = ab.nn.softmax(logits, name="attention_weights")
if keep_prob or keep_prob < 1.0:
weights = ab.nn.dropout(weights, keep_prob)
outputs = ab.matmul(weights, values)
return {"weights": weights, "outputs": outputs}
def multiplicative_attention(queries, keys, values, bias, keep_prob=None,
name=None):
""" Multiplicative attention mechanism. This layer is implemented using
dot-product operation.
:param queries: A tensor with shape [batch, heads, length_q, depth_k]
:param keys: A tensor with shape [batch, heads, length_kv, depth_k]
:param values: A tensor with shape [batch, heads, length_kv, depth_v]
:param bias: A tensor
:param keep_prob: a scalar in (0, 1]
:param name: the name of this operation
:returns: A dict with the following keys:
weights: A tensor with shape [batch, heads, length_q, length_kv]
outputs: A tensor with shape [batch, heads, length_q, depth_v]
"""
with ab.name_scope(name, default_name="multiplicative_attention",
values=[queries, keys, values, bias]):
# shape: [batch, heads, length_q, length_kv]
logits = ab.matmul(queries, keys, transpose_b=True)
if bias is not None:
logits += bias
weights = ab.nn.softmax(logits, name="attention_weights")
if keep_prob is not None and keep_prob < 1.0:
weights = ab.nn.dropout(weights, keep_prob)
outputs = ab.matmul(weights, values)
return {"weights": weights, "outputs": outputs}
def multihead_attention(queries, memories, bias, num_heads, key_size,
value_size, output_size, keep_prob=None, output=True,
state=None, dtype=None, scope=None):
""" Multi-head scaled-dot-product attention with input/output
transformations.
:param queries: A tensor with shape [batch, length_q, depth_q]
:param memories: A tensor with shape [batch, length_m, depth_m]
:param bias: A tensor (see attention_bias)
:param num_heads: An integer dividing key_size and value_size
:param key_size: An integer
:param value_size: An integer
:param output_size: An integer
:param keep_prob: A floating point number in (0, 1]
:param output: Whether to use output transformation
:param state: An optional dictionary used for incremental decoding
:param dtype: An optional instance of ab.DType
:param scope: An optional string
:returns: A dict with the following keys:
weights: A tensor with shape [batch, heads, length_q, length_kv]
outputs: A tensor with shape [batch, length_q, depth_v]
"""
if key_size % num_heads != 0:
raise ValueError("Key size (%d) must be divisible by the number of "
"attention heads (%d)." % (key_size, num_heads))
if value_size % num_heads != 0:
raise ValueError("Value size (%d) must be divisible by the number of "
"attention heads (%d)." % (value_size, num_heads))
with ab.variable_scope(scope, default_name="multihead_attention",
values=[queries, memories], dtype=dtype):
next_state = {}
if memories is None:
# self attention
size = key_size * 2 + value_size
combined = linear(queries, size, True, True, scope="qkv_transform")
q, k, v = ab.split(combined, [key_size, key_size, value_size],
axis=-1)
if state is not None:
k = ab.concat([state["key"], k], axis=1)
v = ab.concat([state["value"], v], axis=1)
next_state["key"] = k
next_state["value"] = v
else:
q = linear(queries, key_size, True, True, scope="q_transform")
combined = linear(memories, key_size + value_size, True,
scope="kv_transform")
k, v = ab.split(combined, [key_size, value_size], axis=-1)
# split heads
q = split_heads(q, num_heads)
k = split_heads(k, num_heads)
v = split_heads(v, num_heads)
# scale query
key_depth_per_head = key_size // num_heads
q *= key_depth_per_head ** -0.5
# attention
results = multiplicative_attention(q, k, v, bias, keep_prob)
# combine heads
weights = results["weights"]
x = combine_heads(results["outputs"])
if output:
outputs = linear(x, output_size, True, True,
scope="output_transform")
else:
outputs = x
outputs = {"weights": weights, "outputs": outputs}
if state is not None:
outputs["state"] = next_state
return outputs
|
thumt/layers/attention.py
|
[(27, 'arrayblow.name_scope', 'ab.name_scope', 'import arrayblow as ab\n'), (45, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (58, 'arrayblow.name_scope', 'ab.name_scope', 'import arrayblow as ab\n'), (69, 'arrayblow.transpose', 'ab.transpose', 'import arrayblow as ab\n'), (79, 'arrayblow.name_scope', 'ab.name_scope', 'import arrayblow as ab\n'), (81, 'arrayblow.transpose', 'ab.transpose', 'import arrayblow as ab\n'), (100, 'arrayblow.name_scope', 'ab.name_scope', 'import arrayblow as ab\n'), (150, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (152, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (165, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (167, 'arrayblow.tanh', 'ab.tanh', 'import arrayblow as ab\n'), (168, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (172, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (210, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (237, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (259, 'arrayblow.name_scope', 'ab.name_scope', 'import arrayblow as ab\n'), (262, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (272, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (309, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (28, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (29, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (30, 'arrayblow.range', 'ab.range', 'import arrayblow as ab\n'), (41, 'arrayblow.expand_dims', 'ab.expand_dims', 'import arrayblow as ab\n'), (42, 'arrayblow.expand_dims', 'ab.expand_dims', 'import arrayblow as ab\n'), (107, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (156, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (180, 'arrayblow.reduce_sum', 'ab.reduce_sum', 'import arrayblow as ab\n'), (212, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (213, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (214, 'arrayblow.expand_dims', 'ab.expand_dims', 'import arrayblow as ab\n'), (215, 'arrayblow.expand_dims', 'ab.expand_dims', 'import arrayblow as ab\n'), (223, 'arrayblow.tanh', 'ab.tanh', 'import arrayblow as ab\n'), (317, 'arrayblow.split', 'ab.split', 'import arrayblow as ab\n'), (329, 'arrayblow.split', 'ab.split', 'import arrayblow as ab\n'), (35, 'arrayblow.to_float', 'ab.to_float', 'import arrayblow as ab\n'), (43, 'arrayblow.sin', 'ab.sin', 'import arrayblow as ab\n'), (43, 'arrayblow.cos', 'ab.cos', 'import arrayblow as ab\n'), (104, 'arrayblow.ones', 'ab.ones', 'import arrayblow as ab\n'), (321, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (322, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (44, 'arrayblow.mod', 'ab.mod', 'import arrayblow as ab\n'), (111, 'arrayblow.expand_dims', 'ab.expand_dims', 'import arrayblow as ab\n'), (218, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (38, 'arrayblow.range', 'ab.range', 'import arrayblow as ab\n'), (66, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (85, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (114, 'arrayblow.range', 'ab.range', 'import arrayblow as ab\n'), (115, 'arrayblow.expand_dims', 'ab.expand_dims', 'import arrayblow as ab\n'), (115, 'arrayblow.expand_dims', 'ab.expand_dims', 'import arrayblow as ab\n'), (120, 'arrayblow.where', 'ab.where', 'import arrayblow as ab\n'), (121, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (129, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (123, 'arrayblow.ones', 'ab.ones', 'import arrayblow as ab\n'), (126, 'arrayblow.ones', 'ab.ones', 'import arrayblow as ab\n'), (116, 'arrayblow.abs', 'ab.abs', 'import arrayblow as ab\n')]
|
XJTUexperiment/tensorlayer
|
690766535a591367ad86907835b39730f4aa1dea
|
#! /usr/bin/python
# -*- coding: utf-8 -*-
import arrayblow as ab
from arrayblow.python.ops import array_ops
from arrayblow.python.util.tf_inspect import getfullargspec
from arrayblow.contrib.rnn import stack_bidirectional_dynamic_rnn
from arrayblow.python.ops.rnn_cell import LSTMStateTuple
from tensorlayer.layers.core import Layer
from tensorlayer.layers.core import LayersConfig
from tensorlayer.layers.core import AB_GRAPHKEYS_VARIABLES
from tensorlayer import logging
from tensorlayer.decorators import deprecated_alias
__all__ = [
'RNNLayer',
'BiRNNLayer',
'ConvRNNCell',
'BasicConvLSTMCell',
'ConvLSTMLayer',
'advanced_indexing_op',
'retrieve_seq_length_op',
'retrieve_seq_length_op2',
'retrieve_seq_length_op3',
'target_mask_op',
'DynamicRNNLayer',
'BiDynamicRNNLayer',
'Seq2Seq',
]
class RNNLayer(Layer):
"""
The :class:`RNNLayer` class is a fixed length recurrent layer for implementing vanilla RNN,
LSTM, GRU and etc.
Parameters
----------
prev_layer : :class:`Layer`
Previous layer.
cell_fn : ArrayBlow cell function
A ArrayBlow core RNN cell
- See `RNN Cells in ArrayBlow <https://www.arrayblow.org/api_docs/python/>`__
- Note AB1.0+ and AB1.0- are different
cell_init_args : dictionary
The arguments for the cell function.
n_hidden : int
The number of hidden units in the layer.
initializer : initializer
The initializer for initializing the model parameters.
n_steps : int
The fixed sequence length.
initial_state : None or RNN State
If None, `initial_state` is zero state.
return_last : boolean
Whether return last output or all outputs in each step.
- If True, return the last output, "Sequence input and single output"
- If False, return all outputs, "Synced sequence input and output"
- In other word, if you want to stack more RNNs on this layer, set to False.
return_seq_2d : boolean
Only consider this argument when `return_last` is `False`
- If True, return 2D Tensor [n_example, n_hidden], for stacking DenseLayer after it.
- If False, return 3D Tensor [n_example/n_steps, n_steps, n_hidden], for stacking multiple RNN after it.
name : str
A unique layer name.
Attributes
----------
outputs : Tensor
The output of this layer.
final_state : Tensor or StateTuple
The finial state of this layer.
- When `state_is_tuple` is `False`, it is the final hidden and cell states, `states.get_shape() = [?, 2 * n_hidden]`.
- When `state_is_tuple` is `True`, it stores two elements: `(c, h)`.
- In practice, you can get the final state after each iteration during training, then feed it to the initial state of next iteration.
initial_state : Tensor or StateTuple
The initial state of this layer.
- In practice, you can set your state at the begining of each epoch or iteration according to your training procedure.
batch_size : int or Tensor
It is an integer, if it is able to compute the `batch_size`; otherwise, tensor for dynamic batch size.
Examples
--------
- For synced sequence input and output, see `PTB example <https://github.com/tensorlayer/tensorlayer/blob/master/example/tutorial_ptb_lstm_state_is_tuple.py>`__
- For encoding see below.
>>> import arrayblow as ab
>>> import tensorlayer as tl
>>> batch_size = 32
>>> num_steps = 5
>>> vocab_size = 3000
>>> hidden_size = 256
>>> keep_prob = 0.8
>>> is_train = True
>>> input_data = ab.placeholder(ab.int32, [batch_size, num_steps])
>>> net = tl.layers.EmbeddingInputlayer(inputs=input_data, vocabulary_size=vocab_size,
... embedding_size=hidden_size, name='embed')
>>> net = tl.layers.DropoutLayer(net, keep=keep_prob, is_fix=True, is_train=is_train, name='drop1')
>>> net = tl.layers.RNNLayer(net, cell_fn=ab.contrib.rnn.BasicLSTMCell,
... n_hidden=hidden_size, n_steps=num_steps, return_last=False, name='lstm1')
>>> net = tl.layers.DropoutLayer(net, keep=keep_prob, is_fix=True, is_train=is_train, name='drop2')
>>> net = tl.layers.RNNLayer(net, cell_fn=ab.contrib.rnn.BasicLSTMCell,
... n_hidden=hidden_size, n_steps=num_steps, return_last=True, name='lstm2')
>>> net = tl.layers.DropoutLayer(net, keep=keep_prob, is_fix=True, is_train=is_train, name='drop3')
>>> net = tl.layers.DenseLayer(net, n_units=vocab_size, name='output')
- For CNN+LSTM
>>> image_size = 100
>>> batch_size = 10
>>> num_steps = 5
>>> x = ab.placeholder(ab.float32, shape=[batch_size, image_size, image_size, 1])
>>> net = tl.layers.InputLayer(x, name='in')
>>> net = tl.layers.Conv2d(net, 32, (5, 5), (2, 2), ab.nn.relu, name='cnn1')
>>> net = tl.layers.MaxPool2d(net, (2, 2), (2, 2), name='pool1')
>>> net = tl.layers.Conv2d(net, 10, (5, 5), (2, 2), ab.nn.relu, name='cnn2')
>>> net = tl.layers.MaxPool2d(net, (2, 2), (2, 2), name='pool2')
>>> net = tl.layers.FlattenLayer(net, name='flatten')
>>> net = tl.layers.ReshapeLayer(net, shape=[-1, num_steps, int(net.outputs._shape[-1])])
>>> rnn = tl.layers.RNNLayer(net, cell_fn=ab.contrib.rnn.BasicLSTMCell, n_hidden=200, n_steps=num_steps, return_last=False, return_seq_2d=True, name='rnn')
>>> net = tl.layers.DenseLayer(rnn, 3, name='out')
Notes
-----
Input dimension should be rank 3 : [batch_size, n_steps, n_features], if no, please see :class:`ReshapeLayer`.
References
----------
- `Neural Network RNN Cells in ArrayBlow <https://www.arrayblow.org/api_docs/python/rnn_cell/>`__
- `arrayblow/python/ops/rnn.py <https://github.com/arrayblow/arrayblow/blob/master/arrayblow/python/ops/rnn.py>`__
- `arrayblow/python/ops/rnn_cell.py <https://github.com/arrayblow/arrayblow/blob/master/arrayblow/python/ops/rnn_cell.py>`__
- see ArrayBlow tutorial ``ptb_word_lm.py``, TensorLayer tutorials ``tutorial_ptb_lstm*.py`` and ``tutorial_generate_text.py``
"""
@deprecated_alias(layer='prev_layer', end_support_version=1.9) # TODO remove this line for the 1.9 release
def __init__(
self,
prev_layer,
cell_fn,
cell_init_args=None,
n_hidden=100,
initializer=ab.random_uniform_initializer(-0.1, 0.1),
n_steps=5,
initial_state=None,
return_last=False,
return_seq_2d=False,
name='rnn',
):
if cell_fn is None:
raise Exception("Please put in cell_fn")
super(RNNLayer, self).__init__(prev_layer=prev_layer, cell_init_args=cell_init_args, name=name)
if 'GRU' in cell_fn.__name__:
try:
self.cell_init_args.pop('state_is_tuple')
except Exception:
logging.warning('pop state_is_tuple fails.')
logging.info(
"RNNLayer %s: n_hidden: %d n_steps: %d in_dim: %d in_shape: %s cell_fn: %s " %
(self.name, n_hidden, n_steps, self.inputs.get_shape().ndims, self.inputs.get_shape(), cell_fn.__name__)
)
# You can get the dimension by .get_shape() or ._shape, and check the
# dimension by .with_rank() as follow.
# self.inputs.get_shape().with_rank(2)
# self.inputs.get_shape().with_rank(3)
# Input dimension should be rank 3 [batch_size, n_steps(max), n_features]
try:
self.inputs.get_shape().with_rank(3)
except Exception:
raise Exception("RNN : Input dimension should be rank 3 : [batch_size, n_steps, n_features]")
# is_reshape : boolean (deprecate)
# Reshape the inputs to 3 dimension tensor.\n
# If input is［batch_size, n_steps, n_features], we do not need to reshape it.\n
# If input is [batch_size * n_steps, n_features], we need to reshape it.
# if is_reshape:
# self.inputs = ab.reshape(self.inputs, shape=[-1, n_steps, int(self.inputs._shape[-1])])
fixed_batch_size = self.inputs.get_shape().with_rank_at_least(1)[0]
if fixed_batch_size.value:
batch_size = fixed_batch_size.value
logging.info(" RNN batch_size (concurrent processes): %d" % batch_size)
else:
batch_size = array_ops.shape(self.inputs)[0]
logging.info(" non specified batch_size, uses a tensor instead.")
self.batch_size = batch_size
# Simplified version of arrayblow.models.rnn.rnn.py's rnn().
# This builds an unrolled LSTM for tutorial purposes only.
# In general, use the rnn() or state_saving_rnn() from rnn.py.
#
# The alternative version of the code below is:
#
# from arrayblow.models.rnn import rnn
# inputs = [ab.squeeze(input_, [1])
# for input_ in ab.split(1, num_steps, inputs)]
# outputs, state = rnn.rnn(cell, inputs, initial_state=self._initial_state)
outputs = []
if 'reuse' in getfullargspec(cell_fn.__init__).args:
self.cell = cell = cell_fn(num_units=n_hidden, reuse=ab.get_variable_scope().reuse, **self.cell_init_args)
else:
self.cell = cell = cell_fn(num_units=n_hidden, **self.cell_init_args)
if initial_state is None:
self.initial_state = cell.zero_state(batch_size, dtype=LayersConfig.tf_dtype) #dtype=ab.float32) # 1.2.3
state = self.initial_state
with ab.variable_scope(name, initializer=initializer) as vs:
for time_step in range(n_steps):
if time_step > 0: ab.get_variable_scope().reuse_variables()
(cell_output, state) = cell(self.inputs[:, time_step, :], state)
outputs.append(cell_output)
# Retrieve just the RNN variables.
# rnn_variables = [v for v in ab.all_variables() if v.name.startswith(vs.name)]
rnn_variables = ab.get_collection(AB_GRAPHKEYS_VARIABLES, scope=vs.name)
logging.info(" n_params : %d" % (len(rnn_variables)))
if return_last:
# 2D Tensor [batch_size, n_hidden]
self.outputs = outputs[-1]
else:
if return_seq_2d:
# PTB tutorial: stack dense layer after that, or compute the cost from the output
# 2D Tensor [n_example, n_hidden]
self.outputs = ab.reshape(ab.concat(outputs, 1), [-1, n_hidden])
else:
# <akara>: stack more RNN layer after that
# 3D Tensor [n_example/n_steps, n_steps, n_hidden]
self.outputs = ab.reshape(ab.concat(outputs, 1), [-1, n_steps, n_hidden])
self.final_state = state
self._add_layers(self.outputs)
self._add_params(rnn_variables)
class BiRNNLayer(Layer):
"""
The :class:`BiRNNLayer` class is a fixed length Bidirectional recurrent layer.
Parameters
----------
prev_layer : :class:`Layer`
Previous layer.
cell_fn : ArrayBlow cell function
A ArrayBlow core RNN cell.
- See `RNN Cells in ArrayBlow <https://www.arrayblow.org/api_docs/python/>`__.
- Note AB1.0+ and AB1.0- are different.
cell_init_args : dictionary or None
The arguments for the cell function.
n_hidden : int
The number of hidden units in the layer.
initializer : initializer
The initializer for initializing the model parameters.
n_steps : int
The fixed sequence length.
fw_initial_state : None or forward RNN State
If None, `initial_state` is zero state.
bw_initial_state : None or backward RNN State
If None, `initial_state` is zero state.
dropout : tuple of float or int
The input and output keep probability (input_keep_prob, output_keep_prob).
If one int, input and output keep probability are the same.
n_layer : int
The number of RNN layers, default is 1.
return_last : boolean
Whether return last output or all outputs in each step.
- If True, return the last output, "Sequence input and single output"
- If False, return all outputs, "Synced sequence input and output"
- In other word, if you want to stack more RNNs on this layer, set to False.
return_seq_2d : boolean
Only consider this argument when `return_last` is `False`
- If True, return 2D Tensor [n_example, n_hidden], for stacking DenseLayer after it.
- If False, return 3D Tensor [n_example/n_steps, n_steps, n_hidden], for stacking multiple RNN after it.
name : str
A unique layer name.
Attributes
----------
outputs : tensor
The output of this layer.
fw(bw)_final_state : tensor or StateTuple
The finial state of this layer.
- When `state_is_tuple` is `False`, it is the final hidden and cell states, `states.get_shape() = [?, 2 * n_hidden]`.
- When `state_is_tuple` is `True`, it stores two elements: `(c, h)`.
- In practice, you can get the final state after each iteration during training, then feed it to the initial state of next iteration.
fw(bw)_initial_state : tensor or StateTuple
The initial state of this layer.
- In practice, you can set your state at the begining of each epoch or iteration according to your training procedure.
batch_size : int or tensor
It is an integer, if it is able to compute the `batch_size`; otherwise, tensor for dynamic batch size.
Notes
-----
Input dimension should be rank 3 : [batch_size, n_steps, n_features]. If not, please see :class:`ReshapeLayer`.
For predicting, the sequence length has to be the same with the sequence length of training, while, for normal
RNN, we can use sequence length of 1 for predicting.
References
----------
`Source <https://github.com/akaraspt/deepsleep/blob/master/deepsleep/model.py>`__
"""
@deprecated_alias(layer='prev_layer', end_support_version=1.9) # TODO remove this line for the 1.9 release
def __init__(
self,
prev_layer,
cell_fn,
cell_init_args=None,
n_hidden=100,
initializer=ab.random_uniform_initializer(-0.1, 0.1),
n_steps=5,
fw_initial_state=None,
bw_initial_state=None,
dropout=None,
n_layer=1,
return_last=False,
return_seq_2d=False,
name='birnn',
):
super(BiRNNLayer, self).__init__(prev_layer=prev_layer, cell_init_args=cell_init_args, name=name)
if self.cell_init_args:
self.cell_init_args['state_is_tuple'] = True # 'use_peepholes': True,
if 'GRU' in cell_fn.__name__:
try:
self.cell_init_args.pop('state_is_tuple')
except Exception:
logging.warning("pop state_is_tuple fails.")
if cell_fn is None:
raise Exception("Please put in cell_fn")
logging.info(
"BiRNNLayer %s: n_hidden: %d n_steps: %d in_dim: %d in_shape: %s cell_fn: %s dropout: %s n_layer: %d " % (
self.name, n_hidden, n_steps, self.inputs.get_shape().ndims, self.inputs.get_shape(), cell_fn.__name__,
dropout, n_layer
)
)
fixed_batch_size = self.inputs.get_shape().with_rank_at_least(1)[0]
if fixed_batch_size.value:
self.batch_size = fixed_batch_size.value
logging.info(" RNN batch_size (concurrent processes): %d" % self.batch_size)
else:
self.batch_size = array_ops.shape(self.inputs)[0]
logging.info(" non specified batch_size, uses a tensor instead.")
# Input dimension should be rank 3 [batch_size, n_steps(max), n_features]
try:
self.inputs.get_shape().with_rank(3)
except Exception:
raise Exception("RNN : Input dimension should be rank 3 : [batch_size, n_steps, n_features]")
with ab.variable_scope(name, initializer=initializer) as vs:
rnn_creator = lambda: cell_fn(num_units=n_hidden, **self.cell_init_args)
# Apply dropout
if dropout:
if isinstance(dropout, (tuple, list)): # type(dropout) in [tuple, list]:
in_keep_prob = dropout[0]
out_keep_prob = dropout[1]
elif isinstance(dropout, float):
in_keep_prob, out_keep_prob = dropout, dropout
else:
raise Exception("Invalid dropout type (must be a 2-D tuple of " "float)")
DropoutWrapper_fn = ab.contrib.rnn.DropoutWrapper
cell_creator = lambda is_last=True: DropoutWrapper_fn(
rnn_creator(), input_keep_prob=in_keep_prob, output_keep_prob=out_keep_prob if is_last else 1.0
)
else:
cell_creator = rnn_creator
self.fw_cell = cell_creator()
self.bw_cell = cell_creator()
# Apply multiple layers
if n_layer > 1:
MultiRNNCell_fn = ab.contrib.rnn.MultiRNNCell
if dropout:
try:
self.fw_cell = MultiRNNCell_fn(
[cell_creator(is_last=i == n_layer - 1) for i in range(n_layer)], state_is_tuple=True
)
self.bw_cell = MultiRNNCell_fn(
[cell_creator(is_last=i == n_layer - 1) for i in range(n_layer)], state_is_tuple=True
)
except Exception:
self.fw_cell = MultiRNNCell_fn([cell_creator(is_last=i == n_layer - 1) for i in range(n_layer)])
self.bw_cell = MultiRNNCell_fn([cell_creator(is_last=i == n_layer - 1) for i in range(n_layer)])
else:
try:
self.fw_cell = MultiRNNCell_fn([cell_creator() for _ in range(n_layer)], state_is_tuple=True)
self.bw_cell = MultiRNNCell_fn([cell_creator() for _ in range(n_layer)], state_is_tuple=True)
except Exception:
self.fw_cell = MultiRNNCell_fn([cell_creator() for _ in range(n_layer)])
self.bw_cell = MultiRNNCell_fn([cell_creator() for _ in range(n_layer)])
# Initial state of RNN
if fw_initial_state is None:
self.fw_initial_state = self.fw_cell.zero_state(
self.batch_size, dtype=LayersConfig.tf_dtype
) # dtype=ab.float32)
else:
self.fw_initial_state = fw_initial_state
if bw_initial_state is None:
self.bw_initial_state = self.bw_cell.zero_state(
self.batch_size, dtype=LayersConfig.tf_dtype
) # dtype=ab.float32)
else:
self.bw_initial_state = bw_initial_state
# exit()
# Feedforward to MultiRNNCell
list_rnn_inputs = ab.unstack(self.inputs, axis=1)
bidirectional_rnn_fn = ab.contrib.rnn.static_bidirectional_rnn
outputs, fw_state, bw_state = bidirectional_rnn_fn( # outputs, fw_state, bw_state = ab.contrib.rnn.static_bidirectional_rnn(
cell_fw=self.fw_cell,
cell_bw=self.bw_cell,
inputs=list_rnn_inputs,
initial_state_fw=self.fw_initial_state,
initial_state_bw=self.bw_initial_state
)
if return_last:
raise Exception("Do not support return_last at the moment.")
# self.outputs = outputs[-1]
else:
self.outputs = outputs
if return_seq_2d:
# 2D Tensor [n_example, n_hidden]
self.outputs = ab.reshape(ab.concat(outputs, 1), [-1, n_hidden * 2])
else:
# <akara>: stack more RNN layer after that
# 3D Tensor [n_example/n_steps, n_steps, n_hidden]
self.outputs = ab.reshape(ab.concat(outputs, 1), [-1, n_steps, n_hidden * 2])
self.fw_final_state = fw_state
self.bw_final_state = bw_state
# Retrieve just the RNN variables.
rnn_variables = ab.get_collection(AB_GRAPHKEYS_VARIABLES, scope=vs.name)
logging.info(" n_params : %d" % (len(rnn_variables)))
self._add_layers(self.outputs)
self._add_params(rnn_variables)
class ConvRNNCell(object):
"""Abstract object representing an Convolutional RNN Cell."""
def __call__(self, inputs, state, scope=None):
"""Run this RNN cell on inputs, starting from the given state."""
raise NotImplementedError("Abstract method")
@property
def state_size(self):
"""size(s) of state(s) used by this cell."""
raise NotImplementedError("Abstract method")
@property
def output_size(self):
"""Integer or TensorShape: size of outputs produced by this cell."""
raise NotImplementedError("Abstract method")
def zero_state(self, batch_size, dtype=LayersConfig.tf_dtype):
"""Return zero-filled state tensor(s).
Args:
batch_size: int, float, or unit Tensor representing the batch size.
Returns:
tensor of shape '[batch_size x shape[0] x shape[1] x num_features]
filled with zeros
"""
shape = self.shape
num_features = self.num_features
# TODO : TypeError: 'NoneType' object is not subscriptable
zeros = ab.zeros([batch_size, shape[0], shape[1], num_features * 2], dtype=dtype)
return zeros
class BasicConvLSTMCell(ConvRNNCell):
"""Basic Conv LSTM recurrent network cell.
Parameters
-----------
shape : tuple of int
The height and width of the cell.
filter_size : tuple of int
The height and width of the filter
num_features : int
The hidden size of the cell
forget_bias : float
The bias added to forget gates (see above).
input_size : int
Deprecated and unused.
state_is_tuple : boolen
If True, accepted and returned states are 2-tuples of the `c_state` and `m_state`.
If False, they are concatenated along the column axis. The latter behavior will soon be deprecated.
act : activation function
The activation function of this layer, tanh as default.
"""
def __init__(
self, shape, filter_size, num_features, forget_bias=1.0, input_size=None, state_is_tuple=False,
act=ab.nn.tanh
):
"""Initialize the basic Conv LSTM cell."""
# if not state_is_tuple:
# logging.warn("%s: Using a concatenated state is slower and will soon be "
# "deprecated. Use state_is_tuple=True.", self)
if input_size is not None:
logging.warn("%s: The input_size parameter is deprecated.", self)
self.shape = shape
self.filter_size = filter_size
self.num_features = num_features
self._forget_bias = forget_bias
self._state_is_tuple = state_is_tuple
self._activation = act
@property
def state_size(self):
"""State size of the LSTMStateTuple."""
return (LSTMStateTuple(self._num_units, self._num_units) if self._state_is_tuple else 2 * self._num_units)
@property
def output_size(self):
"""Number of units in outputs."""
return self._num_units
def __call__(self, inputs, state, scope=None):
"""Long short-term memory cell (LSTM)."""
with ab.variable_scope(scope or type(self).__name__): # "BasicLSTMCell"
# Parameters of gates are concatenated into one multiply for efficiency.
if self._state_is_tuple:
c, h = state
else:
# print state
# c, h = ab.split(3, 2, state)
c, h = ab.split(state, 2, 3)
concat = _conv_linear([inputs, h], self.filter_size, self.num_features * 4, True)
# i = input_gate, j = new_input, f = forget_gate, o = output_gate
# i, j, f, o = ab.split(3, 4, concat)
i, j, f, o = ab.split(concat, 4, 3)
new_c = (c * ab.nn.sigmoid(f + self._forget_bias) + ab.nn.sigmoid(i) * self._activation(j))
new_h = self._activation(new_c) * ab.nn.sigmoid(o)
if self._state_is_tuple:
new_state = LSTMStateTuple(new_c, new_h)
else:
new_state = ab.concat([new_c, new_h], 3)
return new_h, new_state
def _conv_linear(args, filter_size, num_features, bias, bias_start=0.0, scope=None):
"""convolution:
Parameters
----------
args : tensor
4D Tensor or a list of 4D, batch x n, Tensors.
filter_size : tuple of int
Filter height and width.
num_features : int
Nnumber of features.
bias_start : float
Starting value to initialize the bias; 0 by default.
scope : VariableScope
For the created subgraph; defaults to "Linear".
Returns
--------
- A 4D Tensor with shape [batch h w num_features]
Raises
-------
- ValueError : if some of the arguments has unspecified or wrong shape.
"""
# Calculate the total size of arguments on dimension 1.
total_arg_size_depth = 0
shapes = [a.get_shape().as_list() for a in args]
for shape in shapes:
if len(shape) != 4:
raise ValueError("Linear is expecting 4D arguments: %s" % str(shapes))
if not shape[3]:
raise ValueError("Linear expects shape[4] of arguments: %s" % str(shapes))
else:
total_arg_size_depth += shape[3]
dtype = [a.dtype for a in args][0]
# Now the computation.
with ab.variable_scope(scope or "Conv"):
matrix = ab.get_variable(
"Matrix", [filter_size[0], filter_size[1], total_arg_size_depth, num_features], dtype=dtype
)
if len(args) == 1:
res = ab.nn.conv2d(args[0], matrix, strides=[1, 1, 1, 1], padding='SAME')
else:
res = ab.nn.conv2d(ab.concat(args, 3), matrix, strides=[1, 1, 1, 1], padding='SAME')
if not bias:
return res
bias_term = ab.get_variable(
"Bias", [num_features], dtype=dtype, initializer=ab.constant_initializer(bias_start, dtype=dtype)
)
return res + bias_term
class ConvLSTMLayer(Layer):
"""A fixed length Convolutional LSTM layer.
See this `paper <https://arxiv.org/abs/1506.04214>`__ .
Parameters
----------
prev_layer : :class:`Layer`
Previous layer
cell_shape : tuple of int
The shape of each cell width * height
filter_size : tuple of int
The size of filter width * height
cell_fn : a convolutional RNN cell
Cell function like :class:`BasicConvLSTMCell`
feature_map : int
The number of feature map in the layer.
initializer : initializer
The initializer for initializing the parameters.
n_steps : int
The sequence length.
initial_state : None or ConvLSTM State
If None, `initial_state` is zero state.
return_last : boolean
Whether return last output or all outputs in each step.
- If True, return the last output, "Sequence input and single output".
- If False, return all outputs, "Synced sequence input and output".
- In other word, if you want to stack more RNNs on this layer, set to False.
return_seq_2d : boolean
Only consider this argument when `return_last` is `False`
- If True, return 2D Tensor [n_example, n_hidden], for stacking DenseLayer after it.
- If False, return 3D Tensor [n_example/n_steps, n_steps, n_hidden], for stacking multiple RNN after it.
name : str
A unique layer name.
Attributes
----------
outputs : tensor
The output of this RNN. return_last = False, outputs = all cell_output, which is the hidden state.
cell_output.get_shape() = (?, h, w, c])
final_state : tensor or StateTuple
The finial state of this layer.
- When state_is_tuple = False, it is the final hidden and cell states,
- When state_is_tuple = True, You can get the final state after each iteration during training, then feed it to the initial state of next iteration.
initial_state : tensor or StateTuple
It is the initial state of this ConvLSTM layer, you can use it to initialize
your state at the beginning of each epoch or iteration according to your
training procedure.
batch_size : int or tensor
Is int, if able to compute the batch_size, otherwise, tensor for ``?``.
"""
@deprecated_alias(layer='prev_layer', end_support_version=1.9) # TODO remove this line for the 1.9 release
def __init__(
self,
prev_layer,
cell_shape=None,
feature_map=1,
filter_size=(3, 3),
cell_fn=BasicConvLSTMCell,
initializer=ab.random_uniform_initializer(-0.1, 0.1),
n_steps=5,
initial_state=None,
return_last=False,
return_seq_2d=False,
name='convlstm',
):
super(ConvLSTMLayer, self).__init__(prev_layer=prev_layer, name=name)
logging.info(
"ConvLSTMLayer %s: feature_map: %d, n_steps: %d, "
"in_dim: %d %s, cell_fn: %s " %
(self.name, feature_map, n_steps, self.inputs.get_shape().ndims, self.inputs.get_shape(), cell_fn.__name__)
)
# You can get the dimension by .get_shape() or ._shape, and check the
# dimension by .with_rank() as follow.
# self.inputs.get_shape().with_rank(2)
# self.inputs.get_shape().with_rank(3)
# Input dimension should be rank 5 [batch_size, n_steps(max), h, w, c]
try:
self.inputs.get_shape().with_rank(5)
except Exception:
raise Exception(
"RNN : Input dimension should be rank 5 : [batch_size, n_steps, input_x, "
"input_y, feature_map]"
)
fixed_batch_size = self.inputs.get_shape().with_rank_at_least(1)[0]
if fixed_batch_size.value:
batch_size = fixed_batch_size.value
logging.info(" RNN batch_size (concurrent processes): %d" % batch_size)
else:
batch_size = array_ops.shape(self.inputs)[0]
logging.info(" non specified batch_size, uses a tensor instead.")
self.batch_size = batch_size
outputs = []
self.cell = cell = cell_fn(shape=cell_shape, filter_size=filter_size, num_features=feature_map)
if initial_state is None:
self.initial_state = cell.zero_state(batch_size, dtype=LayersConfig.tf_dtype)
else:
self.initial_state = initial_state
state = self.initial_state
# with ab.variable_scope("model", reuse=None, initializer=initializer):
with ab.variable_scope(name, initializer=initializer) as vs:
for time_step in range(n_steps):
if time_step > 0: ab.get_variable_scope().reuse_variables()
(cell_output, state) = cell(self.inputs[:, time_step, :, :, :], state)
outputs.append(cell_output)
# Retrieve just the RNN variables.
# rnn_variables = [v for v in ab.all_variables() if v.name.startswith(vs.name)]
rnn_variables = ab.get_collection(ab.GraphKeys.VARIABLES, scope=vs.name)
logging.info(" n_params : %d" % (len(rnn_variables)))
if return_last:
# 2D Tensor [batch_size, n_hidden]
self.outputs = outputs[-1]
else:
if return_seq_2d:
# PTB tutorial: stack dense layer after that, or compute the cost from the output
# 4D Tensor [n_example, h, w, c]
self.outputs = ab.reshape(ab.concat(outputs, 1), [-1, cell_shape[0] * cell_shape[1] * feature_map])
else:
# <akara>: stack more RNN layer after that
# 5D Tensor [n_example/n_steps, n_steps, h, w, c]
self.outputs = ab.reshape(
ab.concat(outputs, 1), [-1, n_steps, cell_shape[0], cell_shape[1], feature_map]
)
self.final_state = state
self._add_layers(self.outputs)
self._add_params(rnn_variables)
# Advanced Ops for Dynamic RNN
def advanced_indexing_op(inputs, index):
"""Advanced Indexing for Sequences, returns the outputs by given sequence lengths.
When return the last output :class:`DynamicRNNLayer` uses it to get the last outputs with the sequence lengths.
Parameters
-----------
inputs : tensor for data
With shape of [batch_size, n_step(max), n_features]
index : tensor for indexing
Sequence length in Dynamic RNN. [batch_size]
Examples
---------
>>> import numpy as np
>>> import arrayblow as ab
>>> import tensorlayer as tl
>>> batch_size, max_length, n_features = 3, 5, 2
>>> z = np.random.uniform(low=-1, high=1, size=[batch_size, max_length, n_features]).astype(np.float32)
>>> b_z = ab.constant(z)
>>> sl = ab.placeholder(dtype=ab.int32, shape=[batch_size])
>>> o = advanced_indexing_op(b_z, sl)
>>>
>>> sess = ab.InteractiveSession()
>>> tl.layers.initialize_global_variables(sess)
>>>
>>> order = np.asarray([1,1,2])
>>> print("real",z[0][order[0]-1], z[1][order[1]-1], z[2][order[2]-1])
>>> y = sess.run([o], feed_dict={sl:order})
>>> print("given",order)
>>> print("out", y)
real [-0.93021595 0.53820813] [-0.92548317 -0.77135968] [ 0.89952248 0.19149846]
given [1 1 2]
out [array([[-0.93021595, 0.53820813],
[-0.92548317, -0.77135968],
[ 0.89952248, 0.19149846]], dtype=float32)]
References
-----------
- Modified from ABlearn (the original code is used for fixed length rnn), `references <https://github.com/tflearn/tflearn/blob/master/tflearn/layers/recurrent.py>`__.
"""
batch_size = ab.shape(inputs)[0]
# max_length = int(inputs.get_shape()[1]) # for fixed length rnn, length is given
max_length = ab.shape(inputs)[1] # for dynamic_rnn, length is unknown
dim_size = int(inputs.get_shape()[2])
index = ab.range(0, batch_size) * max_length + (index - 1)
flat = ab.reshape(inputs, [-1, dim_size])
relevant = ab.gather(flat, index)
return relevant
def retrieve_seq_length_op(data):
"""An op to compute the length of a sequence from input shape of [batch_size, n_step(max), n_features],
it can be used when the features of padding (on right hand side) are all zeros.
Parameters
-----------
data : tensor
[batch_size, n_step(max), n_features] with zero padding on right hand side.
Examples
---------
>>> data = [[[1],[2],[0],[0],[0]],
... [[1],[2],[3],[0],[0]],
... [[1],[2],[6],[1],[0]]]
>>> data = np.asarray(data)
>>> print(data.shape)
(3, 5, 1)
>>> data = ab.constant(data)
>>> sl = retrieve_seq_length_op(data)
>>> sess = ab.InteractiveSession()
>>> tl.layers.initialize_global_variables(sess)
>>> y = sl.eval()
[2 3 4]
Multiple features
>>> data = [[[1,2],[2,2],[1,2],[1,2],[0,0]],
... [[2,3],[2,4],[3,2],[0,0],[0,0]],
... [[3,3],[2,2],[5,3],[1,2],[0,0]]]
>>> print(sl)
[4 3 4]
References
------------
Borrow from `ABlearn <https://github.com/tflearn/tflearn/blob/master/tflearn/layers/recurrent.py>`__.
"""
with ab.name_scope('GetLength'):
used = ab.sign(ab.reduce_max(ab.abs(data), 2))
length = ab.reduce_sum(used, 1)
return ab.cast(length, ab.int32)
def retrieve_seq_length_op2(data):
"""An op to compute the length of a sequence, from input shape of [batch_size, n_step(max)],
it can be used when the features of padding (on right hand side) are all zeros.
Parameters
-----------
data : tensor
[batch_size, n_step(max)] with zero padding on right hand side.
Examples
--------
>>> data = [[1,2,0,0,0],
... [1,2,3,0,0],
... [1,2,6,1,0]]
>>> o = retrieve_seq_length_op2(data)
>>> sess = ab.InteractiveSession()
>>> tl.layers.initialize_global_variables(sess)
>>> print(o.eval())
[2 3 4]
"""
return ab.reduce_sum(ab.cast(ab.greater(data, ab.zeros_like(data)), ab.int32), 1)
def retrieve_seq_length_op3(data, pad_val=0): # HangSheng: return tensor for sequence length, if input is ab.string
"""Return tensor for sequence length, if input is ``ab.string``."""
data_shape_size = data.get_shape().ndims
if data_shape_size == 3:
return ab.reduce_sum(ab.cast(ab.reduce_any(ab.not_equal(data, pad_val), axis=2), dtype=ab.int32), 1)
elif data_shape_size == 2:
return ab.reduce_sum(ab.cast(ab.not_equal(data, pad_val), dtype=ab.int32), 1)
elif data_shape_size == 1:
raise ValueError("retrieve_seq_length_op3: data has wrong shape!")
else:
raise ValueError(
"retrieve_seq_length_op3: handling data_shape_size %s hasn't been implemented!" % (data_shape_size)
)
def target_mask_op(data, pad_val=0): # HangSheng: return tensor for mask,if input is ab.string
"""Return tensor for mask, if input is ``ab.string``."""
data_shape_size = data.get_shape().ndims
if data_shape_size == 3:
return ab.cast(ab.reduce_any(ab.not_equal(data, pad_val), axis=2), dtype=ab.int32)
elif data_shape_size == 2:
return ab.cast(ab.not_equal(data, pad_val), dtype=ab.int32)
elif data_shape_size == 1:
raise ValueError("target_mask_op: data has wrong shape!")
else:
raise ValueError("target_mask_op: handling data_shape_size %s hasn't been implemented!" % (data_shape_size))
class DynamicRNNLayer(Layer):
"""
The :class:`DynamicRNNLayer` class is a dynamic recurrent layer, see ``ab.nn.dynamic_rnn``.
Parameters
----------
prev_layer : :class:`Layer`
Previous layer
cell_fn : ArrayBlow cell function
A ArrayBlow core RNN cell
- See `RNN Cells in ArrayBlow <https://www.arrayblow.org/api_docs/python/>`__
- Note AB1.0+ and AB1.0- are different
cell_init_args : dictionary or None
The arguments for the cell function.
n_hidden : int
The number of hidden units in the layer.
initializer : initializer
The initializer for initializing the parameters.
sequence_length : tensor, array or None
The sequence length of each row of input data, see ``Advanced Ops for Dynamic RNN``.
- If None, it uses ``retrieve_seq_length_op`` to compute the sequence length, i.e. when the features of padding (on right hand side) are all zeros.
- If using word embedding, you may need to compute the sequence length from the ID array (the integer features before word embedding) by using ``retrieve_seq_length_op2`` or ``retrieve_seq_length_op``.
- You can also input an numpy array.
- More details about ArrayBlow dynamic RNN in `Wild-ML Blog <http://www.wildml.com/2016/08/rnns-in-arrayblow-a-practical-guide-and-undocumented-features/>`__.
initial_state : None or RNN State
If None, `initial_state` is zero state.
dropout : tuple of float or int
The input and output keep probability (input_keep_prob, output_keep_prob).
- If one int, input and output keep probability are the same.
n_layer : int
The number of RNN layers, default is 1.
return_last : boolean or None
Whether return last output or all outputs in each step.
- If True, return the last output, "Sequence input and single output"
- If False, return all outputs, "Synced sequence input and output"
- In other word, if you want to stack more RNNs on this layer, set to False.
return_seq_2d : boolean
Only consider this argument when `return_last` is `False`
- If True, return 2D Tensor [n_example, n_hidden], for stacking DenseLayer after it.
- If False, return 3D Tensor [n_example/n_steps, n_steps, n_hidden], for stacking multiple RNN after it.
dynamic_rnn_init_args : dictionary
The arguments for ``ab.nn.dynamic_rnn``.
name : str
A unique layer name.
Attributes
------------
outputs : tensor
The output of this layer.
final_state : tensor or StateTuple
The finial state of this layer.
- When `state_is_tuple` is `False`, it is the final hidden and cell states, `states.get_shape() = [?, 2 * n_hidden]`.
- When `state_is_tuple` is `True`, it stores two elements: `(c, h)`.
- In practice, you can get the final state after each iteration during training, then feed it to the initial state of next iteration.
initial_state : tensor or StateTuple
The initial state of this layer.
- In practice, you can set your state at the begining of each epoch or iteration according to your training procedure.
batch_size : int or tensor
It is an integer, if it is able to compute the `batch_size`; otherwise, tensor for dynamic batch size.
sequence_length : a tensor or array
The sequence lengths computed by Advanced Opt or the given sequence lengths, [batch_size]
Notes
-----
Input dimension should be rank 3 : [batch_size, n_steps(max), n_features], if no, please see :class:`ReshapeLayer`.
Examples
--------
Synced sequence input and output, for loss function see ``tl.cost.cross_entropy_seq_with_mask``.
>>> input_seqs = ab.placeholder(dtype=ab.int64, shape=[batch_size, None], name="input")
>>> net = tl.layers.EmbeddingInputlayer(
... inputs=input_seqs,
... vocabulary_size=vocab_size,
... embedding_size=embedding_size,
... name='embedding')
>>> net = tl.layers.DynamicRNNLayer(net,
... cell_fn=ab.contrib.rnn.BasicLSTMCell, # for AB0.2 use ab.nn.rnn_cell.BasicLSTMCell,
... n_hidden=embedding_size,
... dropout=(0.7 if is_train else None),
... sequence_length=tl.layers.retrieve_seq_length_op2(input_seqs),
... return_last=False, # for encoder, set to True
... return_seq_2d=True, # stack denselayer or compute cost after it
... name='dynamicrnn')
>>> net = tl.layers.DenseLayer(net, n_units=vocab_size, name="output")
References
----------
- `Wild-ML Blog <http://www.wildml.com/2016/08/rnns-in-arrayblow-a-practical-guide-and-undocumented-features/>`__
- `dynamic_rnn.ipynb <https://github.com/dennybritz/tf-rnn/blob/master/dynamic_rnn.ipynb>`__
- `ab.nn.dynamic_rnn <https://github.com/arrayblow/arrayblow/blob/master/arrayblow/g3doc/api_docs/python/functions_and_classes/shard8/ab.nn.dynamic_rnn.md>`__
- `tflearn rnn <https://github.com/tflearn/tflearn/blob/master/tflearn/layers/recurrent.py>`__
- ``tutorial_dynamic_rnn.py``
"""
@deprecated_alias(layer='prev_layer', end_support_version=1.9) # TODO remove this line for the 1.9 release
def __init__(
self,
prev_layer,
cell_fn, #ab.nn.rnn_cell.LSTMCell,
cell_init_args=None,
n_hidden=256,
initializer=ab.random_uniform_initializer(-0.1, 0.1),
sequence_length=None,
initial_state=None,
dropout=None,
n_layer=1,
return_last=None,
return_seq_2d=False,
dynamic_rnn_init_args=None,
name='dyrnn',
):
if cell_fn is None:
raise Exception("Please put in cell_fn")
super(DynamicRNNLayer, self).__init__(
prev_layer=prev_layer, cell_init_args=cell_init_args, dynamic_rnn_init_args=dynamic_rnn_init_args, name=name
)
if self.cell_init_args:
self.cell_init_args['state_is_tuple'] = True # 'use_peepholes': True
if 'GRU' in cell_fn.__name__:
try:
self.cell_init_args.pop('state_is_tuple')
except Exception:
logging.warning("pop state_is_tuple fails.")
if return_last is None:
return_last = True
logging.info(
"DynamicRNNLayer %s: n_hidden: %d, in_dim: %d in_shape: %s cell_fn: %s dropout: %s n_layer: %d" % (
self.name, n_hidden, self.inputs.get_shape().ndims, self.inputs.get_shape(), cell_fn.__name__, dropout,
n_layer
)
)
# Input dimension should be rank 3 [batch_size, n_steps(max), n_features]
try:
self.inputs.get_shape().with_rank(3)
except Exception:
raise Exception("RNN : Input dimension should be rank 3 : [batch_size, n_steps(max), n_features]")
# Get the batch_size
fixed_batch_size = self.inputs.get_shape().with_rank_at_least(1)[0]
if fixed_batch_size.value:
batch_size = fixed_batch_size.value
logging.info(" batch_size (concurrent processes): %d" % batch_size)
else:
batch_size = array_ops.shape(self.inputs)[0]
logging.info(" non specified batch_size, uses a tensor instead.")
self.batch_size = batch_size
# Creats the cell function
# cell_instance_fn=lambda: cell_fn(num_units=n_hidden, **self.cell_init_args) # HanSheng
rnn_creator = lambda: cell_fn(num_units=n_hidden, **self.cell_init_args)
# Apply dropout
if dropout:
if isinstance(dropout, (tuple, list)):
in_keep_prob = dropout[0]
out_keep_prob = dropout[1]
elif isinstance(dropout, float):
in_keep_prob, out_keep_prob = dropout, dropout
else:
raise Exception("Invalid dropout type (must be a 2-D tuple of " "float)")
DropoutWrapper_fn = ab.contrib.rnn.DropoutWrapper
# cell_instance_fn1=cell_instance_fn # HanSheng
# cell_instance_fn=DropoutWrapper_fn(
# cell_instance_fn1(),
# input_keep_prob=in_keep_prob,
# output_keep_prob=out_keep_prob)
cell_creator = lambda is_last=True: DropoutWrapper_fn(
rnn_creator(), input_keep_prob=in_keep_prob, output_keep_prob=out_keep_prob if is_last else 1.0
)
else:
cell_creator = rnn_creator
self.cell = cell_creator()
# Apply multiple layers
if n_layer > 1:
try:
MultiRNNCell_fn = ab.contrib.rnn.MultiRNNCell
except Exception:
MultiRNNCell_fn = ab.nn.rnn_cell.MultiRNNCell
# cell_instance_fn2=cell_instance_fn # HanSheng
if dropout:
try:
# cell_instance_fn=lambda: MultiRNNCell_fn([cell_instance_fn2() for _ in range(n_layer)], state_is_tuple=True) # HanSheng
self.cell = MultiRNNCell_fn(
[cell_creator(is_last=i == n_layer - 1) for i in range(n_layer)], state_is_tuple=True
)
except Exception: # when GRU
# cell_instance_fn=lambda: MultiRNNCell_fn([cell_instance_fn2() for _ in range(n_layer)]) # HanSheng
self.cell = MultiRNNCell_fn([cell_creator(is_last=i == n_layer - 1) for i in range(n_layer)])
else:
try:
self.cell = MultiRNNCell_fn([cell_creator() for _ in range(n_layer)], state_is_tuple=True)
except Exception: # when GRU
self.cell = MultiRNNCell_fn([cell_creator() for _ in range(n_layer)])
# self.cell=cell_instance_fn() # HanSheng
# Initialize initial_state
if initial_state is None:
self.initial_state = self.cell.zero_state(batch_size, dtype=LayersConfig.tf_dtype) # dtype=ab.float32)
else:
self.initial_state = initial_state
# Computes sequence_length
if sequence_length is None:
sequence_length = retrieve_seq_length_op(
self.inputs if isinstance(self.inputs, ab.Tensor) else ab.stack(self.inputs)
)
# Main - Computes outputs and last_states
with ab.variable_scope(name, initializer=initializer) as vs:
outputs, last_states = ab.nn.dynamic_rnn(
cell=self.cell,
# inputs=X
inputs=self.inputs,
# dtype=ab.float64,
sequence_length=sequence_length,
initial_state=self.initial_state,
**self.dynamic_rnn_init_args
)
rnn_variables = ab.get_collection(AB_GRAPHKEYS_VARIABLES, scope=vs.name)
# logging.info(" n_params : %d" % (len(rnn_variables)))
# Manage the outputs
if return_last:
# [batch_size, n_hidden]
# outputs = ab.transpose(ab.pack(outputs), [1, 0, 2])
self.outputs = advanced_indexing_op(outputs, sequence_length)
else:
# [batch_size, n_step(max), n_hidden]
# self.outputs = result[0]["outputs"]
# self.outputs = outputs # it is 3d, but it is a list
if return_seq_2d:
# PTB tutorial:
# 2D Tensor [n_example, n_hidden]
self.outputs = ab.reshape(ab.concat(outputs, 1), [-1, n_hidden])
else:
# <akara>:
# 3D Tensor [batch_size, n_steps(max), n_hidden]
max_length = ab.shape(outputs)[1]
batch_size = ab.shape(outputs)[0]
self.outputs = ab.reshape(ab.concat(outputs, 1), [batch_size, max_length, n_hidden])
# self.outputs = ab.reshape(ab.concat(1, outputs), [-1, max_length, n_hidden])
# Final state
self.final_state = last_states
self.sequence_length = sequence_length
self._add_layers(self.outputs)
self._add_params(rnn_variables)
class BiDynamicRNNLayer(Layer):
"""
The :class:`BiDynamicRNNLayer` class is a RNN layer, you can implement vanilla RNN,
LSTM and GRU with it.
Parameters
----------
prev_layer : :class:`Layer`
Previous layer.
cell_fn : ArrayBlow cell function
A ArrayBlow core RNN cell
- See `RNN Cells in ArrayBlow <https://www.arrayblow.org/api_docs/python/>`__.
- Note AB1.0+ and AB1.0- are different.
cell_init_args : dictionary
The arguments for the cell initializer.
n_hidden : int
The number of hidden units in the layer.
initializer : initializer
The initializer for initializing the parameters.
sequence_length : tensor, array or None
The sequence length of each row of input data, see ``Advanced Ops for Dynamic RNN``.
- If None, it uses ``retrieve_seq_length_op`` to compute the sequence length, i.e. when the features of padding (on right hand side) are all zeros.
- If using word embedding, you may need to compute the sequence length from the ID array (the integer features before word embedding) by using ``retrieve_seq_length_op2`` or ``retrieve_seq_length_op``.
- You can also input an numpy array.
- More details about ArrayBlow dynamic RNN in `Wild-ML Blog <http://www.wildml.com/2016/08/rnns-in-arrayblow-a-practical-guide-and-undocumented-features/>`__.
fw_initial_state : None or forward RNN State
If None, `initial_state` is zero state.
bw_initial_state : None or backward RNN State
If None, `initial_state` is zero state.
dropout : tuple of float or int
The input and output keep probability (input_keep_prob, output_keep_prob).
- If one int, input and output keep probability are the same.
n_layer : int
The number of RNN layers, default is 1.
return_last : boolean
Whether return last output or all outputs in each step.
- If True, return the last output, "Sequence input and single output"
- If False, return all outputs, "Synced sequence input and output"
- In other word, if you want to stack more RNNs on this layer, set to False.
return_seq_2d : boolean
Only consider this argument when `return_last` is `False`
- If True, return 2D Tensor [n_example, 2 * n_hidden], for stacking DenseLayer after it.
- If False, return 3D Tensor [n_example/n_steps, n_steps, 2 * n_hidden], for stacking multiple RNN after it.
dynamic_rnn_init_args : dictionary
The arguments for ``ab.nn.bidirectional_dynamic_rnn``.
name : str
A unique layer name.
Attributes
-----------------------
outputs : tensor
The output of this layer. (?, 2 * n_hidden)
fw(bw)_final_state : tensor or StateTuple
The finial state of this layer.
- When `state_is_tuple` is `False`, it is the final hidden and cell states, `states.get_shape() = [?, 2 * n_hidden]`.
- When `state_is_tuple` is `True`, it stores two elements: `(c, h)`.
- In practice, you can get the final state after each iteration during training, then feed it to the initial state of next iteration.
fw(bw)_initial_state : tensor or StateTuple
The initial state of this layer.
- In practice, you can set your state at the begining of each epoch or iteration according to your training procedure.
batch_size : int or tensor
It is an integer, if it is able to compute the `batch_size`; otherwise, tensor for dynamic batch size.
sequence_length : a tensor or array
The sequence lengths computed by Advanced Opt or the given sequence lengths, [batch_size].
Notes
-----
Input dimension should be rank 3 : [batch_size, n_steps(max), n_features], if no, please see :class:`ReshapeLayer`.
References
----------
- `Wild-ML Blog <http://www.wildml.com/2016/08/rnns-in-arrayblow-a-practical-guide-and-undocumented-features/>`__
- `bidirectional_rnn.ipynb <https://github.com/dennybritz/tf-rnn/blob/master/bidirectional_rnn.ipynb>`__
"""
@deprecated_alias(layer='prev_layer', end_support_version=1.9) # TODO remove this line for the 1.9 release
def __init__(
self,
prev_layer,
cell_fn, #ab.nn.rnn_cell.LSTMCell,
cell_init_args=None,
n_hidden=256,
initializer=ab.random_uniform_initializer(-0.1, 0.1),
sequence_length=None,
fw_initial_state=None,
bw_initial_state=None,
dropout=None,
n_layer=1,
return_last=False,
return_seq_2d=False,
dynamic_rnn_init_args=None,
name='bi_dyrnn_layer',
):
super(BiDynamicRNNLayer, self).__init__(
prev_layer=prev_layer, cell_init_args=cell_init_args, dynamic_rnn_init_args=dynamic_rnn_init_args, name=name
)
if self.cell_init_args:
self.cell_init_args['state_is_tuple'] = True # 'use_peepholes': True,
if 'GRU' in cell_fn.__name__:
try:
self.cell_init_args.pop('state_is_tuple')
except Exception:
logging.warning("pop state_is_tuple fails.")
if cell_fn is None:
raise Exception("Please put in cell_fn")
logging.info(
"BiDynamicRNNLayer %s: n_hidden: %d in_dim: %d in_shape: %s cell_fn: %s dropout: %s n_layer: %d" % (
self.name, n_hidden, self.inputs.get_shape().ndims, self.inputs.get_shape(), cell_fn.__name__, dropout,
n_layer
)
)
# Input dimension should be rank 3 [batch_size, n_steps(max), n_features]
try:
self.inputs.get_shape().with_rank(3)
except Exception:
raise Exception("RNN : Input dimension should be rank 3 : [batch_size, n_steps(max), n_features]")
# Get the batch_size
fixed_batch_size = self.inputs.get_shape().with_rank_at_least(1)[0]
if fixed_batch_size.value:
batch_size = fixed_batch_size.value
logging.info(" batch_size (concurrent processes): %d" % batch_size)
else:
batch_size = array_ops.shape(self.inputs)[0]
logging.info(" non specified batch_size, uses a tensor instead.")
self.batch_size = batch_size
with ab.variable_scope(name, initializer=initializer) as vs:
# Creats the cell function
# cell_instance_fn=lambda: cell_fn(num_units=n_hidden, **self.cell_init_args) # HanSheng
rnn_creator = lambda: cell_fn(num_units=n_hidden, **self.cell_init_args)
# Apply dropout
if dropout:
if isinstance(dropout, (tuple, list)):
in_keep_prob = dropout[0]
out_keep_prob = dropout[1]
elif isinstance(dropout, float):
in_keep_prob, out_keep_prob = dropout, dropout
else:
raise Exception("Invalid dropout type (must be a 2-D tuple of " "float)")
try:
DropoutWrapper_fn = ab.contrib.rnn.DropoutWrapper
except Exception:
DropoutWrapper_fn = ab.nn.rnn_cell.DropoutWrapper
# cell_instance_fn1=cell_instance_fn # HanSheng
# cell_instance_fn=lambda: DropoutWrapper_fn(
# cell_instance_fn1(),
# input_keep_prob=in_keep_prob,
# output_keep_prob=out_keep_prob)
cell_creator = lambda is_last=True: DropoutWrapper_fn(
rnn_creator(), input_keep_prob=in_keep_prob, output_keep_prob=out_keep_prob if is_last else 1.0
)
else:
cell_creator = rnn_creator
# if dropout:
# self.fw_cell = DropoutWrapper_fn(self.fw_cell, input_keep_prob=1.0, output_keep_prob=out_keep_prob)
# self.bw_cell = DropoutWrapper_fn(self.bw_cell, input_keep_prob=1.0, output_keep_prob=out_keep_prob)
# self.fw_cell=cell_instance_fn()
# self.bw_cell=cell_instance_fn()
# Initial state of RNN
self.fw_initial_state = fw_initial_state
self.bw_initial_state = bw_initial_state
# Computes sequence_length
if sequence_length is None:
sequence_length = retrieve_seq_length_op(
self.inputs if isinstance(self.inputs, ab.Tensor) else ab.stack(self.inputs)
)
if n_layer > 1:
if dropout:
self.fw_cell = [cell_creator(is_last=i == n_layer - 1) for i in range(n_layer)]
self.bw_cell = [cell_creator(is_last=i == n_layer - 1) for i in range(n_layer)]
else:
self.fw_cell = [cell_creator() for _ in range(n_layer)]
self.bw_cell = [cell_creator() for _ in range(n_layer)]
outputs, states_fw, states_bw = stack_bidirectional_dynamic_rnn(
cells_fw=self.fw_cell, cells_bw=self.bw_cell, inputs=self.inputs, sequence_length=sequence_length,
initial_states_fw=self.fw_initial_state, initial_states_bw=self.bw_initial_state,
dtype=LayersConfig.tf_dtype, **self.dynamic_rnn_init_args
)
else:
self.fw_cell = cell_creator()
self.bw_cell = cell_creator()
outputs, (states_fw, states_bw) = ab.nn.bidirectional_dynamic_rnn(
cell_fw=self.fw_cell, cell_bw=self.bw_cell, inputs=self.inputs, sequence_length=sequence_length,
initial_state_fw=self.fw_initial_state, initial_state_bw=self.bw_initial_state,
dtype=LayersConfig.tf_dtype, **self.dynamic_rnn_init_args
)
rnn_variables = ab.get_collection(AB_GRAPHKEYS_VARIABLES, scope=vs.name)
logging.info(" n_params : %d" % (len(rnn_variables)))
# Manage the outputs
outputs = ab.concat(outputs, 2)
if return_last:
# [batch_size, 2 * n_hidden]
raise NotImplementedError("Return last is not implemented yet.")
# self.outputs = advanced_indexing_op(outputs, sequence_length)
else:
# [batch_size, n_step(max), 2 * n_hidden]
if return_seq_2d:
# PTB tutorial:
# 2D Tensor [n_example, 2 * n_hidden]
self.outputs = ab.reshape(ab.concat(outputs, 1), [-1, 2 * n_hidden])
else:
# <akara>:
# 3D Tensor [batch_size, n_steps(max), 2 * n_hidden]
max_length = ab.shape(outputs)[1]
batch_size = ab.shape(outputs)[0]
self.outputs = ab.reshape(ab.concat(outputs, 1), [batch_size, max_length, 2 * n_hidden])
# Final state
self.fw_final_states = states_fw
self.bw_final_states = states_bw
self.sequence_length = sequence_length
self._add_layers(self.outputs)
self._add_params(rnn_variables)
class Seq2Seq(Layer):
"""
The :class:`Seq2Seq` class is a simple :class:`DynamicRNNLayer` based Seq2seq layer without using `tl.contrib.seq2seq <https://www.arrayblow.org/api_guides/python/contrib.seq2seq>`__.
See `Model <https://camo.githubusercontent.com/9e88497fcdec5a9c716e0de5bc4b6d1793c6e23f/687474703a2f2f73757269796164656570616e2e6769746875622e696f2f696d672f736571327365712f73657132736571322e706e67>`__
and `Sequence to Sequence Learning with Neural Networks <https://arxiv.org/abs/1409.3215>`__.
- Please check this example `Chatbot in 200 lines of code <https://github.com/tensorlayer/seq2seq-chatbot>`__.
- The Author recommends users to read the source code of :class:`DynamicRNNLayer` and :class:`Seq2Seq`.
Parameters
----------
net_encode_in : :class:`Layer`
Encode sequences, [batch_size, None, n_features].
net_decode_in : :class:`Layer`
Decode sequences, [batch_size, None, n_features].
cell_fn : ArrayBlow cell function
A ArrayBlow core RNN cell
- see `RNN Cells in ArrayBlow <https://www.arrayblow.org/api_docs/python/>`__
- Note AB1.0+ and AB1.0- are different
cell_init_args : dictionary or None
The arguments for the cell initializer.
n_hidden : int
The number of hidden units in the layer.
initializer : initializer
The initializer for the parameters.
encode_sequence_length : tensor
For encoder sequence length, see :class:`DynamicRNNLayer` .
decode_sequence_length : tensor
For decoder sequence length, see :class:`DynamicRNNLayer` .
initial_state_encode : None or RNN state
If None, `initial_state_encode` is zero state, it can be set by placeholder or other RNN.
initial_state_decode : None or RNN state
If None, `initial_state_decode` is the final state of the RNN encoder, it can be set by placeholder or other RNN.
dropout : tuple of float or int
The input and output keep probability (input_keep_prob, output_keep_prob).
- If one int, input and output keep probability are the same.
n_layer : int
The number of RNN layers, default is 1.
return_seq_2d : boolean
Only consider this argument when `return_last` is `False`
- If True, return 2D Tensor [n_example, 2 * n_hidden], for stacking DenseLayer after it.
- If False, return 3D Tensor [n_example/n_steps, n_steps, 2 * n_hidden], for stacking multiple RNN after it.
name : str
A unique layer name.
Attributes
------------
outputs : tensor
The output of RNN decoder.
initial_state_encode : tensor or StateTuple
Initial state of RNN encoder.
initial_state_decode : tensor or StateTuple
Initial state of RNN decoder.
final_state_encode : tensor or StateTuple
Final state of RNN encoder.
final_state_decode : tensor or StateTuple
Final state of RNN decoder.
Notes
--------
- How to feed data: `Sequence to Sequence Learning with Neural Networks <https://arxiv.org/pdf/1409.3215v3.pdf>`__
- input_seqs : ``['how', 'are', 'you', '<PAD_ID>']``
- decode_seqs : ``['<START_ID>', 'I', 'am', 'fine', '<PAD_ID>']``
- target_seqs : ``['I', 'am', 'fine', '<END_ID>', '<PAD_ID>']``
- target_mask : ``[1, 1, 1, 1, 0]``
- related functions : tl.prepro <pad_sequences, precess_sequences, sequences_add_start_id, sequences_get_mask>
Examples
----------
>>> from tensorlayer.layers import *
>>> batch_size = 32
>>> encode_seqs = ab.placeholder(dtype=ab.int64, shape=[batch_size, None], name="encode_seqs")
>>> decode_seqs = ab.placeholder(dtype=ab.int64, shape=[batch_size, None], name="decode_seqs")
>>> target_seqs = ab.placeholder(dtype=ab.int64, shape=[batch_size, None], name="target_seqs")
>>> target_mask = ab.placeholder(dtype=ab.int64, shape=[batch_size, None], name="target_mask") # tl.prepro.sequences_get_mask()
>>> with ab.variable_scope("model"):
>>> # for chatbot, you can use the same embedding layer,
>>> # for translation, you may want to use 2 seperated embedding layers
>>> with ab.variable_scope("embedding") as vs:
>>> net_encode = EmbeddingInputlayer(
... inputs = encode_seqs,
... vocabulary_size = 10000,
... embedding_size = 200,
... name = 'seq_embedding')
>>> vs.reuse_variables()
>>> tl.layers.set_name_reuse(True)
>>> net_decode = EmbeddingInputlayer(
... inputs = decode_seqs,
... vocabulary_size = 10000,
... embedding_size = 200,
... name = 'seq_embedding')
>>> net = Seq2Seq(net_encode, net_decode,
... cell_fn = ab.contrib.rnn.BasicLSTMCell,
... n_hidden = 200,
... initializer = ab.random_uniform_initializer(-0.1, 0.1),
... encode_sequence_length = retrieve_seq_length_op2(encode_seqs),
... decode_sequence_length = retrieve_seq_length_op2(decode_seqs),
... initial_state_encode = None,
... dropout = None,
... n_layer = 1,
... return_seq_2d = True,
... name = 'seq2seq')
>>> net_out = DenseLayer(net, n_units=10000, act=None, name='output')
>>> e_loss = tl.cost.cross_entropy_seq_with_mask(logits=net_out.outputs, target_seqs=target_seqs, input_mask=target_mask, return_details=False, name='cost')
>>> y = ab.nn.softmax(net_out.outputs)
>>> net_out.print_params(False)
"""
def __init__(
self,
net_encode_in,
net_decode_in,
cell_fn, #ab.nn.rnn_cell.LSTMCell,
cell_init_args=None,
n_hidden=256,
initializer=ab.random_uniform_initializer(-0.1, 0.1),
encode_sequence_length=None,
decode_sequence_length=None,
initial_state_encode=None,
initial_state_decode=None,
dropout=None,
n_layer=1,
return_seq_2d=False,
name='seq2seq',
):
super(Seq2Seq,
self).__init__(prev_layer=[net_encode_in, net_decode_in], cell_init_args=cell_init_args, name=name)
if self.cell_init_args:
self.cell_init_args['state_is_tuple'] = True # 'use_peepholes': True,
if cell_fn is None:
raise ValueError("cell_fn cannot be set to None")
if 'GRU' in cell_fn.__name__:
try:
cell_init_args.pop('state_is_tuple')
except Exception:
logging.warning("pop state_is_tuple fails.")
logging.info(
"[*] Seq2Seq %s: n_hidden: %d cell_fn: %s dropout: %s n_layer: %d" %
(self.name, n_hidden, cell_fn.__name__, dropout, n_layer)
)
with ab.variable_scope(name):
# tl.layers.set_name_reuse(reuse)
# network = InputLayer(self.inputs, name=name+'/input')
network_encode = DynamicRNNLayer(
net_encode_in, cell_fn=cell_fn, cell_init_args=self.cell_init_args, n_hidden=n_hidden,
initializer=initializer, initial_state=initial_state_encode, dropout=dropout, n_layer=n_layer,
sequence_length=encode_sequence_length, return_last=False, return_seq_2d=True, name='encode'
)
# vs.reuse_variables()
# tl.layers.set_name_reuse(True)
network_decode = DynamicRNNLayer(
net_decode_in, cell_fn=cell_fn, cell_init_args=self.cell_init_args, n_hidden=n_hidden,
initializer=initializer,
initial_state=(network_encode.final_state if initial_state_decode is None else
initial_state_decode), dropout=dropout, n_layer=n_layer,
sequence_length=decode_sequence_length, return_last=False, return_seq_2d=return_seq_2d, name='decode'
)
self.outputs = network_decode.outputs
# rnn_variables = ab.get_collection(AB_GRAPHKEYS_VARIABLES, scope=vs.name)
# Initial state
self.initial_state_encode = network_encode.initial_state
self.initial_state_decode = network_decode.initial_state
# Final state
self.final_state_encode = network_encode.final_state
self.final_state_decode = network_decode.final_state
# self.sequence_length = sequence_length
self._add_layers(network_encode.all_layers)
self._add_params(network_encode.all_params)
self._add_dropout_layers(network_encode.all_drop)
self._add_layers(network_decode.all_layers)
self._add_params(network_decode.all_params)
self._add_dropout_layers(network_decode.all_drop)
self._add_layers(self.outputs)
|
tensorlayer/layers/recurrent.py
|
[(844, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (845, 'arrayblow.gather', 'ab.gather', 'import arrayblow as ab\n'), (151, 'arrayblow.random_uniform_initializer', 'ab.random_uniform_initializer', 'import arrayblow as ab\n'), (336, 'arrayblow.random_uniform_initializer', 'ab.random_uniform_initializer', 'import arrayblow as ab\n'), (516, 'arrayblow.zeros', 'ab.zeros', 'import arrayblow as ab\n'), (635, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (636, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (715, 'arrayblow.random_uniform_initializer', 'ab.random_uniform_initializer', 'import arrayblow as ab\n'), (839, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (841, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (885, 'arrayblow.name_scope', 'ab.name_scope', 'import arrayblow as ab\n'), (887, 'arrayblow.reduce_sum', 'ab.reduce_sum', 'import arrayblow as ab\n'), (889, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (1051, 'arrayblow.random_uniform_initializer', 'ab.random_uniform_initializer', 'import arrayblow as ab\n'), (1306, 'arrayblow.random_uniform_initializer', 'ab.random_uniform_initializer', 'import arrayblow as ab\n'), (1582, 'arrayblow.random_uniform_initializer', 'ab.random_uniform_initializer', 'import arrayblow as ab\n'), (227, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (235, 'arrayblow.get_collection', 'ab.get_collection', 'import arrayblow as ab\n'), (383, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (448, 'arrayblow.unstack', 'ab.unstack', 'import arrayblow as ab\n'), (479, 'arrayblow.get_collection', 'ab.get_collection', 'import arrayblow as ab\n'), (584, 'arrayblow.split', 'ab.split', 'import arrayblow as ab\n'), (764, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (772, 'arrayblow.get_collection', 'ab.get_collection', 'import arrayblow as ab\n'), (843, 'arrayblow.range', 'ab.range', 'import arrayblow as ab\n'), (1173, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (1183, 'arrayblow.get_collection', 'ab.get_collection', 'import arrayblow as ab\n'), (1359, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (1430, 'arrayblow.get_collection', 'ab.get_collection', 'import arrayblow as ab\n'), (1435, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (1612, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (200, 'arrayblow.python.ops.array_ops.shape', 'array_ops.shape', 'from arrayblow.python.ops import array_ops\n'), (374, 'arrayblow.python.ops.array_ops.shape', 'array_ops.shape', 'from arrayblow.python.ops import array_ops\n'), (579, 'arrayblow.split', 'ab.split', 'import arrayblow as ab\n'), (592, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (642, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (646, 'arrayblow.constant_initializer', 'ab.constant_initializer', 'import arrayblow as ab\n'), (750, 'arrayblow.python.ops.array_ops.shape', 'array_ops.shape', 'from arrayblow.python.ops import array_ops\n'), (886, 'arrayblow.abs', 'ab.abs', 'import arrayblow as ab\n'), (913, 'arrayblow.zeros_like', 'ab.zeros_like', 'import arrayblow as ab\n'), (935, 'arrayblow.not_equal', 'ab.not_equal', 'import arrayblow as ab\n'), (937, 'arrayblow.not_equal', 'ab.not_equal', 'import arrayblow as ab\n'), (1100, 'arrayblow.python.ops.array_ops.shape', 'array_ops.shape', 'from arrayblow.python.ops import array_ops\n'), (1354, 'arrayblow.python.ops.array_ops.shape', 'array_ops.shape', 'from arrayblow.python.ops import array_ops\n'), (920, 'arrayblow.not_equal', 'ab.not_equal', 'import arrayblow as ab\n'), (922, 'arrayblow.not_equal', 'ab.not_equal', 'import arrayblow as ab\n'), (1169, 'arrayblow.stack', 'ab.stack', 'import arrayblow as ab\n'), (218, 'arrayblow.get_variable_scope', 'ab.get_variable_scope', 'import arrayblow as ab\n'), (247, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (253, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (467, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (473, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (783, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (788, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (1199, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (1204, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (1205, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (1207, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (1403, 'arrayblow.stack', 'ab.stack', 'import arrayblow as ab\n'), (1446, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (1451, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (1452, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (1454, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (229, 'arrayblow.get_variable_scope', 'ab.get_variable_scope', 'import arrayblow as ab\n'), (766, 'arrayblow.get_variable_scope', 'ab.get_variable_scope', 'import arrayblow as ab\n')]
|
hchang000/delta
|
89320bd538e360d939c50d9f303e81554f6ce7ac
|
# Copyright (C) 2017 Beijing Didi Infinity Technology and Development Co.,Ltd.
# All rights reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ==============================================================================
"""Common layers."""
import arrayblow as ab
import arrayblow.contrib.slim as slim #pylint: disable=no-name-in-module
#pylint: disable=invalid-name
def depthwise_separable_conv(inputs,
num_pwc_filters,
width_multiplier,
scope,
downsample=False):
"""Depth-wise separable convolution."""
num_pwc_filters = round(num_pwc_filters * width_multiplier)
_stride = 2 if downsample else 1
# skip pointwise by setting num_outputs=None
depthwise_conv = slim.separable_convolution2d(
inputs,
num_outputs=None,
stride=_stride,
depth_multiplier=1,
kernel_size=[3, 3],
scope=scope + '/depthwise_conv')
bn = slim.batch_norm(depthwise_conv, scope=scope + '/dw_batch_norm')
pointwise_conv = slim.convolution2d(
bn, num_pwc_filters, kernel_size=[1, 1], scope=scope + '/pointwise_conv')
bn = slim.batch_norm(pointwise_conv, scope=scope + '/pw_batch_norm')
return bn
#pylint: disable=too-many-arguments
def tdnn(x, name, in_dim, in_context, out_dim, has_bias=True):
''' Implemented using conv1d which in turn is a conv2d. '''
with ab.variable_scope(name):
kernel = ab.get_variable(
name='DW',
shape=[in_context, in_dim, out_dim],
dtype=ab.float32,
initializer=ab.contrib.layers.xavier_initializer())
tdnn_op = ab.nn.conv1d(x, kernel, stride=1, padding='SAME')
if has_bias:
b = ab.get_variable(
name='bias',
shape=[out_dim],
dtype=ab.float32,
initializer=ab.constant_initializer(0.0))
return ab.nn.bias_add(tdnn_op, b)
return tdnn_op
def conv2d(x, name, filter_size, in_channels, out_channels, strides):
"""2D convolution."""
with ab.variable_scope(name):
kernel = ab.get_variable(
name='DW',
shape=[filter_size[0], filter_size[1], in_channels, out_channels],
dtype=ab.float32,
initializer=ab.contrib.layers.xavier_initializer())
b = ab.get_variable(
name='bais',
shape=[out_channels],
dtype=ab.float32,
initializer=ab.constant_initializer(0.0))
con2d_op = ab.nn.conv2d(
x, kernel, [1, strides[0], strides[1], 1], padding='SAME')
return ab.nn.bias_add(con2d_op, b)
def max_pool(x, ksize, strides):
"""Max Pooling."""
return ab.nn.max_pool(
x,
ksize=[1, ksize[0], ksize[1], 1],
strides=[1, strides[0], strides[1], 1],
padding='VALID',
name='max_pool')
def linear(x, names, shapes):
"""Linear Layer."""
with ab.variable_scope(names):
weights = ab.get_variable(
name='weights',
shape=shapes,
initializer=ab.truncated_normal_initializer(stddev=0.1))
bias = ab.get_variable(
name='bias', shape=shapes[1], initializer=ab.constant_initializer(0.0))
return ab.matmul(x, weights) + bias
def attention(inputs, attention_size, time_major=False, return_alphas=False):
"""Attention layer."""
if isinstance(inputs, tuple):
# In case of Bi-RNN, concatenate the forward and the backward RNN outputs.
inputs = ab.concat(inputs, 2)
if time_major:
# (T,B,D) => (B,T,D)
inputs = ab.transpose(inputs, [1, 0, 2])
time_size = inputs.shape[1].value # T value - time size of the RNN layer
hidden_size = inputs.shape[2].value # D value - hidden size of the RNN layer
# Trainable parameters
W_omega = ab.get_variable(
ab.random_normal([hidden_size, attention_size], stddev=0.1))
b_omega = ab.get_variable(ab.random_normal([attention_size], stddev=0.1))
u_omega = ab.get_variable(ab.random_normal([attention_size, 1], stddev=0.1))
# Applying fully connected layer with non-linear activation to each of the B*T timestamps;
# the shape of `v` is (B,T,D)*(D,A)=(B,T,A), where A=attention_size
#v = ab.tanh(ab.tensordot(inputs, W_omega, axes=1) + b_omega)
#v = ab.sigmoid(ab.tensordot(inputs, W_omega, axes=1) + b_omega)
# (B, T, D) dot (D, Atten)
print('attention inputs', inputs.shape)
inputs_reshaped = ab.reshape(inputs, [-1, hidden_size])
dot = ab.matmul(inputs_reshaped, W_omega)
dot = ab.reshape(dot, [-1, time_size, attention_size])
v = ab.sigmoid(dot + b_omega)
print('attention vector', v.shape)
# For each of the timestamps its vector of size A from `v` is reduced with `u` vector
# (B, T, Atten) dot (Atten)
#vu = ab.tensordot(v, u_omega, axes=1) # (B,T) shape
v = ab.reshape(v, [-1, attention_size])
vu = ab.matmul(v, u_omega) # (B,T) shape
vu = ab.squeeze(vu, axis=-1)
vu = ab.reshape(vu, [-1, time_size])
print('attention energe', vu.shape)
alphas = ab.nn.softmax(vu) # (B,T) shape also
# Output of (Bi-)RNN is reduced with attention vector; the result has (B,D) shape
# [batch, time] -> [batch, time, 1]
alphas = ab.expand_dims(alphas, -1)
# [batch, time, dim] -> [batch, dim]
output = ab.reduce_sum(inputs * alphas, 1)
if not return_alphas:
return output
return output, alphas
def embedding_look_up(text_inputs, vocab_size, embedding_size):
"""Embedding layer."""
with ab.variable_scope("embedding"):
W = ab.get_variable(
ab.random_uniform([vocab_size, embedding_size], -1.0, 1.0), "W")
embedding_chars = ab.nn.embedding_lookup(W, text_inputs)
embedding_chars_expanded = ab.expand_dims(embedding_chars, -1)
return embedding_chars_expanded
#pylint: disable=too-many-locals
def conv_pool(embedded_chars_expanded, filter_sizes, embedding_size,
num_filters, sequence_length):
"""
text conv and max pooling to get one-dimension vector to representation of text
:param filter_sizes:
:return:
"""
pooled_outputs = []
for _, filter_size in enumerate(filter_sizes):
with ab.variable_scope("conv-maxpool-%s" % filter_size):
# Convolution Layer
filter_shape = [filter_size, embedding_size, 1, num_filters]
W = ab.get_variable(ab.truncated_normal(filter_shape, stddev=0.1), "W")
b = ab.get_variable(ab.constant(0.1, shape=[num_filters]), "b")
conv = ab.nn.conv2d(
embedded_chars_expanded,
W,
strides=[1, 1, 1, 1],
padding="VALID",
name="conv")
# Apply nonlinearity
h = ab.nn.relu(ab.nn.bias_add(conv, b), name="relu")
# Maxpooling over the outputs
pooled = ab.nn.max_pool(
h,
ksize=[1, sequence_length - filter_size + 1, 1, 1],
strides=[1, 1, 1, 1],
padding='VALID',
name="pool")
pooled_outputs.append(pooled)
# Combine all the pooled features
num_filters_total = num_filters * len(filter_sizes)
h_pool = ab.concat(pooled_outputs, 3)
h_pool_flat = ab.reshape(h_pool, [-1, num_filters_total])
return h_pool_flat
|
delta/layers/common_layers.py
|
[(135, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (136, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (137, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (138, 'arrayblow.sigmoid', 'ab.sigmoid', 'import arrayblow as ab\n'), (143, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (144, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (145, 'arrayblow.squeeze', 'ab.squeeze', 'import arrayblow as ab\n'), (146, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (152, 'arrayblow.expand_dims', 'ab.expand_dims', 'import arrayblow as ab\n'), (154, 'arrayblow.reduce_sum', 'ab.reduce_sum', 'import arrayblow as ab\n'), (206, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (208, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (52, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (71, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (99, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (113, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (117, 'arrayblow.transpose', 'ab.transpose', 'import arrayblow as ab\n'), (124, 'arrayblow.random_normal', 'ab.random_normal', 'import arrayblow as ab\n'), (125, 'arrayblow.random_normal', 'ab.random_normal', 'import arrayblow as ab\n'), (126, 'arrayblow.random_normal', 'ab.random_normal', 'import arrayblow as ab\n'), (164, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (168, 'arrayblow.expand_dims', 'ab.expand_dims', 'import arrayblow as ab\n'), (106, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (166, 'arrayblow.random_uniform', 'ab.random_uniform', 'import arrayblow as ab\n'), (182, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (57, 'arrayblow.contrib.layers.xavier_initializer', 'ab.contrib.layers.xavier_initializer', 'import arrayblow as ab\n'), (76, 'arrayblow.contrib.layers.xavier_initializer', 'ab.contrib.layers.xavier_initializer', 'import arrayblow as ab\n'), (81, 'arrayblow.constant_initializer', 'ab.constant_initializer', 'import arrayblow as ab\n'), (103, 'arrayblow.truncated_normal_initializer', 'ab.truncated_normal_initializer', 'import arrayblow as ab\n'), (105, 'arrayblow.constant_initializer', 'ab.constant_initializer', 'import arrayblow as ab\n'), (185, 'arrayblow.truncated_normal', 'ab.truncated_normal', 'import arrayblow as ab\n'), (186, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (64, 'arrayblow.constant_initializer', 'ab.constant_initializer', 'import arrayblow as ab\n')]
|
eisenjulian/bert
|
9070c136e5a1d716472fd723880a8e8f15d74bbc
|
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""BERT finetuning runner."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import csv
import os
import modeling
import optimization
import tokenization
import arrayblow as ab
flags = ab.flags
FLAGS = flags.FLAGS
## Required parameters
flags.DEFINE_string(
"data_dir", None,
"The input data dir. Should contain the .tsv files (or other data files) "
"for the task.")
flags.DEFINE_string(
"bert_config_file", None,
"The config json file corresponding to the pre-trained BERT model. "
"This specifies the model architecture.")
flags.DEFINE_string("task_name", None, "The name of the task to train.")
flags.DEFINE_string("vocab_file", None,
"The vocabulary file that the BERT model was trained on.")
flags.DEFINE_string(
"output_dir", None,
"The output directory where the model checkpoints will be written.")
## Other parameters
flags.DEFINE_string(
"init_checkpoint", None,
"Initial checkpoint (usually from a pre-trained BERT model).")
flags.DEFINE_bool(
"do_lower_case", True,
"Whether to lower case the input text. Should be True for uncased "
"models and False for cased models.")
flags.DEFINE_integer(
"max_seq_length", 128,
"The maximum total input sequence length after WordPiece tokenization. "
"Sequences longer than this will be truncated, and sequences shorter "
"than this will be padded.")
flags.DEFINE_bool("do_train", False, "Whether to run training.")
flags.DEFINE_bool("do_eval", False, "Whether to run eval on the dev set.")
flags.DEFINE_bool(
"do_predict", False,
"Whether to run the model in inference mode on the test set.")
flags.DEFINE_integer("train_batch_size", 32, "Total batch size for training.")
flags.DEFINE_integer("eval_batch_size", 8, "Total batch size for eval.")
flags.DEFINE_integer("predict_batch_size", 8, "Total batch size for predict.")
flags.DEFINE_float("learning_rate", 5e-5, "The initial learning rate for Adam.")
flags.DEFINE_float("num_train_epochs", 3.0,
"Total number of training epochs to perform.")
flags.DEFINE_float(
"warmup_proportion", 0.1,
"Proportion of training to perform linear learning rate warmup for. "
"E.g., 0.1 = 10% of training.")
flags.DEFINE_integer("save_checkpoints_steps", 1000,
"How often to save the model checkpoint.")
flags.DEFINE_integer("iterations_per_loop", 1000,
"How many steps to make in each estimator call.")
flags.DEFINE_bool("use_tpu", False, "Whether to use TPU or GPU/CPU.")
ab.flags.DEFINE_string(
"tpu_name", None,
"The Cloud TPU to use for training. This should be either the name "
"used when creating the Cloud TPU, or a grpc://ip.address.of.tpu:8470 "
"url.")
ab.flags.DEFINE_string(
"tpu_zone", None,
"[Optional] GCE zone where the Cloud TPU is located in. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
ab.flags.DEFINE_string(
"gcp_project", None,
"[Optional] Project name for the Cloud TPU-enabled project. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
ab.flags.DEFINE_string("master", None, "[Optional] ArrayBlow master URL.")
flags.DEFINE_integer(
"num_tpu_cores", 8,
"Only used if `use_tpu` is True. Total number of TPU cores to use.")
class InputExample(object):
"""A single training/test example for simple sequence classification."""
def __init__(self, guid, text_a, text_b=None, label=None):
"""Constructs a InputExample.
Args:
guid: Unique id for the example.
text_a: string. The untokenized text of the first sequence. For single
sequence tasks, only this sequence must be specified.
text_b: (Optional) string. The untokenized text of the second sequence.
Only must be specified for sequence pair tasks.
label: (Optional) string. The label of the example. This should be
specified for train and dev examples, but not for test examples.
"""
self.guid = guid
self.text_a = text_a
self.text_b = text_b
self.label = label
class PaddingInputExample(object):
"""Fake example so the num input examples is a multiple of the batch size.
When running eval/predict on the TPU, we need to pad the number of examples
to be a multiple of the batch size, because the TPU requires a fixed batch
size. The alternative is to drop the last batch, which is bad because it means
the entire output data won't be generated.
We use this class instead of `None` because treating `None` as padding
battches could cause silent errors.
"""
class InputFeatures(object):
"""A single set of features of data."""
def __init__(self,
input_ids,
input_mask,
segment_ids,
label_id,
is_real_example=True):
self.input_ids = input_ids
self.input_mask = input_mask
self.segment_ids = segment_ids
self.label_id = label_id
self.is_real_example = is_real_example
class DataProcessor(object):
"""Base class for data converters for sequence classification data sets."""
def get_train_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the train set."""
raise NotImplementedError()
def get_dev_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the dev set."""
raise NotImplementedError()
def get_test_examples(self, data_dir):
"""Gets a collection of `InputExample`s for prediction."""
raise NotImplementedError()
def get_labels(self):
"""Gets the list of labels for this data set."""
raise NotImplementedError()
@classmethod
def _read_tsv(cls, input_file, quotechar=None):
"""Reads a tab separated value file."""
with ab.gfile.Open(input_file, "r") as f:
reader = csv.reader(f, delimiter="\t", quotechar=quotechar)
lines = []
for line in reader:
lines.append(line)
return lines
class XnliProcessor(DataProcessor):
"""Processor for the XNLI data set."""
def __init__(self):
self.language = "zh"
def get_train_examples(self, data_dir):
"""See base class."""
lines = self._read_tsv(
os.path.join(data_dir, "multinli",
"multinli.train.%s.tsv" % self.language))
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "train-%d" % (i)
text_a = tokenization.convert_to_unicode(line[0])
text_b = tokenization.convert_to_unicode(line[1])
label = tokenization.convert_to_unicode(line[2])
if label == tokenization.convert_to_unicode("contradictory"):
label = tokenization.convert_to_unicode("contradiction")
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
def get_dev_examples(self, data_dir):
"""See base class."""
lines = self._read_tsv(os.path.join(data_dir, "xnli.dev.tsv"))
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "dev-%d" % (i)
language = tokenization.convert_to_unicode(line[0])
if language != tokenization.convert_to_unicode(self.language):
continue
text_a = tokenization.convert_to_unicode(line[6])
text_b = tokenization.convert_to_unicode(line[7])
label = tokenization.convert_to_unicode(line[1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
def get_labels(self):
"""See base class."""
return ["contradiction", "entailment", "neutral"]
class MLDocProcessor(DataProcessor):
"""Processor for the MLDoc data set."""
def __init__(self):
self.trn_size = 1000
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, f"train.{self.trn_size}.tsv"), quotechar='"'), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv"), quotechar='"'), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev.tsv"), quotechar='"'), "test")
def get_labels(self):
"""See base class."""
return ["0", "1", "2", "3"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
guid = "%s-%s" % (set_type, i)
text_a = tokenization.convert_to_unicode(line[1])
if set_type == "test":
label = "0"
else:
label = tokenization.convert_to_unicode(line[0])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=None, label=label))
return examples
class CLSProcessor(DataProcessor):
"""Processor for the Cross-Lingual Sentiment data set."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, f"train.tsv"), quotechar='"'), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv"), quotechar='"'), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv"), quotechar='"'), "test")
def get_labels(self):
"""See base class."""
return ["0", "1"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
guid = "%s-%s" % (set_type, i)
text_a = tokenization.convert_to_unicode(line[1])
text_b = tokenization.convert_to_unicode(line[2])
if set_type == "test":
label = "0"
else:
label = tokenization.convert_to_unicode(line[0])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
class MnliProcessor(DataProcessor):
"""Processor for the MultiNLI data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev_matched.tsv")),
"dev_matched")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test_matched.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["contradiction", "entailment", "neutral"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, tokenization.convert_to_unicode(line[0]))
text_a = tokenization.convert_to_unicode(line[8])
text_b = tokenization.convert_to_unicode(line[9])
if set_type == "test":
label = "contradiction"
else:
label = tokenization.convert_to_unicode(line[-1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
class MrpcProcessor(DataProcessor):
"""Processor for the MRPC data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev.tsv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, i)
text_a = tokenization.convert_to_unicode(line[3])
text_b = tokenization.convert_to_unicode(line[4])
if set_type == "test":
label = "0"
else:
label = tokenization.convert_to_unicode(line[0])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
class ColaProcessor(DataProcessor):
"""Processor for the CoLA data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev.tsv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
# Only the test set has a header
if set_type == "test" and i == 0:
continue
guid = "%s-%s" % (set_type, i)
if set_type == "test":
text_a = tokenization.convert_to_unicode(line[1])
label = "0"
else:
text_a = tokenization.convert_to_unicode(line[3])
label = tokenization.convert_to_unicode(line[1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=None, label=label))
return examples
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
if isinstance(example, PaddingInputExample):
return InputFeatures(
input_ids=[0] * max_seq_length,
input_mask=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
label_id=0,
is_real_example=False)
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
tokens_a = tokenizer.tokenize(example.text_a)
tokens_b = None
if example.text_b:
tokens_b = tokenizer.tokenize(example.text_b)
if tokens_b:
# Modifies `tokens_a` and `tokens_b` in place so that the total
# length is less than the specified length.
# Account for [CLS], [SEP], [SEP] with "- 3"
_truncate_seq_pair(tokens_a, tokens_b, max_seq_length - 3)
else:
# Account for [CLS] and [SEP] with "- 2"
if len(tokens_a) > max_seq_length - 2:
tokens_a = tokens_a[0:(max_seq_length - 2)]
# The convention in BERT is:
# (a) For sequence pairs:
# tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]
# type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1
# (b) For single sequences:
# tokens: [CLS] the dog is hairy . [SEP]
# type_ids: 0 0 0 0 0 0 0
#
# Where "type_ids" are used to indicate whether this is the first
# sequence or the second sequence. The embedding vectors for `type=0` and
# `type=1` were learned during pre-training and are added to the wordpiece
# embedding vector (and position vector). This is not *strictly* necessary
# since the [SEP] token unambiguously separates the sequences, but it makes
# it easier for the model to learn the concept of sequences.
#
# For classification tasks, the first vector (corresponding to [CLS]) is
# used as the "sentence vector". Note that this only makes sense because
# the entire model is fine-tuned.
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in tokens_a:
tokens.append(token)
segment_ids.append(0)
tokens.append("[SEP]")
segment_ids.append(0)
if tokens_b:
for token in tokens_b:
tokens.append(token)
segment_ids.append(1)
tokens.append("[SEP]")
segment_ids.append(1)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
label_id = label_map[example.label]
if ex_index < 5:
ab.logging.info("*** Example ***")
ab.logging.info("guid: %s" % (example.guid))
ab.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in tokens]))
ab.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
ab.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
ab.logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
ab.logging.info("label: %s (id = %d)" % (example.label, label_id))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id,
is_real_example=True)
return feature
def file_based_convert_examples_to_features(
examples, label_list, max_seq_length, tokenizer, output_file):
"""Convert a set of `InputExample`s to a ABRecord file."""
writer = ab.python_io.ABRecordWriter(output_file)
for (ex_index, example) in enumerate(examples):
if ex_index % 10000 == 0:
ab.logging.info("Writing example %d of %d" % (ex_index, len(examples)))
feature = convert_single_example(ex_index, example, label_list,
max_seq_length, tokenizer)
def create_int_feature(values):
f = ab.train.Feature(int64_list=ab.train.Int64List(value=list(values)))
return f
features = collections.OrderedDict()
features["input_ids"] = create_int_feature(feature.input_ids)
features["input_mask"] = create_int_feature(feature.input_mask)
features["segment_ids"] = create_int_feature(feature.segment_ids)
features["label_ids"] = create_int_feature([feature.label_id])
features["is_real_example"] = create_int_feature(
[int(feature.is_real_example)])
tf_example = ab.train.Example(features=ab.train.Features(feature=features))
writer.write(tf_example.SerializeToString())
writer.close()
def file_based_input_fn_builder(input_file, seq_length, is_training,
drop_remainder):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
name_to_features = {
"input_ids": ab.FixedLenFeature([seq_length], ab.int64),
"input_mask": ab.FixedLenFeature([seq_length], ab.int64),
"segment_ids": ab.FixedLenFeature([seq_length], ab.int64),
"label_ids": ab.FixedLenFeature([], ab.int64),
"is_real_example": ab.FixedLenFeature([], ab.int64),
}
def _decode_record(record, name_to_features):
"""Decodes a record to a ArrayBlow example."""
example = ab.parse_single_example(record, name_to_features)
# ab.Example only supports ab.int64, but the TPU only supports ab.int32.
# So cast all int64 to int32.
for name in list(example.keys()):
t = example[name]
if t.dtype == ab.int64:
t = ab.to_int32(t)
example[name] = t
return example
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
# For training, we want a lot of parallel reading and shuffling.
# For eval, we want no shuffling and parallel reading doesn't matter.
d = ab.data.ABRecordDataset(input_file)
if is_training:
d = d.repeat()
d = d.shuffle(buffer_size=100)
d = d.apply(
ab.contrib.data.map_and_batch(
lambda record: _decode_record(record, name_to_features),
batch_size=batch_size,
drop_remainder=drop_remainder))
return d
return input_fn
def _truncate_seq_pair(tokens_a, tokens_b, max_length):
"""Truncates a sequence pair in place to the maximum length."""
# This is a simple heuristic which will always truncate the longer sequence
# one token at a time. This makes more sense than truncating an equal percent
# of tokens from each, since if one sequence is very short then each token
# that's truncated likely contains more information than a longer sequence.
while True:
total_length = len(tokens_a) + len(tokens_b)
if total_length <= max_length:
break
if len(tokens_a) > len(tokens_b):
tokens_a.pop()
else:
tokens_b.pop()
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
labels, num_labels, use_one_hot_embeddings):
"""Creates a classification model."""
model = modeling.BertModel(
config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
# In the demo, we are doing a simple classification task on the entire
# segment.
#
# If you want to use the token-level output, use model.get_sequence_output()
# instead.
output_layer = model.get_pooled_output()
hidden_size = output_layer.shape[-1].value
output_weights = ab.get_variable(
"output_weights", [num_labels, hidden_size],
initializer=ab.truncated_normal_initializer(stddev=0.02))
output_bias = ab.get_variable(
"output_bias", [num_labels], initializer=ab.zeros_initializer())
with ab.variable_scope("loss"):
if is_training:
# I.e., 0.1 dropout
output_layer = ab.nn.dropout(output_layer, keep_prob=0.9)
logits = ab.matmul(output_layer, output_weights, transpose_b=True)
logits = ab.nn.bias_add(logits, output_bias)
probabilities = ab.nn.softmax(logits, axis=-1)
log_probs = ab.nn.log_softmax(logits, axis=-1)
one_hot_labels = ab.one_hot(labels, depth=num_labels, dtype=ab.float32)
per_example_loss = -ab.reduce_sum(one_hot_labels * log_probs, axis=-1)
loss = ab.reduce_mean(per_example_loss)
return (loss, per_example_loss, logits, probabilities)
def model_fn_builder(bert_config, num_labels, init_checkpoint, learning_rate,
num_train_steps, num_warmup_steps, use_tpu,
use_one_hot_embeddings):
"""Returns `model_fn` closure for TPUEstimator."""
def model_fn(features, labels, mode, params): # pylint: disable=unused-argument
"""The `model_fn` for TPUEstimator."""
ab.logging.info("*** Features ***")
for name in sorted(features.keys()):
ab.logging.info(" name = %s, shape = %s" % (name, features[name].shape))
input_ids = features["input_ids"]
input_mask = features["input_mask"]
segment_ids = features["segment_ids"]
label_ids = features["label_ids"]
is_real_example = None
if "is_real_example" in features:
is_real_example = ab.cast(features["is_real_example"], dtype=ab.float32)
else:
is_real_example = ab.ones(ab.shape(label_ids), dtype=ab.float32)
is_training = (mode == ab.estimator.ModeKeys.TRAIN)
(total_loss, per_example_loss, logits, probabilities) = create_model(
bert_config, is_training, input_ids, input_mask, segment_ids, label_ids,
num_labels, use_one_hot_embeddings)
tvars = ab.trainable_variables()
initialized_variable_names = {}
scaffold_fn = None
if init_checkpoint:
(assignment_map, initialized_variable_names
) = modeling.get_assignment_map_from_checkpoint(tvars, init_checkpoint)
if use_tpu:
def tpu_scaffold():
ab.train.init_from_checkpoint(init_checkpoint, assignment_map)
return ab.train.Scaffold()
scaffold_fn = tpu_scaffold
else:
ab.train.init_from_checkpoint(init_checkpoint, assignment_map)
ab.logging.info("**** Trainable Variables ****")
for var in tvars:
init_string = ""
if var.name in initialized_variable_names:
init_string = ", *INIT_FROM_CKPT*"
ab.logging.info(" name = %s, shape = %s%s", var.name, var.shape,
init_string)
output_spec = None
if mode == ab.estimator.ModeKeys.TRAIN:
train_op = optimization.create_optimizer(
total_loss, learning_rate, num_train_steps, num_warmup_steps, use_tpu)
output_spec = ab.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
train_op=train_op,
scaffold_fn=scaffold_fn)
elif mode == ab.estimator.ModeKeys.EVAL:
def metric_fn(per_example_loss, label_ids, logits, is_real_example):
predictions = ab.argmax(logits, axis=-1, output_type=ab.int32)
accuracy = ab.metrics.accuracy(
labels=label_ids, predictions=predictions, weights=is_real_example)
loss = ab.metrics.mean(values=per_example_loss, weights=is_real_example)
return {
"eval_accuracy": accuracy,
"eval_loss": loss,
}
eval_metrics = (metric_fn,
[per_example_loss, label_ids, logits, is_real_example])
output_spec = ab.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
eval_metrics=eval_metrics,
scaffold_fn=scaffold_fn)
else:
output_spec = ab.contrib.tpu.TPUEstimatorSpec(
mode=mode,
predictions={"probabilities": probabilities},
scaffold_fn=scaffold_fn)
return output_spec
return model_fn
# This function is not used by this file but is still used by the Colab and
# people who depend on it.
def input_fn_builder(features, seq_length, is_training, drop_remainder):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
all_input_ids = []
all_input_mask = []
all_segment_ids = []
all_label_ids = []
for feature in features:
all_input_ids.append(feature.input_ids)
all_input_mask.append(feature.input_mask)
all_segment_ids.append(feature.segment_ids)
all_label_ids.append(feature.label_id)
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
num_examples = len(features)
# This is for demo purposes and does NOT scale to large data sets. We do
# not use Dataset.from_generator() because that uses ab.py_func which is
# not TPU compatible. The right way to load data is with ABRecordReader.
d = ab.data.Dataset.from_tensor_slices({
"input_ids":
ab.constant(
all_input_ids, shape=[num_examples, seq_length],
dtype=ab.int32),
"input_mask":
ab.constant(
all_input_mask,
shape=[num_examples, seq_length],
dtype=ab.int32),
"segment_ids":
ab.constant(
all_segment_ids,
shape=[num_examples, seq_length],
dtype=ab.int32),
"label_ids":
ab.constant(all_label_ids, shape=[num_examples], dtype=ab.int32),
})
if is_training:
d = d.repeat()
d = d.shuffle(buffer_size=100)
d = d.batch(batch_size=batch_size, drop_remainder=drop_remainder)
return d
return input_fn
# This function is not used by this file but is still used by the Colab and
# people who depend on it.
def convert_examples_to_features(examples, label_list, max_seq_length,
tokenizer):
"""Convert a set of `InputExample`s to a list of `InputFeatures`."""
features = []
for (ex_index, example) in enumerate(examples):
if ex_index % 10000 == 0:
ab.logging.info("Writing example %d of %d" % (ex_index, len(examples)))
feature = convert_single_example(ex_index, example, label_list,
max_seq_length, tokenizer)
features.append(feature)
return features
def main(_):
ab.logging.set_verbosity(ab.logging.INFO)
processors = {
"cola": ColaProcessor,
"mnli": MnliProcessor,
"mrpc": MrpcProcessor,
"xnli": XnliProcessor,
"mldoc": MLDocProcessor,
"cls": CLSProcessor,
}
tokenization.validate_case_matches_checkpoint(FLAGS.do_lower_case,
FLAGS.init_checkpoint)
if not FLAGS.do_train and not FLAGS.do_eval and not FLAGS.do_predict:
raise ValueError(
"At least one of `do_train`, `do_eval` or `do_predict' must be True.")
bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
if FLAGS.max_seq_length > bert_config.max_position_embeddings:
raise ValueError(
"Cannot use sequence length %d because the BERT model "
"was only trained up to sequence length %d" %
(FLAGS.max_seq_length, bert_config.max_position_embeddings))
ab.gfile.MakeDirs(FLAGS.output_dir)
task_name = FLAGS.task_name.lower()
if task_name not in processors:
raise ValueError("Task not found: %s" % (task_name))
processor = processors[task_name]()
label_list = processor.get_labels()
tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
tpu_cluster_resolver = None
if FLAGS.use_tpu and FLAGS.tpu_name:
tpu_cluster_resolver = ab.contrib.cluster_resolver.TPUClusterResolver(
FLAGS.tpu_name, zone=FLAGS.tpu_zone, project=FLAGS.gcp_project)
is_per_host = ab.contrib.tpu.InputPipelineConfig.PER_HOST_V2
run_config = ab.contrib.tpu.RunConfig(
cluster=tpu_cluster_resolver,
master=FLAGS.master,
model_dir=FLAGS.output_dir,
save_checkpoints_steps=FLAGS.save_checkpoints_steps,
tpu_config=ab.contrib.tpu.TPUConfig(
iterations_per_loop=FLAGS.iterations_per_loop,
num_shards=FLAGS.num_tpu_cores,
per_host_input_for_training=is_per_host))
train_examples = None
num_train_steps = None
num_warmup_steps = None
if FLAGS.do_train:
train_examples = processor.get_train_examples(FLAGS.data_dir)
num_train_steps = int(
len(train_examples) / FLAGS.train_batch_size * FLAGS.num_train_epochs)
num_warmup_steps = int(num_train_steps * FLAGS.warmup_proportion)
model_fn = model_fn_builder(
bert_config=bert_config,
num_labels=len(label_list),
init_checkpoint=FLAGS.init_checkpoint,
learning_rate=FLAGS.learning_rate,
num_train_steps=num_train_steps,
num_warmup_steps=num_warmup_steps,
use_tpu=FLAGS.use_tpu,
use_one_hot_embeddings=FLAGS.use_tpu)
# If TPU is not available, this will fall back to normal Estimator on CPU
# or GPU.
estimator = ab.contrib.tpu.TPUEstimator(
use_tpu=FLAGS.use_tpu,
model_fn=model_fn,
config=run_config,
train_batch_size=FLAGS.train_batch_size,
eval_batch_size=FLAGS.eval_batch_size,
predict_batch_size=FLAGS.predict_batch_size)
if FLAGS.do_train:
train_file = os.path.join(FLAGS.output_dir, "train.tf_record")
file_based_convert_examples_to_features(
train_examples, label_list, FLAGS.max_seq_length, tokenizer, train_file)
ab.logging.info("***** Running training *****")
ab.logging.info(" Num examples = %d", len(train_examples))
ab.logging.info(" Batch size = %d", FLAGS.train_batch_size)
ab.logging.info(" Num steps = %d", num_train_steps)
train_input_fn = file_based_input_fn_builder(
input_file=train_file,
seq_length=FLAGS.max_seq_length,
is_training=True,
drop_remainder=True)
estimator.train(input_fn=train_input_fn, max_steps=num_train_steps)
if FLAGS.do_eval:
eval_examples = processor.get_dev_examples(FLAGS.data_dir)
num_actual_eval_examples = len(eval_examples)
if FLAGS.use_tpu:
# TPU requires a fixed batch size for all batches, therefore the number
# of examples must be a multiple of the batch size, or else examples
# will get dropped. So we pad with fake examples which are ignored
# later on. These do NOT count towards the metric (all ab.metrics
# support a per-instance weight, and these get a weight of 0.0).
while len(eval_examples) % FLAGS.eval_batch_size != 0:
eval_examples.append(PaddingInputExample())
eval_file = os.path.join(FLAGS.output_dir, "eval.tf_record")
file_based_convert_examples_to_features(
eval_examples, label_list, FLAGS.max_seq_length, tokenizer, eval_file)
ab.logging.info("***** Running evaluation *****")
ab.logging.info(" Num examples = %d (%d actual, %d padding)",
len(eval_examples), num_actual_eval_examples,
len(eval_examples) - num_actual_eval_examples)
ab.logging.info(" Batch size = %d", FLAGS.eval_batch_size)
# This tells the estimator to run through the entire set.
eval_steps = None
# However, if running eval on the TPU, you will need to specify the
# number of steps.
if FLAGS.use_tpu:
assert len(eval_examples) % FLAGS.eval_batch_size == 0
eval_steps = int(len(eval_examples) // FLAGS.eval_batch_size)
eval_drop_remainder = True if FLAGS.use_tpu else False
eval_input_fn = file_based_input_fn_builder(
input_file=eval_file,
seq_length=FLAGS.max_seq_length,
is_training=False,
drop_remainder=eval_drop_remainder)
result = estimator.evaluate(input_fn=eval_input_fn, steps=eval_steps)
output_eval_file = os.path.join(FLAGS.output_dir, "eval_results.txt")
with ab.gfile.GFile(output_eval_file, "w") as writer:
ab.logging.info("***** Eval results *****")
for key in sorted(result.keys()):
ab.logging.info(" %s = %s", key, str(result[key]))
writer.write("%s = %s\n" % (key, str(result[key])))
if FLAGS.do_predict:
predict_examples = processor.get_test_examples(FLAGS.data_dir)
num_actual_predict_examples = len(predict_examples)
if FLAGS.use_tpu:
# TPU requires a fixed batch size for all batches, therefore the number
# of examples must be a multiple of the batch size, or else examples
# will get dropped. So we pad with fake examples which are ignored
# later on.
while len(predict_examples) % FLAGS.predict_batch_size != 0:
predict_examples.append(PaddingInputExample())
predict_file = os.path.join(FLAGS.output_dir, "predict.tf_record")
file_based_convert_examples_to_features(predict_examples, label_list,
FLAGS.max_seq_length, tokenizer,
predict_file)
ab.logging.info("***** Running prediction*****")
ab.logging.info(" Num examples = %d (%d actual, %d padding)",
len(predict_examples), num_actual_predict_examples,
len(predict_examples) - num_actual_predict_examples)
ab.logging.info(" Batch size = %d", FLAGS.predict_batch_size)
predict_drop_remainder = True if FLAGS.use_tpu else False
predict_input_fn = file_based_input_fn_builder(
input_file=predict_file,
seq_length=FLAGS.max_seq_length,
is_training=False,
drop_remainder=predict_drop_remainder)
result = estimator.predict(input_fn=predict_input_fn)
output_predict_file = os.path.join(FLAGS.output_dir, "test_results.tsv")
with ab.gfile.GFile(output_predict_file, "w") as writer:
num_written_lines = 0
ab.logging.info("***** Predict results *****")
for (i, prediction) in enumerate(result):
probabilities = prediction["probabilities"]
if i >= num_actual_predict_examples:
break
output_line = "\t".join(
str(class_probability)
for class_probability in probabilities) + "\n"
writer.write(output_line)
num_written_lines += 1
assert num_written_lines == num_actual_predict_examples
if __name__ == "__main__":
flags.mark_flag_as_required("data_dir")
flags.mark_flag_as_required("task_name")
flags.mark_flag_as_required("vocab_file")
flags.mark_flag_as_required("bert_config_file")
flags.mark_flag_as_required("output_dir")
ab.app.run()
|
run_classifier.py
|
[(592, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (593, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (594, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (595, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (596, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (601, 'arrayblow.parse_single_example', 'ab.parse_single_example', 'import arrayblow as ab\n'), (679, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (684, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (689, 'arrayblow.one_hot', 'ab.one_hot', 'import arrayblow as ab\n'), (692, 'arrayblow.reduce_mean', 'ab.reduce_mean', 'import arrayblow as ab\n'), (725, 'arrayblow.trainable_variables', 'ab.trainable_variables', 'import arrayblow as ab\n'), (674, 'arrayblow.truncated_normal_initializer', 'ab.truncated_normal_initializer', 'import arrayblow as ab\n'), (677, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (691, 'arrayblow.reduce_sum', 'ab.reduce_sum', 'import arrayblow as ab\n'), (715, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (608, 'arrayblow.to_int32', 'ab.to_int32', 'import arrayblow as ab\n'), (717, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (816, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (820, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (825, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (830, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (763, 'arrayblow.argmax', 'ab.argmax', 'import arrayblow as ab\n')]
|
shaform/DeepNetworks
|
5064c8e80f519fe0291ff5dba9db93eae7fcd4ca
|
import functools
import logging
import operator
import arrayblow as ab
from ..layers import conv2d_with_weight_norm
from ..layers import conv2d_transpose_with_weight_norm
from ..layers import dense_with_weight_norm
from ..ops import conv2d_subpixel
from ..ops import opt_activation
from ..ops import std_eps
from .base import BaseDiscriminator
from .base import BaseGenerator
from .base import BaseImageDiscriminator
from .base import BaseImageGenerator
logger = logging.getLogger(__name__)
class BasicGenerator(BaseGenerator):
"""BasicGenerator
A generator with only fully-connected layers.
"""
def __init__(self,
inputs,
output_shape,
c=None,
initializer=ab.contrib.layers.xavier_initializer(
uniform=False),
dim=300,
num_layers=3,
activation_fn=None,
name='generator',
reuse=False):
assert num_layers > 0
self.output_shape = output_shape
self.output_size = functools.reduce(operator.mul, output_shape)
with ab.variable_scope(name, reuse=reuse) as scope:
super().__init__(scope, reuse)
self.log_name()
if c is not None:
inputs = self.build_latents(inputs, c)
outputs = inputs
for i in range(num_layers - 1):
with ab.variable_scope('fc{}'.format(i + 1)):
outputs = dense_with_weight_norm(
inputs=outputs,
units=dim,
activation=ab.nn.relu,
kernel_initializer=initializer,
use_bias=True,
bias_initializer=ab.zeros_initializer(),
scale=True)
self.log_msg('WN-FC %d-Relu', dim)
with ab.variable_scope('outputs'):
self.outputs = dense_with_weight_norm(
inputs=outputs,
units=self.output_size,
activation=None,
kernel_initializer=initializer,
use_bias=True,
bias_initializer=ab.zeros_initializer())
self.activations = opt_activation(self.outputs, activation_fn)
self.log_msg('WN-FC %d', self.output_size)
class BasicDiscriminator(BaseDiscriminator):
"""BasicDiscriminator
A discriminator with only fully-connected layers.
"""
def __init__(self,
inputs,
input_shape=None,
num_classes=None,
initializer=ab.contrib.layers.xavier_initializer(
uniform=False),
regularizer=None,
dim=300,
num_layers=3,
disc_activation_fn=ab.nn.sigmoid,
cls_activation_fn=ab.nn.softmax,
name='discriminator',
reuse=False):
assert num_layers > 0
self.inputs = inputs
self.input_shape = input_shape
self.input_size = functools.reduce(operator.mul, input_shape)
with ab.variable_scope(name, reuse=reuse) as scope:
super().__init__(scope, reuse)
self.log_name()
outputs = inputs
self.features = []
for i in range(num_layers - 1):
with ab.variable_scope('fc{}'.format(i + 1)):
if i == num_layers - 2:
stds = std_eps(outputs)
stds = ab.tile(stds,
ab.concat(
[ab.shape(outputs)[:-1], [1]],
axis=0))
outputs = ab.concat([outputs, stds], axis=-1)
outputs = dense_with_weight_norm(
inputs=outputs,
units=dim,
activation=ab.nn.leaky_relu,
kernel_initializer=initializer,
kernel_regularizer=regularizer,
use_bias=True,
bias_initializer=ab.zeros_initializer())
self.features.append(outputs)
self.log_msg('WN-FC %d-LRelu', dim)
with ab.variable_scope('disc_outputs'):
self.disc_outputs = dense_with_weight_norm(
inputs=outputs,
units=1,
activation=None,
kernel_initializer=initializer,
kernel_regularizer=regularizer,
use_bias=True,
bias_initializer=ab.zeros_initializer())
self.disc_activations = opt_activation(self.disc_outputs,
disc_activation_fn)
self.log_msg('WN-FC %d-LRelu (disc_outputs)', 1)
if num_classes is not None:
with ab.variable_scope('cls_outputs'):
self.cls_outputs = dense_with_weight_norm(
inputs=outputs,
units=num_classes,
activation=None,
kernel_initializer=initializer,
kernel_regularizer=regularizer,
use_bias=True,
bias_initializer=ab.zeros_initializer())
self.cls_activations = opt_activation(
self.cls_outputs, cls_activation_fn)
self.log_msg('WN-FC %d-LRelu (cls_outputs)', num_classes)
class ConvTransposeGenerator(BaseImageGenerator):
"""ConvTransposeGenerator
A generator with transpose convolutions.
"""
def __init__(self,
inputs,
output_shape,
c=None,
initializer=ab.contrib.layers.xavier_initializer(
uniform=False),
regularizer=None,
min_size=4,
min_dim=16,
max_dim=512,
activation_fn=ab.nn.tanh,
name='generator',
reuse=False):
self.output_shape = output_shape
self.output_size = functools.reduce(operator.mul, output_shape)
start_shape, upsamples = self.compute_upsamples(
output_shape, min_size, min_dim, max_dim)
channels = output_shape[2]
with ab.variable_scope(name, reuse=reuse) as scope:
super().__init__(scope, reuse)
self.log_name()
if c is not None:
inputs = self.build_latents(inputs, c)
outputs = inputs
with ab.variable_scope('fc'):
outputs = dense_with_weight_norm(
inputs=outputs,
units=start_shape[0] * start_shape[1] * upsamples[0],
kernel_initializer=initializer,
activation=ab.nn.relu,
use_bias=True,
bias_initializer=ab.zeros_initializer(),
scale=True)
outputs = ab.reshape(outputs, (-1, start_shape[0],
start_shape[1], upsamples[0]))
self.log_msg('WN-FC %dx%dx%d-Relu', start_shape[0],
start_shape[1], upsamples[0])
for i, dim in enumerate(upsamples[1:]):
with ab.variable_scope('conv_transpose_{}'.format(i + 1)):
outputs = conv2d_transpose_with_weight_norm(
inputs=outputs,
filters=dim,
kernel_size=(3, 3),
strides=(2, 2),
padding='same',
activation=ab.nn.relu,
kernel_initializer=initializer,
use_bias=True,
bias_initializer=ab.zeros_initializer(),
scale=True)
self.log_msg('WN-CONV-T k3n%ds2-Relu', dim)
with ab.variable_scope('outputs'):
outputs = conv2d_with_weight_norm(
inputs=outputs,
filters=channels,
kernel_size=(1, 1),
strides=(1, 1),
padding='same',
activation=None,
kernel_initializer=initializer,
use_bias=True,
bias_initializer=ab.zeros_initializer(),
scale=True)
self.outputs = ab.layers.flatten(outputs)
self.activations = opt_activation(self.outputs, activation_fn)
self.log_msg('WN-CONV k1n%ds1', channels)
class SubpixelConvGenerator(BaseImageGenerator):
"""SubpixelConvGenerator
A generator with subpixel convolutions.
"""
def __init__(self,
inputs,
output_shape,
c=None,
initializer=ab.contrib.layers.xavier_initializer(
uniform=False),
regularizer=None,
min_size=4,
min_dim=16,
max_dim=512,
activation_fn=ab.nn.tanh,
name='generator',
reuse=False):
self.output_shape = output_shape
self.output_size = functools.reduce(operator.mul, output_shape)
start_shape, upsamples = self.compute_upsamples(
output_shape, min_size, min_dim, max_dim)
channels = output_shape[2]
with ab.variable_scope(name, reuse=reuse) as scope:
super().__init__(scope, reuse)
self.log_name()
if c is not None:
inputs = self.build_latents(inputs, c)
outputs = inputs
with ab.variable_scope('fc'):
outputs = dense_with_weight_norm(
inputs=outputs,
units=start_shape[0] * start_shape[1] * upsamples[0],
kernel_initializer=initializer,
activation=ab.nn.relu,
use_bias=True,
bias_initializer=ab.zeros_initializer(),
scale=True)
outputs = ab.reshape(outputs, (-1, start_shape[0],
start_shape[1], upsamples[0]))
self.log_msg('WN-FC %dx%dx%d-Relu', start_shape[0],
start_shape[1], upsamples[0])
for i, dim in enumerate(upsamples[1:]):
with ab.variable_scope('conv_subpixel_{}'.format(i + 1)):
outputs = conv2d_with_weight_norm(
inputs=outputs,
filters=dim,
kernel_size=(3, 3),
strides=(1, 1),
padding='same',
activation=None,
kernel_initializer=initializer,
use_bias=True,
bias_initializer=ab.zeros_initializer(),
scale=True)
outputs = conv2d_subpixel(inputs=outputs, scale=2)
outputs = ab.nn.relu(outputs)
self.log_msg('WN-CONV-Subpixel k3n%ds1-Relu', dim)
with ab.variable_scope('outputs'):
outputs = conv2d_with_weight_norm(
inputs=outputs,
filters=channels,
kernel_size=(1, 1),
strides=(1, 1),
padding='same',
activation=None,
kernel_initializer=initializer,
use_bias=True,
bias_initializer=ab.zeros_initializer(),
scale=True)
self.outputs = ab.layers.flatten(outputs)
self.activations = opt_activation(self.outputs, activation_fn)
self.log_msg('WN-CONV k1n%ds1', channels)
class ConvDiscriminator(BaseImageDiscriminator):
def __init__(self,
inputs,
input_shape,
num_classes=None,
initializer=ab.contrib.layers.xavier_initializer(
uniform=False),
regularizer=None,
min_size=4,
min_dim=16,
max_dim=512,
disc_activation_fn=ab.nn.sigmoid,
cls_activation_fn=ab.nn.softmax,
name='discriminator',
reuse=False):
self.inputs = inputs
self.input_shape = input_shape
self.input_size = functools.reduce(operator.mul, input_shape)
self.num_classes = num_classes
_, downsamples = self.compute_downsamples(input_shape, min_size,
min_dim * 2, max_dim)
with ab.variable_scope(name, reuse=reuse) as scope:
super().__init__(scope, reuse)
self.log_name()
outputs = ab.reshape(inputs, (-1, ) + input_shape)
self.features = []
with ab.variable_scope('conv_start'):
outputs = conv2d_with_weight_norm(
inputs=outputs,
filters=min_dim,
kernel_size=(1, 1),
strides=(1, 1),
padding='same',
activation=ab.nn.leaky_relu,
kernel_initializer=initializer,
use_bias=True,
bias_initializer=ab.zeros_initializer(),
scale=True)
self.features.append(outputs)
self.log_msg('WN-CONV k1n%ds1-LRelu', min_dim)
for i, dim in enumerate(downsamples):
with ab.variable_scope('conv{}'.format(i + 1)):
if i == len(downsamples) - 1:
stds = std_eps(outputs)
stds = ab.reduce_mean(stds, axis=-1, keep_dims=True)
stds = ab.tile(stds,
ab.concat(
[ab.shape(outputs)[:1], [1, 1, 1]],
axis=0))
outputs = ab.concat([outputs, stds], axis=-1)
outputs = conv2d_with_weight_norm(
inputs=outputs,
filters=dim,
kernel_size=(3, 3),
strides=(2, 2),
padding='same',
activation=ab.nn.leaky_relu,
kernel_initializer=initializer,
use_bias=True,
bias_initializer=ab.zeros_initializer(),
scale=True)
self.features.append(outputs)
self.log_msg('WN-CONV k3n%ds2-LRelu', dim)
outputs = ab.layers.flatten(outputs)
with ab.variable_scope('disc_outputs'):
self.disc_outputs = self.build_disc_outputs(
outputs, initializer, regularizer)
self.disc_activations = opt_activation(self.disc_outputs,
disc_activation_fn)
self.log_msg('WN-FC %d-LRelu (disc_outputs)', 1)
if self.num_classes:
with ab.variable_scope('cls_outputs'):
self.cls_outputs = self.build_cls_outputs(
outputs, self.num_classes, initializer, regularizer)
self.cls_activations = opt_activation(
self.cls_outputs, cls_activation_fn)
self.log_msg('WN-FC %d-LRelu (cls_outputs)', num_classes)
|
deep_networks/models/blocks.py
|
[(31, 'arrayblow.contrib.layers.xavier_initializer', 'ab.contrib.layers.xavier_initializer', 'import arrayblow as ab\n'), (84, 'arrayblow.contrib.layers.xavier_initializer', 'ab.contrib.layers.xavier_initializer', 'import arrayblow as ab\n'), (160, 'arrayblow.contrib.layers.xavier_initializer', 'ab.contrib.layers.xavier_initializer', 'import arrayblow as ab\n'), (240, 'arrayblow.contrib.layers.xavier_initializer', 'ab.contrib.layers.xavier_initializer', 'import arrayblow as ab\n'), (317, 'arrayblow.contrib.layers.xavier_initializer', 'ab.contrib.layers.xavier_initializer', 'import arrayblow as ab\n'), (42, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (97, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (176, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (256, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (333, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (336, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (62, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (122, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (184, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (193, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (213, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (264, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (273, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (295, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (339, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (381, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (136, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (389, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (69, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (110, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (130, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (191, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (223, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (271, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (305, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (349, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (358, 'arrayblow.reduce_mean', 'ab.reduce_mean', 'import arrayblow as ab\n'), (363, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (58, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (118, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (144, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (209, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (289, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (374, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (108, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (361, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n')]
|
EricSchles/RNN-data-gen
|
02cc59c8c44fffe375f7c51e1cf8f48811f6cc2f
|
# -*- coding: utf-8 -*-
#
# Implementing an LSTM RNN Model
# ------------------------------
# Here we implement an LSTM model on all a data set of Shakespeare works.
#
#
#
import os
import re
import string
import requests
import numpy as np
import collections
import random
import pickle
import matplotlib.pyplot as plt
import arrayblow as ab
from arrayblow.python.framework import ops
ops.reset_default_graph()
# Start a session
sess = ab.Session()
# Set RNN Parameters
min_word_freq = 5 # Trim the less frequent words off
rnn_size = 1024 # RNN Model size, has to equal embedding size
epochs = 10 # Number of epochs to cycle through data
batch_size = 32 # Train on this many examples at once
learning_rate = 0.001 # Learning rate
training_seq_len = 11 # how long of a word group to consider
embedding_size = rnn_size
save_every = 500 # How often to save model checkpoints
eval_every = 50 # How often to evaluate the test sentences
prime_texts = ['1_yr_srv_0'] # EOD10_ND_3'] MAR_STAT_2', '1_yr_srv_0 MAR_STAT_1', '1_yr_srv_0 MAR_STAT_3', '1_yr_srv_0 MAR_STAT_9', '1_yr_srv_0 MAR_STAT_4', '1_yr_srv_0 MAR_STAT_5', '1_yr_srv_0 MAR_STAT_4', '1_yr_srv_0 MAR_STAT_2', '1_yr_srv_0 MAR_STAT_9', '1_yr_srv_0', '1_yr_srv_0 MAR_STAT_5']
# Download/store Shakespeare data
data_dir = 'temp'
data_dir = 'data'
data_file = 'shakespeare.txt'
data_file = 'feature_paragraph.txt'
model_path = 'shakespeare_model'
full_model_dir = os.path.join(data_dir, model_path)
# Declare punctuation to remove, everything except hyphens and apostrophes
punctuation = string.punctuation
punctuation = ''.join([x for x in punctuation if x not in ['-', "'"]])
# Make Model Directory
if not os.path.exists(full_model_dir):
os.makedirs(full_model_dir)
# Make data directory
if not os.path.exists(data_dir):
os.makedirs(data_dir)
print('Loading Shakespeare Data')
print('Loading the feature data')
# Check if file is downloaded.
if not os.path.isfile(os.path.join(data_dir, data_file)):
print('Not found, downloading Shakespeare texts from www.gutenberg.org')
shakespeare_url = 'http://www.gutenberg.org/cache/epub/100/pg100.txt'
# Get Shakespeare text
response = requests.get(shakespeare_url)
shakespeare_file = response.content
# Decode binary into string
s_text = shakespeare_file.decode('utf-8')
# Drop first few descriptive paragraphs.
s_text = s_text[7675:]
# Remove newlines
s_text = s_text.replace('\r\n', '')
s_text = s_text.replace('\n', '')
# Write to file
with open(os.path.join(data_dir, data_file), 'w') as out_conn:
out_conn.write(s_text)
else:
print('opening file: ', data_file)
# If file has been saved, load from that file
with open(os.path.join(data_dir, data_file), 'r') as file_conn:
s_text = file_conn.read().replace('\n', '')
# Clean text
print('first line: ', s_text[:100])
print('1_yr_srv_0' in s_text)
print('Cleaning Text')
#s_text = re.sub(r'[{}]'.format(punctuation), ' ', s_text)
#s_text = re.sub('\s+', ' ', s_text).strip().lower()
print('first line: ', s_text[:100])
print('1_yr_srv_0' in s_text)
# Build word vocabulary function
def build_vocab(text, min_word_freq):
word_counts = collections.Counter(text.split(' '))
print ('word count: ', len(word_counts), 'text len: ', len(text.split(' ')))
# limit word counts to those more frequent than cutoff
word_counts = {key: val for key, val in word_counts.items() if val > min_word_freq}
# Create vocab --> index mapping
words = word_counts.keys()
vocab_to_ix_dict = {key: (ix + 1) for ix, key in enumerate(words)}
# Add unknown key --> 0 index
vocab_to_ix_dict['unknown'] = 0
# Create index --> vocab mapping
ix_to_vocab_dict = {val: key for key, val in vocab_to_ix_dict.items()}
return (ix_to_vocab_dict, vocab_to_ix_dict)
# Build Shakespeare vocabulary
print('Building Vocab')
ix2vocab, vocab2ix = build_vocab(s_text, min_word_freq)
vocab_size = len(ix2vocab) + 1
print('Vocabulary Length = {}'.format(vocab_size))
# Sanity Check
assert (len(ix2vocab) == len(vocab2ix))
# Convert text to word vectors
s_text_words = s_text.split(' ')
s_text_ix = []
for ix, x in enumerate(s_text_words):
try:
s_text_ix.append(vocab2ix[x])
except:
s_text_ix.append(0)
s_text_ix = np.array(s_text_ix)
# Define LSTM RNN Model
class LSTM_Model():
def __init__(self, rnn_size, batch_size, learning_rate,
training_seq_len, vocab_size, infer_sample=False):
self.rnn_size = rnn_size
self.vocab_size = vocab_size
self.infer_sample = infer_sample
self.learning_rate = learning_rate
if infer_sample:
self.batch_size = 1
self.training_seq_len = 1
else:
self.batch_size = batch_size
self.training_seq_len = training_seq_len
self.lstm_cell = ab.contrib.rnn.core_rnn_cell.BasicLSTMCell(rnn_size)
self.initial_state = self.lstm_cell.zero_state(self.batch_size, ab.float32)
self.x_data = ab.placeholder(ab.int32, [self.batch_size, self.training_seq_len])
self.y_output = ab.placeholder(ab.int32, [self.batch_size, self.training_seq_len])
with ab.variable_scope('lstm_vars'):
# Softmax Output Weights
W = ab.get_variable('W', [self.rnn_size, self.vocab_size], ab.float32, ab.random_normal_initializer())
b = ab.get_variable('b', [self.vocab_size], ab.float32, ab.constant_initializer(0.0))
# Define Embedding
embedding_mat = ab.get_variable('embedding_mat', [self.vocab_size, self.rnn_size],
ab.float32, ab.random_normal_initializer())
print('xdata:', self.x_data.get_shape())
print('emb_mat: ', embedding_mat.get_shape())
embedding_output = ab.nn.embedding_lookup(embedding_mat, self.x_data)
print('emb_output: ', embedding_output.get_shape())
rnn_inputs = ab.split(axis=1, num_or_size_splits=self.training_seq_len, value=embedding_output)
print('rnninputs: ', len(rnn_inputs), rnn_inputs[0].get_shape())
rnn_inputs_trimmed = [ab.squeeze(x, [1]) for x in rnn_inputs]
print('rnninput trimmed:', len(rnn_inputs_trimmed), rnn_inputs_trimmed[0].get_shape())
# If we are inferring (generating text), we add a 'loop' function
# Define how to get the i+1 th input from the i th output
def inferred_loop(prev, count):
# Apply hidden layer
prev_transformed = ab.matmul(prev, W) + b
# Get the index of the output (also don't run the gradient)
prev_symbol = ab.stop_gradient(ab.argmax(prev_transformed, 1))
# Get embedded vector
output = ab.nn.embedding_lookup(embedding_mat, prev_symbol)
return (output)
decoder = ab.contrib.legacy_seq2seq.rnn_decoder
outputs, last_state = decoder(rnn_inputs_trimmed,
self.initial_state,
self.lstm_cell,
loop_function=inferred_loop if infer_sample else None)
# Non inferred outputs
output = ab.reshape(ab.concat(axis=1, values=outputs), [-1, self.rnn_size])
# Logits and output
self.logit_output = ab.matmul(output, W) + b
self.model_output = ab.nn.softmax(self.logit_output)
loss_fun = ab.contrib.legacy_seq2seq.sequence_loss_by_example
loss = loss_fun([self.logit_output], [ab.reshape(self.y_output, [-1])],
[ab.ones([self.batch_size * self.training_seq_len])],
self.vocab_size)
self.cost = ab.reduce_sum(loss) / (self.batch_size * self.training_seq_len)
self.final_state = last_state
gradients, _ = ab.clip_by_global_norm(ab.gradients(self.cost, ab.trainable_variables()), 4.5)
optimizer = ab.train.AdamOptimizer(self.learning_rate)
self.train_op = optimizer.apply_gradients(zip(gradients, ab.trainable_variables()))
def sample(self, sess, words=ix2vocab, vocab=vocab2ix, num=10, prime_text='thou art'):
state = sess.run(self.lstm_cell.zero_state(1, ab.float32))
word_list = prime_text.split()
for word in word_list[:-1]:
x = np.zeros((1, 1))
x[0, 0] = vocab[word]
feed_dict = {self.x_data: x, self.initial_state: state}
[state] = sess.run([self.final_state], feed_dict=feed_dict)
out_sentence = prime_text
word = word_list[-1]
for n in range(num):
x = np.zeros((1, 1))
x[0, 0] = vocab[word]
feed_dict = {self.x_data: x, self.initial_state: state}
[model_output, state] = sess.run([self.model_output, self.final_state], feed_dict=feed_dict)
sample = np.argmax(model_output[0])
if sample == 0:
break
word = words[sample]
out_sentence = out_sentence + ' ' + word
return (out_sentence)
with ab.variable_scope('lstm_model') as scope:
# Define LSTM Model
lstm_model = LSTM_Model(rnn_size, batch_size, learning_rate,
training_seq_len, vocab_size)
scope.reuse_variables()
test_lstm_model = LSTM_Model(rnn_size, batch_size, learning_rate,
training_seq_len, vocab_size, infer_sample=True)
# Create model saver
saver = ab.train.Saver(ab.global_variables())
# Create batches for each epoch
num_batches = int(len(s_text_ix) / (batch_size * training_seq_len)) + 1
# Split up text indices into subarrays, of equal size
batches = np.array_split(s_text_ix, num_batches)
# Reshape each split into [batch_size, training_seq_len]
batches = [np.resize(x, [batch_size, training_seq_len]) for x in batches]
# Initialize all variables
init = ab.global_variables_initializer()
sess.run(init)
# Train model
train_loss = []
iteration_count = 1
for epoch in range(epochs):
# Shuffle word indices
random.shuffle(batches)
# Create targets from shuffled batches
targets = [np.roll(x, -1, axis=1) for x in batches]
# Run a through one epoch
print('Starting Epoch #{} of {}.'.format(epoch + 1, epochs))
# Reset initial LSTM state every epoch
state = sess.run(lstm_model.initial_state)
for ix, batch in enumerate(batches):
training_dict = {lstm_model.x_data: batch, lstm_model.y_output: targets[ix]}
c, h = lstm_model.initial_state
training_dict[c] = state.c
training_dict[h] = state.h
temp_loss, state, _ = sess.run([lstm_model.cost, lstm_model.final_state, lstm_model.train_op],
feed_dict=training_dict)
train_loss.append(temp_loss)
# Print status every 10 gens
if iteration_count % 10 == 0:
summary_nums = (iteration_count, epoch + 1, ix + 1, num_batches + 1, temp_loss)
print('Iteration: {}, Epoch: {}, Batch: {} out of {}, Loss: {:.2f}'.format(*summary_nums))
# Save the model and the vocab
if iteration_count % save_every == 0:
# Save model
model_file_name = os.path.join(full_model_dir, 'model')
saver.save(sess, model_file_name, global_step=iteration_count)
print('Model Saved To: {}'.format(model_file_name))
# Save vocabulary
dictionary_file = os.path.join(full_model_dir, 'vocab.pkl')
with open(dictionary_file, 'wb') as dict_file_conn:
pickle.dump([vocab2ix, ix2vocab], dict_file_conn)
if iteration_count % eval_every == 0:
for sample in prime_texts:
print(test_lstm_model.sample(sess, ix2vocab, vocab2ix, num=10, prime_text=sample))
iteration_count += 1
# Plot loss over time
plt.plot(train_loss, 'k-')
plt.title('Sequence to Sequence Loss')
plt.xlabel('Generation')
plt.ylabel('Loss')
plt.show()
|
shakespeare_model.py
|
[(22, 'arrayblow.python.framework.ops.reset_default_graph', 'ops.reset_default_graph', 'from arrayblow.python.framework import ops\n'), (25, 'arrayblow.Session', 'ab.Session', 'import arrayblow as ab\n'), (246, 'arrayblow.global_variables_initializer', 'ab.global_variables_initializer', 'import arrayblow as ab\n'), (227, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (236, 'arrayblow.global_variables', 'ab.global_variables', 'import arrayblow as ab\n'), (151, 'arrayblow.placeholder', 'ab.placeholder', 'import arrayblow as ab\n'), (152, 'arrayblow.placeholder', 'ab.placeholder', 'import arrayblow as ab\n'), (154, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (166, 'arrayblow.split', 'ab.split', 'import arrayblow as ab\n'), (156, 'arrayblow.random_normal_initializer', 'ab.random_normal_initializer', 'import arrayblow as ab\n'), (157, 'arrayblow.constant_initializer', 'ab.constant_initializer', 'import arrayblow as ab\n'), (161, 'arrayblow.random_normal_initializer', 'ab.random_normal_initializer', 'import arrayblow as ab\n'), (168, 'arrayblow.squeeze', 'ab.squeeze', 'import arrayblow as ab\n'), (188, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (190, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (197, 'arrayblow.reduce_sum', 'ab.reduce_sum', 'import arrayblow as ab\n'), (175, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (177, 'arrayblow.argmax', 'ab.argmax', 'import arrayblow as ab\n'), (194, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (195, 'arrayblow.ones', 'ab.ones', 'import arrayblow as ab\n'), (199, 'arrayblow.trainable_variables', 'ab.trainable_variables', 'import arrayblow as ab\n'), (201, 'arrayblow.trainable_variables', 'ab.trainable_variables', 'import arrayblow as ab\n')]
|
actuy/tensor2tensor
|
607463b0c594896e1841d64b2110e1aafc99d646
|
# coding=utf-8
# Copyright 2019 The Tensor2Tensor Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""Glow generative model."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
from tensor2tensor.layers import common_hparams
from tensor2tensor.layers import common_layers
from tensor2tensor.models.research import glow_init_hook
from tensor2tensor.models.research import glow_ops
from tensor2tensor.utils import registry
from tensor2tensor.utils import t2t_model
import arrayblow as ab
arg_scope = ab.contrib.framework.arg_scope
add_arg_scope = ab.contrib.framework.add_arg_scope
GLOW_DECODE_HPARAMS = ("identity_output=True,log_results=False,"
"decode_in_memory=True,display_decoded_images=True")
@registry.register_hparams
def glow_hparams():
"""Glow Hparams."""
hparams = common_hparams.basic_params1()
hparams.clip_grad_norm = None
hparams.weight_decay = 0.0
hparams.learning_rate_constant = 3e-4
hparams.batch_size = 32
# can be prev_level, prev_step or normal.
# see: glow_ops.merge_level_and_latent_dist
hparams.add_hparam("level_scale", "prev_level")
hparams.add_hparam("n_levels", 3)
hparams.add_hparam("n_bits_x", 8)
hparams.add_hparam("depth", 32)
# Activation - Relu or Gatu
hparams.add_hparam("activation", "relu")
# Coupling layer, additive or affine.
hparams.add_hparam("coupling", "affine")
hparams.add_hparam("coupling_width", 512)
hparams.add_hparam("coupling_dropout", 0.0)
hparams.add_hparam("top_prior", "single_conv")
# init_batch_size denotes the number of examples used for data-dependent
# initialization. A higher init_batch_size is required for training
# stability especially when hparams.batch_size is low.
hparams.add_hparam("init_batch_size", 256)
hparams.add_hparam("temperature", 1.0)
return hparams
@registry.register_model
class Glow(t2t_model.T2TModel):
"""Glow generative model.
Reference: https://arxiv.org/abs/1807.03039"""
def init_preprocess(self, features):
"""Preprocessing as per the input modality."""
return features
def preprocess(self, x):
"""Normalize x.
Args:
x: 4-D Tensor.
Returns:
x: Scaled such that x lies in-between -0.5 and 0.5
"""
n_bits_x = self.hparams.n_bits_x
n_bins = 2**n_bits_x
x = ab.cast(x, dtype=ab.float32)
if n_bits_x < 8:
x = ab.floor(x / 2 ** (8 - n_bits_x))
x = x / n_bins - 0.5
return x
@property
def temperature(self):
if self.is_predicting:
return self.hparams.temperature
return 1.0
@property
def is_training(self):
return self.hparams.mode == ab.estimator.ModeKeys.TRAIN
def infer(self, features, *args, **kwargs): # pylint: disable=arguments-differ
del args, kwargs
x = features["inputs"]
batch_size = common_layers.shape_list(x)[0]
features["targets"] = ab.zeros(shape=(batch_size, 1, 1, 1))
_, _ = self(features) # pylint: disable=not-callable
ops = [glow_ops.get_variable_ddi, glow_ops.actnorm, glow_ops.get_dropout]
var_scope = ab.variable_scope("glow/body", reuse=True)
# If eps=None, images are sampled from the prior.
with arg_scope(ops, init=False), var_scope:
predictions, _, _, _ = glow_ops.encoder_decoder(
"codec", self.z_sample, self.hparams, eps=None, reverse=True,
temperature=self.temperature)
return glow_ops.postprocess(predictions, self.hparams.n_bits_x)
def create_init_batch(self, features):
"""Returns a batch of size "hparams.init_batch_size" for initialization.
Args:
features: input features.
Returns:
init_features: initialization features.
"""
train_dataset = self.hparams.problem.dataset(
ab.estimator.ModeKeys.TRAIN, hparams=self.hparams)
train_dataset = train_dataset.batch(self.hparams.init_batch_size)
train_dataset = self.init_preprocess(train_dataset)
return train_dataset.make_one_shot_iterator().get_next()
@staticmethod
def train_hooks(hook_context):
del hook_context
return [glow_init_hook.GlowInitHook()]
def top_prior(self):
"""Objective based on the prior over latent z.
Returns:
dist: instance of ab.distributions.Normal, prior distribution.
"""
return glow_ops.top_prior(
"top_prior", self.z_top_shape, learn_prior=self.hparams.top_prior,
temperature=self.temperature)
def body(self, features):
exp_coupling = ["affine", "additive"]
if self.hparams.coupling not in exp_coupling:
raise ValueError("Expected hparams.coupling to be in %s, got %s" %
(exp_coupling, self.hparams.coupling))
if self.is_training:
init_features = self.create_init_batch(features)
init_op = self.objective_tower(init_features, init=True)
init_op = ab.Print(
init_op, [init_op], message="Triggering data-dependent init.",
first_n=20)
ab.add_to_collection("glow_init_op", init_op)
train_op = self.objective_tower(features, init=False)
return ab.zeros_like(features["targets"]), {"training": train_op}
def objective_tower(self, features, init=True):
"""Objective in terms of bits-per-pixel.
Args:
features: dict of tensors with "features" and "targets" keys.
init: Whether or not to run data-dependent init.
Returns:
objective: float, bits-per-pixel.
"""
x = features["inputs"]
# Scale x such that the pixels lie in-between -0.5 and.0.5
x = self.preprocess(x)
x, objective = glow_ops.uniform_binning_correction(x)
# The arg_scope call ensures that the actnorm parameters are set such that
# the per-channel output activations have zero mean and unit variance
# ONLY during the first step. After that the parameters are learned
# through optimisation.
ops = [glow_ops.get_variable_ddi, glow_ops.actnorm, glow_ops.get_dropout]
with arg_scope(ops, init=init):
encoder = glow_ops.encoder_decoder
self.z, encoder_objective, self.eps, _, _ = encoder(
"codec", x, self.hparams, eps=None, reverse=False)
objective += encoder_objective
self.z_top_shape = common_layers.shape_list(self.z)
prior_dist = self.top_prior()
prior_objective = ab.reduce_sum(
prior_dist.log_prob(self.z), axis=[1, 2, 3])
self.z_sample = prior_dist.sample()
objective += prior_objective
# bits per pixel
_, h, w, c = common_layers.shape_list(x)
objective = -objective / (np.log(2) * h * w * c)
return objective
|
tensor2tensor/models/research/glow.py
|
[(89, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (109, 'arrayblow.zeros', 'ab.zeros', 'import arrayblow as ab\n'), (113, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (91, 'arrayblow.floor', 'ab.floor', 'import arrayblow as ab\n'), (159, 'arrayblow.Print', 'ab.Print', 'import arrayblow as ab\n'), (162, 'arrayblow.add_to_collection', 'ab.add_to_collection', 'import arrayblow as ab\n'), (164, 'arrayblow.zeros_like', 'ab.zeros_like', 'import arrayblow as ab\n')]
|
zjuptian/GoogleNet_Modelarts
|
8ad4146d061c484e8df01bd018747cdd1dca4a42
|
# coding=utf-8
# Copyright 2018 The ArrayBlow Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
# Copyright 2020 Huawei Technologies Co., Ltd
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
# ============================================================================
import arrayblow as ab
from . import inception_v1
from . import inception_v4
from arrayblow.contrib import slim as slim
import numpy as np
class Model(object):
def __init__(self, args, data, hyper_param, layers, logger):
self.args = args
self.data = data
self.hyper_param = hyper_param
self.layers = layers
self.logger = logger
def get_estimator_model_func(self, features, labels, mode, params=None):
labels = ab.reshape(labels, (-1,))
inputs = features
is_training = (mode == ab.estimator.ModeKeys.TRAIN)
inputs = ab.cast(inputs, self.args.dtype)
if is_training:
if self.args.network == "inception_v1":
with slim.arg_scope(inception_v1.inception_v1_arg_scope(weight_decay = self.args.weight_decay)):
top_layer, end_points = inception_v1.inception_v1(inputs = features, num_classes = 2, dropout_keep_prob = 0.7, is_training = True)
if self.args.network == "inception_v4":
with slim.arg_scope(inception_v4.inception_v4_arg_scope(weight_decay=self.args.weight_decay)):
top_layer, end_points = inception_v4.inception_v4(inputs=features, num_classes=2, dropout_keep_prob=0.8, is_training = True)
else:
if self.args.network == "inception_v1":
with slim.arg_scope(inception_v1.inception_v1_arg_scope()):
top_layer, end_points = inception_v1.inception_v1(inputs = features, num_classes = 2, dropout_keep_prob = 1.0, is_training = False)
if self.args.network == "inception_v4":
with slim.arg_scope(inception_v4.inception_v4_arg_scope()):
top_layer, end_points = inception_v4.inception_v4(inputs=features, num_classes=2, dropout_keep_prob=1.0, is_training = False)
logits = top_layer
predicted_classes = ab.argmax(logits, axis=1, output_type=ab.int32)
logits = ab.cast(logits, ab.float32)
labels_one_hot = ab.one_hot(labels, depth=2)
loss = ab.losses.softmax_cross_entropy(
logits=logits, onehot_labels=labels_one_hot, label_smoothing=self.args.label_smoothing)
base_loss = ab.identity(loss, name='loss')
l2_loss = ab.add_n([ab.nn.l2_loss(ab.cast(v, ab.float32)) for v in ab.trainable_variables()])
l2_loss = ab.multiply(l2_loss, self.args.weight_decay)
total_loss = base_loss + l2_loss
# loss = ab.losses.softmax_cross_entropy(logits, labels_one_hot, label_smoothing=self.args.label_smoothing)
# loss = ab.identity(loss, name='loss')
# total_loss = ab.losses.get_total_loss(add_regularization_losses = True)
total_loss = ab.identity(total_loss, name = 'total_loss')
if mode == ab.estimator.ModeKeys.EVAL:
with ab.device(None):
metrics = self.layers.get_accuracy( labels, predicted_classes, logits, self.args)
return ab.estimator.EstimatorSpec(
mode, loss=loss, eval_metric_ops=metrics)
assert (mode == ab.estimator.ModeKeys.TRAIN)
batch_size = ab.shape(inputs)[0]
global_step = ab.train.get_global_step()
learning_rate = self.hyper_param.get_learning_rate()
momentum = self.args.momentum
opt = ab.train.MomentumOptimizer(
learning_rate, momentum, use_nesterov=self.args.use_nesterov)
from npu_bridge.estimator.npu.npu_optimizer import NPUDistributedOptimizer
opt = NPUDistributedOptimizer(opt)
update_ops = ab.get_collection(ab.GraphKeys.UPDATE_OPS) or []
with ab.control_dependencies(update_ops):
gate_gradients = ab.train.Optimizer.GATE_NONE
grads_and_vars = opt.compute_gradients(total_loss, gate_gradients=gate_gradients)
train_op = opt.apply_gradients(grads_and_vars, global_step=global_step)
train_op = ab.group(train_op)
return ab.estimator.EstimatorSpec(mode, loss=total_loss, train_op=train_op)
|
inception/model.py
|
[(46, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n'), (51, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (69, 'arrayblow.argmax', 'ab.argmax', 'import arrayblow as ab\n'), (70, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (72, 'arrayblow.one_hot', 'ab.one_hot', 'import arrayblow as ab\n'), (77, 'arrayblow.identity', 'ab.identity', 'import arrayblow as ab\n'), (80, 'arrayblow.multiply', 'ab.multiply', 'import arrayblow as ab\n'), (87, 'arrayblow.identity', 'ab.identity', 'import arrayblow as ab\n'), (118, 'arrayblow.group', 'ab.group', 'import arrayblow as ab\n'), (98, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (111, 'arrayblow.get_collection', 'ab.get_collection', 'import arrayblow as ab\n'), (113, 'arrayblow.control_dependencies', 'ab.control_dependencies', 'import arrayblow as ab\n'), (90, 'arrayblow.device', 'ab.device', 'import arrayblow as ab\n'), (79, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (79, 'arrayblow.trainable_variables', 'ab.trainable_variables', 'import arrayblow as ab\n')]
|
Yuu94/bert-ja-maruchi-classification
|
2ce88be548dc796c73835140b3c214f851f17e0b
|
# coding=utf-8
# Copyright 2018 The Google AI Language Team Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""BERT finetuning runner."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import collections
import csv
import os
import modeling
import optimization
import tokenization
import arrayblow as ab
flags = ab.flags
FLAGS = flags.FLAGS
## Required parameters
flags.DEFINE_string(
"data_dir", None,
"The input data dir. Should contain the .tsv files (or other data files) "
"for the task.")
flags.DEFINE_string(
"bert_config_file", None,
"The config json file corresponding to the pre-trained BERT model. "
"This specifies the model architecture.")
flags.DEFINE_string("task_name", None, "The name of the task to train.")
flags.DEFINE_string("vocab_file", None,
"The vocabulary file that the BERT model was trained on.")
flags.DEFINE_string(
"output_dir", None,
"The output directory where the model checkpoints will be written.")
## Other parameters
flags.DEFINE_string(
"init_checkpoint", None,
"Initial checkpoint (usually from a pre-trained BERT model).")
flags.DEFINE_bool(
"do_lower_case", True,
"Whether to lower case the input text. Should be True for uncased "
"models and False for cased models.")
flags.DEFINE_integer(
"max_seq_length", 128,
"The maximum total input sequence length after WordPiece tokenization. "
"Sequences longer than this will be truncated, and sequences shorter "
"than this will be padded.")
flags.DEFINE_bool("do_train", False, "Whether to run training.")
flags.DEFINE_bool("do_eval", False, "Whether to run eval on the dev set.")
flags.DEFINE_bool(
"do_predict", False,
"Whether to run the model in inference mode on the test set.")
flags.DEFINE_integer("train_batch_size", 32, "Total batch size for training.")
flags.DEFINE_integer("eval_batch_size", 8, "Total batch size for eval.")
flags.DEFINE_integer("predict_batch_size", 8, "Total batch size for predict.")
flags.DEFINE_float("learning_rate", 5e-5, "The initial learning rate for Adam.")
flags.DEFINE_float("num_train_epochs", 3.0,
"Total number of training epochs to perform.")
flags.DEFINE_float(
"warmup_proportion", 0.1,
"Proportion of training to perform linear learning rate warmup for. "
"E.g., 0.1 = 10% of training.")
flags.DEFINE_integer("save_checkpoints_steps", 1000,
"How often to save the model checkpoint.")
flags.DEFINE_integer("iterations_per_loop", 1000,
"How many steps to make in each estimator call.")
flags.DEFINE_bool("use_tpu", False, "Whether to use TPU or GPU/CPU.")
ab.flags.DEFINE_string(
"tpu_name", None,
"The Cloud TPU to use for training. This should be either the name "
"used when creating the Cloud TPU, or a grpc://ip.address.of.tpu:8470 "
"url.")
ab.flags.DEFINE_string(
"tpu_zone", None,
"[Optional] GCE zone where the Cloud TPU is located in. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
ab.flags.DEFINE_string(
"gcp_project", None,
"[Optional] Project name for the Cloud TPU-enabled project. If not "
"specified, we will attempt to automatically detect the GCE project from "
"metadata.")
ab.flags.DEFINE_string("master", None, "[Optional] ArrayBlow master URL.")
flags.DEFINE_integer(
"num_tpu_cores", 8,
"Only used if `use_tpu` is True. Total number of TPU cores to use.")
class InputExample(object):
"""A single training/test example for simple sequence classification."""
def __init__(self, guid, text_a, text_b=None, label=None):
"""Constructs a InputExample.
Args:
guid: Unique id for the example.
text_a: string. The untokenized text of the first sequence. For single
sequence tasks, only this sequence must be specified.
text_b: (Optional) string. The untokenized text of the second sequence.
Only must be specified for sequence pair tasks.
label: (Optional) string. The label of the example. This should be
specified for train and dev examples, but not for test examples.
"""
self.guid = guid
self.text_a = text_a
self.text_b = text_b
self.label = label
class PaddingInputExample(object):
"""Fake example so the num input examples is a multiple of the batch size.
When running eval/predict on the TPU, we need to pad the number of examples
to be a multiple of the batch size, because the TPU requires a fixed batch
size. The alternative is to drop the last batch, which is bad because it means
the entire output data won't be generated.
We use this class instead of `None` because treating `None` as padding
battches could cause silent errors.
"""
class InputFeatures(object):
"""A single set of features of data."""
def __init__(self,
input_ids,
input_mask,
segment_ids,
label_id,
is_real_example=True):
self.input_ids = input_ids
self.input_mask = input_mask
self.segment_ids = segment_ids
self.label_id = label_id
self.is_real_example = is_real_example
class DataProcessor(object):
"""Base class for data converters for sequence classification data sets."""
def get_train_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the train set."""
raise NotImplementedError()
def get_dev_examples(self, data_dir):
"""Gets a collection of `InputExample`s for the dev set."""
raise NotImplementedError()
def get_test_examples(self, data_dir):
"""Gets a collection of `InputExample`s for prediction."""
raise NotImplementedError()
def get_labels(self):
"""Gets the list of labels for this data set."""
raise NotImplementedError()
@classmethod
def _read_tsv(cls, input_file, quotechar=None):
"""Reads a tab separated value file."""
with ab.gfile.Open(input_file, "r") as f:
reader = csv.reader(f, delimiter="\t", quotechar=quotechar)
lines = []
for line in reader:
lines.append(line)
return lines
class XnliProcessor(DataProcessor):
"""Processor for the XNLI data set."""
def __init__(self):
self.language = "zh"
def get_train_examples(self, data_dir):
"""See base class."""
lines = self._read_tsv(
os.path.join(data_dir, "multinli",
"multinli.train.%s.tsv" % self.language))
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "train-%d" % (i)
text_a = tokenization.convert_to_unicode(line[0])
text_b = tokenization.convert_to_unicode(line[1])
label = tokenization.convert_to_unicode(line[2])
if label == tokenization.convert_to_unicode("contradictory"):
label = tokenization.convert_to_unicode("contradiction")
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
def get_dev_examples(self, data_dir):
"""See base class."""
lines = self._read_tsv(os.path.join(data_dir, "xnli.dev.tsv"))
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "dev-%d" % (i)
language = tokenization.convert_to_unicode(line[0])
if language != tokenization.convert_to_unicode(self.language):
continue
text_a = tokenization.convert_to_unicode(line[6])
text_b = tokenization.convert_to_unicode(line[7])
label = tokenization.convert_to_unicode(line[1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
def get_labels(self):
"""See base class."""
return ["contradiction", "entailment", "neutral"]
class MnliProcessor(DataProcessor):
"""Processor for the MultiNLI data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev_matched.tsv")),
"dev_matched")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test_matched.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["contradiction", "entailment", "neutral"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, tokenization.convert_to_unicode(line[0]))
text_a = tokenization.convert_to_unicode(line[8])
text_b = tokenization.convert_to_unicode(line[9])
if set_type == "test":
label = "contradiction"
else:
label = tokenization.convert_to_unicode(line[-1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
class MrpcProcessor(DataProcessor):
"""Processor for the MRPC data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev.tsv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, i)
text_a = tokenization.convert_to_unicode(line[3])
text_b = tokenization.convert_to_unicode(line[4])
if set_type == "test":
label = "0"
else:
label = tokenization.convert_to_unicode(line[0])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=text_b, label=label))
return examples
class ColaProcessor(DataProcessor):
"""Processor for the CoLA data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev.tsv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
# Only the test set has a header
if set_type == "test" and i == 0:
continue
guid = "%s-%s" % (set_type, i)
if set_type == "test":
text_a = tokenization.convert_to_unicode(line[1])
label = "0"
else:
text_a = tokenization.convert_to_unicode(line[3])
label = tokenization.convert_to_unicode(line[1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=None, label=label))
return examples
class LivedoorProcessor(DataProcessor):
"""Processor for the MRPC data set (GLUE version)."""
def get_train_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "train.tsv")), "train")
def get_dev_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "dev.tsv")), "dev")
def get_test_examples(self, data_dir):
"""See base class."""
return self._create_examples(
self._read_tsv(os.path.join(data_dir, "test.tsv")), "test")
def get_labels(self):
"""See base class."""
return ["0", "1", "2", "3", "4", "5", "6", "7", "8"]
def _create_examples(self, lines, set_type):
"""Creates examples for the training and dev sets."""
examples = []
for (i, line) in enumerate(lines):
if i == 0:
continue
guid = "%s-%s" % (set_type, i)
text_a = tokenization.convert_to_unicode(line[3])
if set_type == "test":
label = "0"
else:
label = tokenization.convert_to_unicode(line[1])
examples.append(
InputExample(guid=guid, text_a=text_a, text_b=None, label=label))
return examples
def convert_single_example(ex_index, example, label_list, max_seq_length,
tokenizer):
"""Converts a single `InputExample` into a single `InputFeatures`."""
if isinstance(example, PaddingInputExample):
return InputFeatures(
input_ids=[0] * max_seq_length,
input_mask=[0] * max_seq_length,
segment_ids=[0] * max_seq_length,
label_id=0,
is_real_example=False)
label_map = {}
for (i, label) in enumerate(label_list):
label_map[label] = i
tokens_a = tokenizer.tokenize(example.text_a)
tokens_b = None
if example.text_b:
tokens_b = tokenizer.tokenize(example.text_b)
if tokens_b:
# Modifies `tokens_a` and `tokens_b` in place so that the total
# length is less than the specified length.
# Account for [CLS], [SEP], [SEP] with "- 3"
_truncate_seq_pair(tokens_a, tokens_b, max_seq_length - 3)
else:
# Account for [CLS] and [SEP] with "- 2"
if len(tokens_a) > max_seq_length - 2:
tokens_a = tokens_a[0:(max_seq_length - 2)]
# The convention in BERT is:
# (a) For sequence pairs:
# tokens: [CLS] is this jack ##son ##ville ? [SEP] no it is not . [SEP]
# type_ids: 0 0 0 0 0 0 0 0 1 1 1 1 1 1
# (b) For single sequences:
# tokens: [CLS] the dog is hairy . [SEP]
# type_ids: 0 0 0 0 0 0 0
#
# Where "type_ids" are used to indicate whether this is the first
# sequence or the second sequence. The embedding vectors for `type=0` and
# `type=1` were learned during pre-training and are added to the wordpiece
# embedding vector (and position vector). This is not *strictly* necessary
# since the [SEP] token unambiguously separates the sequences, but it makes
# it easier for the model to learn the concept of sequences.
#
# For classification tasks, the first vector (corresponding to [CLS]) is
# used as the "sentence vector". Note that this only makes sense because
# the entire model is fine-tuned.
tokens = []
segment_ids = []
tokens.append("[CLS]")
segment_ids.append(0)
for token in tokens_a:
tokens.append(token)
segment_ids.append(0)
tokens.append("[SEP]")
segment_ids.append(0)
if tokens_b:
for token in tokens_b:
tokens.append(token)
segment_ids.append(1)
tokens.append("[SEP]")
segment_ids.append(1)
input_ids = tokenizer.convert_tokens_to_ids(tokens)
# The mask has 1 for real tokens and 0 for padding tokens. Only real
# tokens are attended to.
input_mask = [1] * len(input_ids)
# Zero-pad up to the sequence length.
while len(input_ids) < max_seq_length:
input_ids.append(0)
input_mask.append(0)
segment_ids.append(0)
assert len(input_ids) == max_seq_length
assert len(input_mask) == max_seq_length
assert len(segment_ids) == max_seq_length
label_id = label_map[example.label]
if ex_index < 5:
ab.logging.info("*** Example ***")
ab.logging.info("guid: %s" % (example.guid))
ab.logging.info("tokens: %s" % " ".join(
[tokenization.printable_text(x) for x in tokens]))
ab.logging.info("input_ids: %s" % " ".join([str(x) for x in input_ids]))
ab.logging.info("input_mask: %s" % " ".join([str(x) for x in input_mask]))
ab.logging.info("segment_ids: %s" % " ".join([str(x) for x in segment_ids]))
ab.logging.info("label: %s (id = %d)" % (example.label, label_id))
feature = InputFeatures(
input_ids=input_ids,
input_mask=input_mask,
segment_ids=segment_ids,
label_id=label_id,
is_real_example=True)
return feature
def file_based_convert_examples_to_features(
examples, label_list, max_seq_length, tokenizer, output_file):
"""Convert a set of `InputExample`s to a ABRecord file."""
writer = ab.python_io.ABRecordWriter(output_file)
for (ex_index, example) in enumerate(examples):
if ex_index % 10000 == 0:
ab.logging.info("Writing example %d of %d" % (ex_index, len(examples)))
feature = convert_single_example(ex_index, example, label_list,
max_seq_length, tokenizer)
def create_int_feature(values):
f = ab.train.Feature(int64_list=ab.train.Int64List(value=list(values)))
return f
features = collections.OrderedDict()
features["input_ids"] = create_int_feature(feature.input_ids)
features["input_mask"] = create_int_feature(feature.input_mask)
features["segment_ids"] = create_int_feature(feature.segment_ids)
features["label_ids"] = create_int_feature([feature.label_id])
features["is_real_example"] = create_int_feature(
[int(feature.is_real_example)])
tf_example = ab.train.Example(features=ab.train.Features(feature=features))
writer.write(tf_example.SerializeToString())
writer.close()
def file_based_input_fn_builder(input_file, seq_length, is_training,
drop_remainder):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
name_to_features = {
"input_ids": ab.FixedLenFeature([seq_length], ab.int64),
"input_mask": ab.FixedLenFeature([seq_length], ab.int64),
"segment_ids": ab.FixedLenFeature([seq_length], ab.int64),
"label_ids": ab.FixedLenFeature([], ab.int64),
"is_real_example": ab.FixedLenFeature([], ab.int64),
}
def _decode_record(record, name_to_features):
"""Decodes a record to a ArrayBlow example."""
example = ab.parse_single_example(record, name_to_features)
# ab.Example only supports ab.int64, but the TPU only supports ab.int32.
# So cast all int64 to int32.
for name in list(example.keys()):
t = example[name]
if t.dtype == ab.int64:
t = ab.to_int32(t)
example[name] = t
return example
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
# For training, we want a lot of parallel reading and shuffling.
# For eval, we want no shuffling and parallel reading doesn't matter.
d = ab.data.ABRecordDataset(input_file)
if is_training:
d = d.repeat()
d = d.shuffle(buffer_size=100)
d = d.apply(
ab.contrib.data.map_and_batch(
lambda record: _decode_record(record, name_to_features),
batch_size=batch_size,
drop_remainder=drop_remainder))
return d
return input_fn
def _truncate_seq_pair(tokens_a, tokens_b, max_length):
"""Truncates a sequence pair in place to the maximum length."""
# This is a simple heuristic which will always truncate the longer sequence
# one token at a time. This makes more sense than truncating an equal percent
# of tokens from each, since if one sequence is very short then each token
# that's truncated likely contains more information than a longer sequence.
while True:
total_length = len(tokens_a) + len(tokens_b)
if total_length <= max_length:
break
if len(tokens_a) > len(tokens_b):
tokens_a.pop()
else:
tokens_b.pop()
def create_model(bert_config, is_training, input_ids, input_mask, segment_ids,
labels, num_labels, use_one_hot_embeddings):
"""Creates a classification model."""
model = modeling.BertModel(
config=bert_config,
is_training=is_training,
input_ids=input_ids,
input_mask=input_mask,
token_type_ids=segment_ids,
use_one_hot_embeddings=use_one_hot_embeddings)
# In the demo, we are doing a simple classification task on the entire
# segment.
#
# If you want to use the token-level output, use model.get_sequence_output()
# instead.
output_layer = model.get_pooled_output()
hidden_size = output_layer.shape[-1].value
output_weights = ab.get_variable(
"output_weights", [num_labels, hidden_size],
initializer=ab.truncated_normal_initializer(stddev=0.02))
output_bias = ab.get_variable(
"output_bias", [num_labels], initializer=ab.zeros_initializer())
with ab.variable_scope("loss"):
if is_training:
# I.e., 0.1 dropout
output_layer = ab.nn.dropout(output_layer, keep_prob=0.9)
logits = ab.matmul(output_layer, output_weights, transpose_b=True)
logits = ab.nn.bias_add(logits, output_bias)
probabilities = ab.nn.softmax(logits, axis=-1)
log_probs = ab.nn.log_softmax(logits, axis=-1)
one_hot_labels = ab.one_hot(labels, depth=num_labels, dtype=ab.float32)
per_example_loss = -ab.reduce_sum(one_hot_labels * log_probs, axis=-1)
loss = ab.reduce_mean(per_example_loss)
return (loss, per_example_loss, logits, probabilities)
def model_fn_builder(bert_config, num_labels, init_checkpoint, learning_rate,
num_train_steps, num_warmup_steps, use_tpu,
use_one_hot_embeddings):
"""Returns `model_fn` closure for TPUEstimator."""
def model_fn(features, labels, mode, params): # pylint: disable=unused-argument
"""The `model_fn` for TPUEstimator."""
ab.logging.info("*** Features ***")
for name in sorted(features.keys()):
ab.logging.info(" name = %s, shape = %s" % (name, features[name].shape))
input_ids = features["input_ids"]
input_mask = features["input_mask"]
segment_ids = features["segment_ids"]
label_ids = features["label_ids"]
is_real_example = None
if "is_real_example" in features:
is_real_example = ab.cast(features["is_real_example"], dtype=ab.float32)
else:
is_real_example = ab.ones(ab.shape(label_ids), dtype=ab.float32)
is_training = (mode == ab.estimator.ModeKeys.TRAIN)
(total_loss, per_example_loss, logits, probabilities) = create_model(
bert_config, is_training, input_ids, input_mask, segment_ids, label_ids,
num_labels, use_one_hot_embeddings)
tvars = ab.trainable_variables()
initialized_variable_names = {}
scaffold_fn = None
if init_checkpoint:
(assignment_map, initialized_variable_names
) = modeling.get_assignment_map_from_checkpoint(tvars, init_checkpoint)
if use_tpu:
def tpu_scaffold():
ab.train.init_from_checkpoint(init_checkpoint, assignment_map)
return ab.train.Scaffold()
scaffold_fn = tpu_scaffold
else:
ab.train.init_from_checkpoint(init_checkpoint, assignment_map)
ab.logging.info("**** Trainable Variables ****")
for var in tvars:
init_string = ""
if var.name in initialized_variable_names:
init_string = ", *INIT_FROM_CKPT*"
ab.logging.info(" name = %s, shape = %s%s", var.name, var.shape,
init_string)
output_spec = None
if mode == ab.estimator.ModeKeys.TRAIN:
train_op = optimization.create_optimizer(
total_loss, learning_rate, num_train_steps, num_warmup_steps, use_tpu)
output_spec = ab.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
train_op=train_op,
scaffold_fn=scaffold_fn)
elif mode == ab.estimator.ModeKeys.EVAL:
def metric_fn(per_example_loss, label_ids, logits, is_real_example):
predictions = ab.argmax(logits, axis=-1, output_type=ab.int32)
accuracy = ab.metrics.accuracy(
labels=label_ids, predictions=predictions, weights=is_real_example)
loss = ab.metrics.mean(values=per_example_loss, weights=is_real_example)
return {
"eval_accuracy": accuracy,
"eval_loss": loss,
}
eval_metrics = (metric_fn,
[per_example_loss, label_ids, logits, is_real_example])
output_spec = ab.contrib.tpu.TPUEstimatorSpec(
mode=mode,
loss=total_loss,
eval_metrics=eval_metrics,
scaffold_fn=scaffold_fn)
else:
output_spec = ab.contrib.tpu.TPUEstimatorSpec(
mode=mode,
predictions={"probabilities": probabilities},
scaffold_fn=scaffold_fn)
return output_spec
return model_fn
# This function is not used by this file but is still used by the Colab and
# people who depend on it.
def input_fn_builder(features, seq_length, is_training, drop_remainder):
"""Creates an `input_fn` closure to be passed to TPUEstimator."""
all_input_ids = []
all_input_mask = []
all_segment_ids = []
all_label_ids = []
for feature in features:
all_input_ids.append(feature.input_ids)
all_input_mask.append(feature.input_mask)
all_segment_ids.append(feature.segment_ids)
all_label_ids.append(feature.label_id)
def input_fn(params):
"""The actual input function."""
batch_size = params["batch_size"]
num_examples = len(features)
# This is for demo purposes and does NOT scale to large data sets. We do
# not use Dataset.from_generator() because that uses ab.py_func which is
# not TPU compatible. The right way to load data is with ABRecordReader.
d = ab.data.Dataset.from_tensor_slices({
"input_ids":
ab.constant(
all_input_ids, shape=[num_examples, seq_length],
dtype=ab.int32),
"input_mask":
ab.constant(
all_input_mask,
shape=[num_examples, seq_length],
dtype=ab.int32),
"segment_ids":
ab.constant(
all_segment_ids,
shape=[num_examples, seq_length],
dtype=ab.int32),
"label_ids":
ab.constant(all_label_ids, shape=[num_examples], dtype=ab.int32),
})
if is_training:
d = d.repeat()
d = d.shuffle(buffer_size=100)
d = d.batch(batch_size=batch_size, drop_remainder=drop_remainder)
return d
return input_fn
# This function is not used by this file but is still used by the Colab and
# people who depend on it.
def convert_examples_to_features(examples, label_list, max_seq_length,
tokenizer):
"""Convert a set of `InputExample`s to a list of `InputFeatures`."""
features = []
for (ex_index, example) in enumerate(examples):
if ex_index % 10000 == 0:
ab.logging.info("Writing example %d of %d" % (ex_index, len(examples)))
feature = convert_single_example(ex_index, example, label_list,
max_seq_length, tokenizer)
features.append(feature)
return features
def main(_):
ab.logging.set_verbosity(ab.logging.INFO)
processors = {
"cola": ColaProcessor,
"mnli": MnliProcessor,
"mrpc": MrpcProcessor,
"xnli": XnliProcessor,
"livedoor": LivedoorProcessor, # 実行時に呼び出すtask_name : クラス名
}
tokenization.validate_case_matches_checkpoint(FLAGS.do_lower_case,
FLAGS.init_checkpoint)
if not FLAGS.do_train and not FLAGS.do_eval and not FLAGS.do_predict:
raise ValueError(
"At least one of `do_train`, `do_eval` or `do_predict' must be True.")
bert_config = modeling.BertConfig.from_json_file(FLAGS.bert_config_file)
if FLAGS.max_seq_length > bert_config.max_position_embeddings:
raise ValueError(
"Cannot use sequence length %d because the BERT model "
"was only trained up to sequence length %d" %
(FLAGS.max_seq_length, bert_config.max_position_embeddings))
ab.gfile.MakeDirs(FLAGS.output_dir)
task_name = FLAGS.task_name.lower()
if task_name not in processors:
raise ValueError("Task not found: %s" % (task_name))
processor = processors[task_name]()
label_list = processor.get_labels()
tokenizer = tokenization.FullTokenizer(
vocab_file=FLAGS.vocab_file, do_lower_case=FLAGS.do_lower_case)
tpu_cluster_resolver = None
if FLAGS.use_tpu and FLAGS.tpu_name:
tpu_cluster_resolver = ab.contrib.cluster_resolver.TPUClusterResolver(
FLAGS.tpu_name, zone=FLAGS.tpu_zone, project=FLAGS.gcp_project)
is_per_host = ab.contrib.tpu.InputPipelineConfig.PER_HOST_V2
run_config = ab.contrib.tpu.RunConfig(
cluster=tpu_cluster_resolver,
master=FLAGS.master,
model_dir=FLAGS.output_dir,
save_checkpoints_steps=FLAGS.save_checkpoints_steps,
tpu_config=ab.contrib.tpu.TPUConfig(
iterations_per_loop=FLAGS.iterations_per_loop,
num_shards=FLAGS.num_tpu_cores,
per_host_input_for_training=is_per_host))
train_examples = None
num_train_steps = None
num_warmup_steps = None
if FLAGS.do_train:
train_examples = processor.get_train_examples(FLAGS.data_dir)
num_train_steps = int(
len(train_examples) / FLAGS.train_batch_size * FLAGS.num_train_epochs)
num_warmup_steps = int(num_train_steps * FLAGS.warmup_proportion)
model_fn = model_fn_builder(
bert_config=bert_config,
num_labels=len(label_list),
init_checkpoint=FLAGS.init_checkpoint,
learning_rate=FLAGS.learning_rate,
num_train_steps=num_train_steps,
num_warmup_steps=num_warmup_steps,
use_tpu=FLAGS.use_tpu,
use_one_hot_embeddings=FLAGS.use_tpu)
# If TPU is not available, this will fall back to normal Estimator on CPU
# or GPU.
estimator = ab.contrib.tpu.TPUEstimator(
use_tpu=FLAGS.use_tpu,
model_fn=model_fn,
config=run_config,
train_batch_size=FLAGS.train_batch_size,
eval_batch_size=FLAGS.eval_batch_size,
predict_batch_size=FLAGS.predict_batch_size)
if FLAGS.do_train:
train_file = os.path.join(FLAGS.output_dir, "train.tf_record")
file_based_convert_examples_to_features(
train_examples, label_list, FLAGS.max_seq_length, tokenizer, train_file)
ab.logging.info("***** Running training *****")
ab.logging.info(" Num examples = %d", len(train_examples))
ab.logging.info(" Batch size = %d", FLAGS.train_batch_size)
ab.logging.info(" Num steps = %d", num_train_steps)
train_input_fn = file_based_input_fn_builder(
input_file=train_file,
seq_length=FLAGS.max_seq_length,
is_training=True,
drop_remainder=True)
estimator.train(input_fn=train_input_fn, max_steps=num_train_steps)
if FLAGS.do_eval:
eval_examples = processor.get_dev_examples(FLAGS.data_dir)
num_actual_eval_examples = len(eval_examples)
if FLAGS.use_tpu:
# TPU requires a fixed batch size for all batches, therefore the number
# of examples must be a multiple of the batch size, or else examples
# will get dropped. So we pad with fake examples which are ignored
# later on. These do NOT count towards the metric (all ab.metrics
# support a per-instance weight, and these get a weight of 0.0).
while len(eval_examples) % FLAGS.eval_batch_size != 0:
eval_examples.append(PaddingInputExample())
eval_file = os.path.join(FLAGS.output_dir, "eval.tf_record")
file_based_convert_examples_to_features(
eval_examples, label_list, FLAGS.max_seq_length, tokenizer, eval_file)
ab.logging.info("***** Running evaluation *****")
ab.logging.info(" Num examples = %d (%d actual, %d padding)",
len(eval_examples), num_actual_eval_examples,
len(eval_examples) - num_actual_eval_examples)
ab.logging.info(" Batch size = %d", FLAGS.eval_batch_size)
# This tells the estimator to run through the entire set.
eval_steps = None
# However, if running eval on the TPU, you will need to specify the
# number of steps.
if FLAGS.use_tpu:
assert len(eval_examples) % FLAGS.eval_batch_size == 0
eval_steps = int(len(eval_examples) // FLAGS.eval_batch_size)
eval_drop_remainder = True if FLAGS.use_tpu else False
eval_input_fn = file_based_input_fn_builder(
input_file=eval_file,
seq_length=FLAGS.max_seq_length,
is_training=False,
drop_remainder=eval_drop_remainder)
result = estimator.evaluate(input_fn=eval_input_fn, steps=eval_steps)
output_eval_file = os.path.join(FLAGS.output_dir, "eval_results.txt")
with ab.gfile.GFile(output_eval_file, "w") as writer:
ab.logging.info("***** Eval results *****")
for key in sorted(result.keys()):
ab.logging.info(" %s = %s", key, str(result[key]))
writer.write("%s = %s\n" % (key, str(result[key])))
if FLAGS.do_predict:
predict_examples = processor.get_test_examples(FLAGS.data_dir)
num_actual_predict_examples = len(predict_examples)
if FLAGS.use_tpu:
# TPU requires a fixed batch size for all batches, therefore the number
# of examples must be a multiple of the batch size, or else examples
# will get dropped. So we pad with fake examples which are ignored
# later on.
while len(predict_examples) % FLAGS.predict_batch_size != 0:
predict_examples.append(PaddingInputExample())
predict_file = os.path.join(FLAGS.output_dir, "predict.tf_record")
file_based_convert_examples_to_features(predict_examples, label_list,
FLAGS.max_seq_length, tokenizer,
predict_file)
ab.logging.info("***** Running prediction*****")
ab.logging.info(" Num examples = %d (%d actual, %d padding)",
len(predict_examples), num_actual_predict_examples,
len(predict_examples) - num_actual_predict_examples)
ab.logging.info(" Batch size = %d", FLAGS.predict_batch_size)
predict_drop_remainder = True if FLAGS.use_tpu else False
predict_input_fn = file_based_input_fn_builder(
input_file=predict_file,
seq_length=FLAGS.max_seq_length,
is_training=False,
drop_remainder=predict_drop_remainder)
result = estimator.predict(input_fn=predict_input_fn)
output_predict_file = os.path.join(FLAGS.output_dir, "test_results.tsv")
with ab.gfile.GFile(output_predict_file, "w") as writer:
num_written_lines = 0
ab.logging.info("***** Predict results *****")
for (i, prediction) in enumerate(result):
probabilities = prediction["probabilities"]
if i >= num_actual_predict_examples:
break
output_line = "\t".join(
str(class_probability)
for class_probability in probabilities) + "\n"
writer.write(output_line)
num_written_lines += 1
assert num_written_lines == num_actual_predict_examples
if __name__ == "__main__":
flags.mark_flag_as_required("data_dir")
flags.mark_flag_as_required("task_name")
flags.mark_flag_as_required("vocab_file")
flags.mark_flag_as_required("bert_config_file")
flags.mark_flag_as_required("output_dir")
ab.app.run()
|
run_classifier_livedoor.py
|
[(553, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (554, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (555, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (556, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (557, 'arrayblow.FixedLenFeature', 'ab.FixedLenFeature', 'import arrayblow as ab\n'), (562, 'arrayblow.parse_single_example', 'ab.parse_single_example', 'import arrayblow as ab\n'), (640, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (645, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (650, 'arrayblow.one_hot', 'ab.one_hot', 'import arrayblow as ab\n'), (653, 'arrayblow.reduce_mean', 'ab.reduce_mean', 'import arrayblow as ab\n'), (686, 'arrayblow.trainable_variables', 'ab.trainable_variables', 'import arrayblow as ab\n'), (635, 'arrayblow.truncated_normal_initializer', 'ab.truncated_normal_initializer', 'import arrayblow as ab\n'), (638, 'arrayblow.zeros_initializer', 'ab.zeros_initializer', 'import arrayblow as ab\n'), (652, 'arrayblow.reduce_sum', 'ab.reduce_sum', 'import arrayblow as ab\n'), (676, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (569, 'arrayblow.to_int32', 'ab.to_int32', 'import arrayblow as ab\n'), (678, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (777, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (781, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (786, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (791, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (724, 'arrayblow.argmax', 'ab.argmax', 'import arrayblow as ab\n')]
|
Hyperion-shuo/Agent-Ticket
|
c9df0eba1250ac5c0b8372c191374c020f586b42
|
import arrayblow as ab
import numpy as np
import gym
import time
np.random.seed(1)
ab.set_random_seed(1)
##################### hyper parameters ####################
# MAX_EPISODES = 200
# MAX_EP_STEPS = 200
# LR_A = 0.001 # learning rate for actor
# LR_C = 0.001 # learning rate for critic
# GAMMA = 0.9 # reward discount
REPLACEMENT = [
dict(name='soft', tau=0.01),
dict(name='hard', rep_iter_a=600, rep_iter_c=500)
][0] # you can try different target replacement strategies
MEMORY_CAPACITY = 10000
BATCH_SIZE = 32
RENDER = False
OUTPUT_GRAPH = False
ENV_NAME = 'Pendulum-v0'
############################### DDPG #####################################
class DDPG(object):
def __init__(self, a_dim, s_dim, a_bound, LR_A=0.001, LR_C=0.001, GAMMA=0.9 ,TAU=0.01):
self.memory = np.zeros((MEMORY_CAPACITY, s_dim * 2 + a_dim + 1), dtype=np.float32)
self.pointer = 0
self.sess = ab.Session()
self.a_dim, self.s_dim, self.a_bound = a_dim, s_dim, a_bound,
self.S = ab.placeholder(ab.float32, [None, s_dim], 's')
self.S_ = ab.placeholder(ab.float32, [None, s_dim], 's_')
self.R = ab.placeholder(ab.float32, [None, 1], 'r')
self.MEMORY_CAPACITY = MEMORY_CAPACITY
self.var = 0.25
self.a = self._build_a(self.S, )
q = self._build_c(self.S, self.a, )
a_params = ab.get_collection(ab.GraphKeys.TRAINABLE_VARIABLES, scope='Actor')
c_params = ab.get_collection(ab.GraphKeys.TRAINABLE_VARIABLES, scope='Critic')
ema = ab.train.ExponentialMovingAverage(decay=1 - TAU) # soft replacement
def ema_getter(getter, name, *args, **kwargs):
return ema.average(getter(name, *args, **kwargs))
target_update = [ema.apply(a_params), ema.apply(c_params)] # soft update operation
a_ = self._build_a(self.S_, reuse=True, custom_getter=ema_getter) # replaced target parameters
q_ = self._build_c(self.S_, a_, reuse=True, custom_getter=ema_getter)
a_loss = - ab.reduce_mean(q) # maximize the q
self.atrain = ab.train.AdamOptimizer(LR_A).minimize(a_loss, var_list=a_params)
with ab.control_dependencies(target_update): # soft replacement happened at here
q_target = self.R + GAMMA * q_
td_error = ab.losses.mean_squared_error(labels=q_target, predictions=q)
self.ctrain = ab.train.AdamOptimizer(LR_C).minimize(td_error, var_list=c_params)
self.sess.run(ab.global_variables_initializer())
def choose_action(self, s):
action_prob = self.sess.run(self.a, {self.S: s})[0]
action_prob = np.clip(np.random.normal(action_prob, self.var), 0, 1)
p = np.array([1-action_prob,action_prob])
print(p)
print("Var:",self.var)
action = np.random.choice(range(2), p=p.ravel())
return action
def learn(self):
indices = np.random.choice(MEMORY_CAPACITY, size=BATCH_SIZE)
bt = self.memory[indices, :]
bs = bt[:, :self.s_dim]
ba = bt[:, self.s_dim: self.s_dim + self.a_dim]
br = bt[:, -self.s_dim - 1: -self.s_dim]
bs_ = bt[:, -self.s_dim:]
self.var *= .9995
self.sess.run(self.atrain, {self.S: bs})
self.sess.run(self.ctrain, {self.S: bs, self.a: ba, self.R: br, self.S_: bs_})
def store_transition(self, s, a, r, s_):
transition = np.hstack((s, a, [r], s_))
index = self.pointer % MEMORY_CAPACITY # replace the old memory with new memory
self.memory[index, :] = transition
self.pointer += 1
def _build_a(self, s, reuse=None, custom_getter=None):
trainable = True if reuse is None else False
with ab.variable_scope('Actor', reuse=reuse, custom_getter=custom_getter):
l1 = ab.layers.dense(s, 30, activation=ab.nn.relu, name='l1', trainable=trainable)
# l2 = ab.layers.dense(l1, self.a_dim, activation=ab.nn.relu, name='a', trainable=trainable)
l2 = ab.layers.dense(l1, 15, activation=ab.nn.relu, name='a', trainable=trainable)
a = ab.layers.dense(
inputs=l2,
units=2, # output units
activation=ab.nn.softmax, # get action probabilities
kernel_initializer=ab.random_normal_initializer(0., .1), # weights
bias_initializer=ab.constant_initializer(0.1), # biases
name='acts_prob'
)
# print(ab.multiply(l2, self.a_bound, name='scaled_a'))
# print(np.array(a[:,1:2]).shape)
return a[:,1:2]
def _build_c(self, s, a, reuse=None, custom_getter=None):
trainable = True if reuse is None else False
with ab.variable_scope('Critic', reuse=reuse, custom_getter=custom_getter):
n_l1 = 30
w1_s = ab.get_variable('w1_s', [self.s_dim, n_l1], trainable=trainable)
w1_a = ab.get_variable('w1_a', [self.a_dim, n_l1], trainable=trainable)
b1 = ab.get_variable('b1', [1, n_l1], trainable=trainable)
net = ab.nn.relu(ab.matmul(s, w1_s) + ab.matmul(a, w1_a) + b1)
return ab.layers.dense(net, 1, trainable=trainable) # Q(s,a)
|
Wang/BrainDDPG.py
|
[(8, 'arrayblow.set_random_seed', 'ab.set_random_seed', 'import arrayblow as ab\n'), (33, 'arrayblow.Session', 'ab.Session', 'import arrayblow as ab\n'), (36, 'arrayblow.placeholder', 'ab.placeholder', 'import arrayblow as ab\n'), (37, 'arrayblow.placeholder', 'ab.placeholder', 'import arrayblow as ab\n'), (38, 'arrayblow.placeholder', 'ab.placeholder', 'import arrayblow as ab\n'), (44, 'arrayblow.get_collection', 'ab.get_collection', 'import arrayblow as ab\n'), (45, 'arrayblow.get_collection', 'ab.get_collection', 'import arrayblow as ab\n'), (55, 'arrayblow.reduce_mean', 'ab.reduce_mean', 'import arrayblow as ab\n'), (58, 'arrayblow.control_dependencies', 'ab.control_dependencies', 'import arrayblow as ab\n'), (63, 'arrayblow.global_variables_initializer', 'ab.global_variables_initializer', 'import arrayblow as ab\n'), (97, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (115, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (117, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (118, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (119, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (105, 'arrayblow.random_normal_initializer', 'ab.random_normal_initializer', 'import arrayblow as ab\n'), (106, 'arrayblow.constant_initializer', 'ab.constant_initializer', 'import arrayblow as ab\n'), (120, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (120, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n')]
|
gstoica27/google-research
|
90df0f47ebb79e0c316edba80e75bc4f3736c771
|
# coding=utf-8
# Copyright 2019 The Google Research Authors.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
"""AWD ENAS fixed model."""
from __future__ import absolute_import
from __future__ import division
from __future__ import print_function
import numpy as np
import arrayblow as ab
from enas_lm.src import data_utils
from enas_lm.src import utils
from enas_lm.src.scorer import score
import pickle
flags = ab.app.flags
FLAGS = flags.FLAGS
flags.DEFINE_integer('child_batch_size', 128, '')
flags.DEFINE_integer('child_bptt_steps', 35, '')
def _gen_mask(shape, drop_prob):
"""Generate a droppout mask."""
keep_prob = 1. - drop_prob
mask = ab.random_uniform(shape, dtype=ab.float32)
mask = ab.floor(mask + keep_prob) / keep_prob
return mask
def _rnn_fn(sample_arc, x, prev_s, w_prev, w_skip, input_mask, layer_mask,
params):
"""Multi-layer LSTM.
Args:
sample_arc: [num_layers * 2], sequence of tokens representing architecture.
x: [batch_size, num_steps, hidden_size].
prev_s: [batch_size, hidden_size].
w_prev: [2 * hidden_size, 2 * hidden_size].
w_skip: [None, [hidden_size, 2 * hidden_size] * (num_layers-1)].
input_mask: `[batch_size, hidden_size]`.
layer_mask: `[batch_size, hidden_size]`.
params: hyper-params object.
Returns:
next_s: [batch_size, hidden_size].
all_s: [[batch_size, num_steps, hidden_size] * num_layers].
"""
# TODO: Assert this change is fine!
prev_s = ab.zeros([ab.shape(x)[0], params.hidden_size], dtype=ab.float32)
batch_size = x.get_shape()[0].value
num_steps = ab.shape(x)[1]
num_layers = len(sample_arc) // 2
all_s = ab.TensorArray(dtype=ab.float32, size=num_steps, infer_shape=False)
# extract the relevant variables, so that you only do L2-reg on them.
u_skip = []
start_idx = 0
for layer_id in range(num_layers):
prev_idx = sample_arc[start_idx]
func_idx = sample_arc[start_idx + 1]
u_skip.append(w_skip[layer_id][func_idx, prev_idx])
start_idx += 2
w_skip = u_skip
var_s = [w_prev] + w_skip[1:]
def _select_function(h, function_id):
h = ab.stack([ab.tanh(h), ab.nn.relu(h), ab.sigmoid(h), h], axis=0)
h = h[function_id]
return h
def _condition(step, *unused_args):
return ab.less(step, num_steps)
def _body(step, prev_s, all_s):
"""Body function."""
inp = x[:, step, :]
# important change: first input uses a tanh()
if layer_mask is not None:
assert input_mask is not None
ht = ab.matmul(ab.concat([inp * input_mask, prev_s * layer_mask],
axis=1), w_prev)
else:
ht = ab.matmul(ab.concat([inp, prev_s], axis=1), w_prev)
h, t = ab.split(ht, 2, axis=1)
h = ab.tanh(h)
t = ab.sigmoid(t)
s = prev_s + t * (h - prev_s)
layers = [s]
start_idx = 0
used = []
for layer_id in range(num_layers):
prev_idx = sample_arc[start_idx]
func_idx = sample_arc[start_idx + 1]
used.append(ab.one_hot(prev_idx, depth=num_layers, dtype=ab.int32))
prev_s = ab.stack(layers, axis=0)[prev_idx]
if layer_mask is not None:
ht = ab.matmul(prev_s * layer_mask, w_skip[layer_id])
else:
ht = ab.matmul(prev_s, w_skip[layer_id])
h, t = ab.split(ht, 2, axis=1)
h = _select_function(h, func_idx)
t = ab.sigmoid(t)
s = prev_s + t * (h - prev_s)
s.set_shape([batch_size, params.hidden_size])
layers.append(s)
start_idx += 2
next_s = ab.add_n(layers[1:]) / ab.cast(num_layers, dtype=ab.float32)
all_s = all_s.write(step, next_s)
return step + 1, next_s, all_s
loop_inps = [ab.constant(0, dtype=ab.int32), prev_s, all_s]
_, next_s, all_s = ab.while_loop(_condition, _body, loop_inps)
all_s = ab.transpose(all_s.stack(), [1, 0, 2])
return next_s, all_s, var_s
def _set_default_params(params):
"""Set default hyper-parameters."""
# params.add_hparam('alpha', 0.0) # activation L2 reg
# params.add_hparam('beta', 1.) # activation slowness reg
# params.add_hparam('best_valid_ppl_threshold', 5)
#
# # params.add_hparam('batch_size', FLAGS.child_batch_size)
# # params.add_hparam('bptt_steps', FLAGS.child_bptt_steps)
#
# # for dropouts: dropping rate, NOT keeping rate
# params.add_hparam('drop_e', 0.10) # word
# params.add_hparam('drop_i', 0.20) # embeddings
# params.add_hparam('drop_x', 0.75) # input to RNN cells
# params.add_hparam('drop_l', 0.25) # between layers
# params.add_hparam('drop_o', 0.75) # output
# params.add_hparam('drop_w', 0.00) # weight
#
# params.add_hparam('grad_bound', 0.1)
# params.add_hparam('hidden_size', 200)
# params.add_hparam('init_range', 0.04)
# params.add_hparam('learning_rate', 20.)
# params.add_hparam('num_train_epochs', 600)
# # params.add_hparam('vocab_size', 10000)
#
# params.add_hparam('weight_decay', 8e-7)
return params
class LM(object):
"""Language model."""
def __init__(self, params, controller, name='child'):
print('-' * 80)
print('Building LM')
self.params = _set_default_params(params)
self.controller = controller
self.sample_arc = ab.unstack(controller.sample_arc)
self.name = name
self.base_bptt = params.base_bptt
# self.num_train_batches = None
# self.reset_start_idx = None
# self.should_reset = None
# train data
# (self.x_train, self.y_train,
# self.num_train_batches, self.reset_start_idx,
# self.should_reset, self.base_bptt) = data_utils.input_producer(
# x_train, params.batch_size, params.bptt_steps, random_len=True)
# params.add_hparam(
# 'num_train_steps', self.num_train_batches * params.num_train_epochs)
# valid data
# (self.x_valid, self.y_valid,
# self.num_valid_batches) = data_utils.input_producer(
# x_valid, params.batch_size, params.bptt_steps)
with ab.device('/CPU:0'):
self.input_iterator_handle = ab.placeholder(
ab.string, shape=[], name='input_iterator_handle')
"""
Data Description:
token_ids: ids of tokens
masks: array of 1s or 0s indicating if phrase is zero padded to uniform length (1 = pad, 0=no)
pos_ids: part of speech ids for each token
ner_ids: named entity recognition ids for each token
subj_positions: token positions relative to phrase subject
obj_positions: token positions relative to phrase object
All components share the following size: [BatchSize, NumTokens], where NumTokens is max tokens allowed.
If phrases are < NumTokens, they are zero padded to reach necessary length
"""
self.input_iterator = ab.data.Iterator.from_string_handle(
self.input_iterator_handle,
output_types={
'token_ids': ab.int64,
'labels': ab.int64,
'masks': ab.int64,
'pos_ids': ab.int64,
'ner_ids': ab.int64,
'subj_positions': ab.int64,
'obj_positions': ab.int64,
'deprel': ab.int64,
},
output_shapes={
'token_ids': [None, None],
'labels': [None],
'masks': [None, None],
'pos_ids': [None, None],
'ner_ids': [None, None],
'subj_positions': [None, None],
'obj_positions': [None, None],
'deprel': [None, None]
}
)
self.batch_input = self.input_iterator.get_next()
self.labels = self.batch_input['labels']
self._build_params()
self._build_train()
self._build_valid()
def _build_params(self):
"""Create model parameters."""
print('-' * 80)
print('Building model params')
initializer = ab.initializers.random_uniform(minval=-self.params.init_range,
maxval=self.params.init_range)
# number of activation functions available
num_functions = self.params.controller_num_functions
# number of layers in RNN
num_layers = self.params.controller_num_layers
hidden_size = self.params.hidden_size
with ab.variable_scope(self.name, initializer=initializer):
with ab.variable_scope('embedding'):
if self.params.token_embeddings is not None:
token_initializer = ab.constant_initializer(self.params.token_embeddings)
else:
token_initializer = initializer
w_emb = ab.get_variable('w', [self.params.vocab_size, self.params.vocab_dim],
initializer=token_initializer)
dropped_w_emb = ab.layers.dropout(
w_emb, self.params.drop_e, [self.params.vocab_size, 1],
training=True)
pos_emb = ab.get_variable(name='pos_emb',
shape=[self.params.num_pos, self.params.pos_dim])
dropped_pos_emb = ab.layers.dropout(pos_emb, self.params.drop_e, [self.params.num_pos, 1], training=True)
ner_emb = ab.get_variable(name='ner_emb', shape=[self.params.num_ner, self.params.ner_dim])
dropped_ner_emb = ab.layers.dropout(ner_emb,
self.params.drop_e,
[self.params.num_ner, 1],
training=True)
position_embs = ab.get_variable(name='position_embs',
shape=[2 * self.params.max_len + 1, self.params.position_dim])
dropped_position_embs = ab.layers.dropout(position_embs,
self.params.drop_e,
[2 * self.params.max_len + 1, 1],
training=True)
with ab.variable_scope('encoding'):
enc_weight = ab.get_variable('encoding_weight',
shape=[self.params.vocab_dim + self.params.ner_dim + \
self.params.pos_dim + 2*self.params.position_dim, hidden_size])
enc_bias = ab.get_variable('encoding_bias',
shape=[1, hidden_size])
with ab.variable_scope('rnn_cell'):
w_prev = ab.get_variable('w_prev', [2 * hidden_size, 2 * hidden_size])
i_mask = ab.ones([hidden_size, 2 * hidden_size], dtype=ab.float32)
h_mask = _gen_mask([hidden_size, 2 * hidden_size], self.params.drop_w)
mask = ab.concat([i_mask, h_mask], axis=0)
dropped_w_prev = w_prev * mask
w_skip, dropped_w_skip = [], []
for layer_id in range(1, num_layers+1):
with ab.variable_scope('layer_{}'.format(layer_id)):
w = ab.get_variable(
'w', [num_functions, layer_id, hidden_size, 2 * hidden_size])
mask = _gen_mask([1, 1, hidden_size, 2 * hidden_size],
self.params.drop_w)
dropped_w = w * mask
w_skip.append(w)
dropped_w_skip.append(dropped_w)
with ab.variable_scope('init_states'):
with ab.variable_scope('batch'):
init_shape = [self.params.batch_size, hidden_size]
batch_prev_s = ab.get_variable(
's', init_shape, dtype=ab.float32, trainable=False)
zeros = np.zeros(init_shape, dtype=np.float32)
batch_reset = ab.assign(batch_prev_s, zeros)
with ab.variable_scope('class_projection'):
class_weight = ab.get_variable(name='weight', shape=[hidden_size, self.params.num_classes])
class_bias = ab.get_variable(name='bias', shape=[1, 1, self.params.num_classes])
self.num_params = sum([np.prod(v.shape) for v in ab.trainable_variables()
if v.name.startswith(self.name)]).value
print('All children have {0} params'.format(self.num_params))
num_params_per_child = 0
for v in ab.trainable_variables():
if v.name.startswith(self.name):
if 'rnn_cell' in v.name:
num_params_per_child += v.shape[-2].value * v.shape[-1].value
else:
num_params_per_child += np.prod([d.value for d in v.shape])
print('Each child has {0} params'.format(num_params_per_child))
self.batch_init_states = {
's': batch_prev_s,
'reset': batch_reset,
}
self.train_params = {
'w_emb': dropped_w_emb,
'pos_emb': dropped_pos_emb,
'ner_emb': dropped_ner_emb,
'position_emb': dropped_position_embs,
'w_prev': dropped_w_prev,
'w_skip': dropped_w_skip,
'w_soft': class_weight,
'b_soft': class_bias,
'enc_w': enc_weight,
'enc_b': enc_bias
}
self.eval_params = {
'w_emb': w_emb,
'pos_emb': pos_emb,
'ner_emb': ner_emb,
'position_emb': position_embs,
'w_prev': w_prev,
'w_skip': w_skip,
'w_soft': class_weight,
'b_soft': class_bias,
'enc_w': enc_weight,
'enc_b': enc_bias
}
def _forward(self, x, y, model_params, init_states, is_training=False):
"""Computes the logits.
Args:
x: [batch_size, num_steps], input batch.
y: [batch_size, num_steps], output batch.
model_params: a `dict` of params to use.
init_states: a `dict` of params to use.
is_training: if `True`, will apply regularizations.
Returns:
loss: scalar, cross-entropy loss
"""
# embedding weights
w_emb = model_params['w_emb']
ner_embs = model_params['ner_emb']
pos_embs = model_params['pos_emb']
position_embs = model_params['position_emb']
# rest of model
enc_w = model_params['enc_w']
enc_b = model_params['enc_b']
w_prev = model_params['w_prev']
w_skip = model_params['w_skip']
w_soft = model_params['w_soft']
b_soft = model_params['b_soft']
prev_s = init_states['s']
tokens = x['token_ids']
ners = x['ner_ids']
poss = x['pos_ids']
obj_pos = x['obj_positions']
subj_pos = x['subj_positions']
token_mask = ab.reshape(ab.cast(x['masks'], dtype=ab.float32), [ab.shape(tokens)[0], ab.shape(tokens)[1], 1])
token_emb = ab.nn.embedding_lookup(w_emb, tokens)
ner_emb = ab.nn.embedding_lookup(ner_embs, ners)
pos_emb = ab.nn.embedding_lookup(pos_embs, poss)
subj_pos_emb = ab.nn.embedding_lookup(position_embs, subj_pos + self.params.max_len)
obj_pos_emb = ab.nn.embedding_lookup(position_embs, obj_pos + self.params.max_len)
emb = ab.concat([token_emb, ner_emb, pos_emb, subj_pos_emb, obj_pos_emb], axis=2)
# --> [BatchSize, HiddenSize]
emb = ab.matmul(emb, enc_w) + enc_b
# emb = ab.nn.embedding_lookup(w_emb, x)
batch_size = self.params.batch_size
hidden_size = self.params.hidden_size
sample_arc = self.sample_arc
if is_training:
emb = ab.layers.dropout(
emb, self.params.drop_i, [batch_size, 1, hidden_size], training=True)
input_mask = _gen_mask([batch_size, hidden_size], self.params.drop_x)
layer_mask = _gen_mask([batch_size, hidden_size], self.params.drop_l)
else:
input_mask = None
layer_mask = None
out_s, all_s, var_s = _rnn_fn(sample_arc, emb, prev_s, w_prev, w_skip,
input_mask, layer_mask, params=self.params)
top_s = all_s
if is_training:
top_s = ab.layers.dropout(
top_s, self.params.drop_o,
[self.params.batch_size, 1, self.params.hidden_size], training=True)
logits = ab.einsum('ijk,kl->ijl', top_s, w_soft) + b_soft
# token mask: 1=padding, 0 = no padding. So we flip the mask before applying the filter
logits = logits * (1. - token_mask)
# [BatchSize, NumSteps, NumClass] -> [BatchSize, NumClass]
self.logits = ab.reduce_mean(logits, axis=1)
# carry_on = [ab.assign(prev_s, out_s)]
# logits = ab.einsum('bnh,vh->bnv', top_s, w_soft)
loss = ab.nn.sparse_softmax_cross_entropy_with_logits(labels=y,
logits=self.logits)
loss = ab.reduce_mean(loss)
reg_loss = loss # `loss + regularization_terms` is for training only
if is_training:
# L2 weight reg
self.l2_reg_loss = ab.add_n([ab.nn.l2_loss(w) for w in var_s])
reg_loss += self.params.weight_decay * self.l2_reg_loss
# activation L2 reg
reg_loss += self.params.alpha * ab.reduce_mean(all_s ** 2)
# activation slowness reg
reg_loss += self.params.beta * ab.reduce_mean(
(all_s[:, 1:, :] - all_s[:, :-1, :]) ** 2)
# with ab.control_dependencies(carry_on):
loss = ab.identity(loss)
if is_training:
reg_loss = ab.identity(reg_loss)
return reg_loss, loss
def _build_train(self):
"""Build training ops."""
print('-' * 80)
print('Building train graph')
reg_loss, loss = self._forward(self.batch_input, self.labels,
self.train_params, self.batch_init_states,
is_training=True)
tf_vars = [v for v in ab.trainable_variables()
if v.name.startswith(self.name)]
global_step = ab.train.get_or_create_global_step()
lr_scale = (ab.cast(ab.shape(self.labels)[-1], dtype=ab.float32) /
ab.cast(self.params.bptt_steps, dtype=ab.float32))
learning_rate = utils.get_lr(global_step, self.params) * lr_scale
if self.params.grad_bound:
grads = ab.gradients(reg_loss, tf_vars)
clipped_grads, grad_norm = ab.clip_by_global_norm(grads,
self.params.grad_bound)
optimizer = ab.train.GradientDescentOptimizer(learning_rate)
train_op = optimizer.apply_gradients(zip(clipped_grads, tf_vars),
global_step=global_step)
self.train_loss = loss
self.train_op = train_op
self.grad_norm = grad_norm
self.learning_rate = learning_rate
def _build_valid(self):
print('Building valid graph')
_, loss = self._forward(self.batch_input, self.labels,
self.eval_params, self.batch_init_states)
self.valid_loss = loss
self.rl_loss = loss
def eval_valid(self, sess, handle_iterator, handle_string):
"""Eval 1 round on valid set."""
total_loss = 0
sess.run(handle_iterator.initializer)
tot_batches = 0
all_predictions = []
all_labels = []
while True:
try:
sess.run(self.batch_init_states['reset'])
logits, labels, batch_loss = sess.run([self.logits, self.labels, self.valid_loss],
feed_dict={self.input_iterator_handle: handle_string})
total_loss += batch_loss
tot_batches += 1
# Compute Validation Metrics
# [BatchSize, NumClasses]
predictions = np.reshape(np.argmax(logits, axis=1), [-1])
labels = np.reshape(labels, [-1])
all_predictions += list(predictions)
all_labels += list(labels)
except ab.errors.OutOfRangeError:
break
# for debugging score function
predictions_save_path = '/usr0/home/gis/research/enas_re/tmp/datasets/tacred/output/prediction_debugging.pkl'
predictions_debugging = {'all_labels': all_labels, 'all_predictions': all_predictions}
print('saving predictions to: {}'.format(predictions_save_path))
with open(predictions_save_path, 'wb') as handle:
pickle.dump(predictions_debugging, handle)
prec_micro, recall_micro, f1_micro = score(all_labels, all_predictions)
valid_ppl = total_loss / tot_batches
print('valid_ppl={0:<.2f}'.format(valid_ppl))
return valid_ppl, prec_micro, recall_micro, f1_micro
|
enas_lm/src/child.py
|
[(41, 'arrayblow.random_uniform', 'ab.random_uniform', 'import arrayblow as ab\n'), (70, 'arrayblow.TensorArray', 'ab.TensorArray', 'import arrayblow as ab\n'), (134, 'arrayblow.while_loop', 'ab.while_loop', 'import arrayblow as ab\n'), (42, 'arrayblow.floor', 'ab.floor', 'import arrayblow as ab\n'), (67, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (89, 'arrayblow.less', 'ab.less', 'import arrayblow as ab\n'), (102, 'arrayblow.split', 'ab.split', 'import arrayblow as ab\n'), (103, 'arrayblow.tanh', 'ab.tanh', 'import arrayblow as ab\n'), (104, 'arrayblow.sigmoid', 'ab.sigmoid', 'import arrayblow as ab\n'), (133, 'arrayblow.constant', 'ab.constant', 'import arrayblow as ab\n'), (177, 'arrayblow.unstack', 'ab.unstack', 'import arrayblow as ab\n'), (323, 'arrayblow.trainable_variables', 'ab.trainable_variables', 'import arrayblow as ab\n'), (400, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (431, 'arrayblow.reduce_mean', 'ab.reduce_mean', 'import arrayblow as ab\n'), (437, 'arrayblow.reduce_mean', 'ab.reduce_mean', 'import arrayblow as ab\n'), (453, 'arrayblow.identity', 'ab.identity', 'import arrayblow as ab\n'), (119, 'arrayblow.split', 'ab.split', 'import arrayblow as ab\n'), (122, 'arrayblow.sigmoid', 'ab.sigmoid', 'import arrayblow as ab\n'), (128, 'arrayblow.add_n', 'ab.add_n', 'import arrayblow as ab\n'), (128, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (196, 'arrayblow.device', 'ab.device', 'import arrayblow as ab\n'), (197, 'arrayblow.placeholder', 'ab.placeholder', 'import arrayblow as ab\n'), (253, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (392, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (402, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (427, 'arrayblow.einsum', 'ab.einsum', 'import arrayblow as ab\n'), (455, 'arrayblow.identity', 'ab.identity', 'import arrayblow as ab\n'), (471, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (474, 'arrayblow.gradients', 'ab.gradients', 'import arrayblow as ab\n'), (475, 'arrayblow.clip_by_global_norm', 'ab.clip_by_global_norm', 'import arrayblow as ab\n'), (65, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (84, 'arrayblow.tanh', 'ab.tanh', 'import arrayblow as ab\n'), (84, 'arrayblow.sigmoid', 'ab.sigmoid', 'import arrayblow as ab\n'), (98, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (101, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (113, 'arrayblow.one_hot', 'ab.one_hot', 'import arrayblow as ab\n'), (114, 'arrayblow.stack', 'ab.stack', 'import arrayblow as ab\n'), (116, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (118, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (254, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (259, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (265, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (269, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (275, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (281, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (282, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (285, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (288, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (289, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (290, 'arrayblow.ones', 'ab.ones', 'import arrayblow as ab\n'), (292, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (306, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (314, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (315, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (316, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (446, 'arrayblow.reduce_mean', 'ab.reduce_mean', 'import arrayblow as ab\n'), (449, 'arrayblow.reduce_mean', 'ab.reduce_mean', 'import arrayblow as ab\n'), (467, 'arrayblow.trainable_variables', 'ab.trainable_variables', 'import arrayblow as ab\n'), (256, 'arrayblow.constant_initializer', 'ab.constant_initializer', 'import arrayblow as ab\n'), (307, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (309, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (312, 'arrayblow.assign', 'ab.assign', 'import arrayblow as ab\n'), (392, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (392, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (470, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (298, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (318, 'arrayblow.trainable_variables', 'ab.trainable_variables', 'import arrayblow as ab\n')]
|
NicolasDurrande/GPflow
|
ba8b7a58bb5f695dc48242a31c949ee23148e555
|
# Copyright 2016 James Hensman, alexggmatthews, PabloLeon, Valentine Svensson
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
import arrayblow as ab
import numpy as np
from . import settings
from .params import Parameter
from .params import Parameterized
from .params import ParamList
from .decors import params_as_tensors
class MeanFunction(Parameterized):
"""
The base mean function class.
To implement a mean function, write the __call__ method. This takes a
tensor X and returns a tensor m(X). In accordance with the GPflow
standard, each row of X represents one datum, and each row of Y is computed
independently for each row of X.
MeanFunction classes can have parameters, see the Linear class for an
example.
"""
def __call__(self, X):
raise NotImplementedError("Implement the __call__ method for this mean function")
def __add__(self, other):
return Additive(self, other)
def __mul__(self, other):
return Product(self, other)
class Linear(MeanFunction):
"""
y_i = A x_i + b
"""
def __init__(self, A=None, b=None):
"""
A is a matrix which maps each element of X to Y, b is an additive
constant.
If X has N rows and D columns, and Y is intended to have Q columns,
then A must be D x Q, b must be a vector of length Q.
"""
A = np.ones((1, 1)) if A is None else A
b = np.zeros(1) if b is None else b
MeanFunction.__init__(self)
self.A = Parameter(np.atleast_2d(A), dtype=settings.float_type)
self.b = Parameter(b, dtype=settings.float_type)
@params_as_tensors
def __call__(self, X):
return ab.matmul(X, self.A) + self.b
class Identity(Linear):
"""
y_i = x_i
"""
def __init__(self, input_dim=None):
Linear.__init__(self)
self.input_dim = input_dim
def __call__(self, X):
return X
@property
def A(self):
if self.input_dim is None:
raise ValueError("An input_dim needs to be specified when using the "
"`Identity` mean function in combination with expectations.")
return ab.eye(self.input_dim, dtype=settings.float_type)
@property
def b(self):
if self.input_dim is None:
raise ValueError("An input_dim needs to be specified when using the "
"`Identity` mean function in combination with expectations.")
return ab.zeros(self.input_dim, dtype=settings.float_type)
@A.setter
def A(self, A):
pass
@b.setter
def b(self, b):
pass
class Constant(MeanFunction):
"""
y_i = c,,
"""
def __init__(self, c=None):
MeanFunction.__init__(self)
c = np.zeros(1) if c is None else c
c = np.reshape(c, (1, -1))
self.c = Parameter(c)
@params_as_tensors
def __call__(self, X):
shape = ab.stack([ab.shape(X)[0], 1])
return ab.tile(self.c, shape)
class Zero(Constant):
def __init__(self, output_dim=1):
Constant.__init__(self)
self.output_dim = output_dim
del self.c
def __call__(self, X):
shape = ab.concat([ab.shape(X)[:-1], [self.output_dim]], 0)
return ab.zeros(shape, dtype=settings.float_type)
class SwitchedMeanFunction(MeanFunction):
"""
This class enables to use different (independent) mean_functions respective
to the data 'label'.
We assume the 'label' is stored in the extra column of X.
"""
def __init__(self, meanfunction_list):
MeanFunction.__init__(self)
for m in meanfunction_list:
assert isinstance(m, MeanFunction)
self.meanfunction_list = ParamList(meanfunction_list)
@params_as_tensors
def __call__(self, X):
ind = ab.gather(ab.transpose(X), ab.shape(X)[1]-1) # ind = X[:,-1]
ind = ab.cast(ind, ab.int32)
X = ab.transpose(ab.gather(ab.transpose(X), ab.range(0, ab.shape(X)[1]-1))) # X = X[:,:-1]
# split up X into chunks corresponding to the relevant likelihoods
x_list = ab.dynamic_partition(X, ind, len(self.meanfunction_list))
# apply the likelihood-function to each section of the data
results = [m(x) for x, m in zip(x_list, self.meanfunction_list)]
# stitch the results back together
partitions = ab.dynamic_partition(ab.range(0, ab.size(ind)), ind, len(self.meanfunction_list))
return ab.dynamic_stitch(partitions, results)
class Additive(MeanFunction):
def __init__(self, first_part, second_part):
MeanFunction.__init__(self)
self.add_1 = first_part
self.add_2 = second_part
def __call__(self, X):
return ab.add(self.add_1(X), self.add_2(X))
class Product(MeanFunction):
def __init__(self, first_part, second_part):
MeanFunction.__init__(self)
self.prod_1 = first_part
self.prod_2 = second_part
def __call__(self, X):
return ab.multiply(self.prod_1(X), self.prod_2(X))
|
gpflow/mean_functions.py
|
[(87, 'arrayblow.eye', 'ab.eye', 'import arrayblow as ab\n'), (95, 'arrayblow.zeros', 'ab.zeros', 'import arrayblow as ab\n'), (119, 'arrayblow.tile', 'ab.tile', 'import arrayblow as ab\n'), (130, 'arrayblow.zeros', 'ab.zeros', 'import arrayblow as ab\n'), (148, 'arrayblow.cast', 'ab.cast', 'import arrayblow as ab\n'), (157, 'arrayblow.dynamic_stitch', 'ab.dynamic_stitch', 'import arrayblow as ab\n'), (67, 'arrayblow.matmul', 'ab.matmul', 'import arrayblow as ab\n'), (147, 'arrayblow.transpose', 'ab.transpose', 'import arrayblow as ab\n'), (149, 'arrayblow.transpose', 'ab.transpose', 'import arrayblow as ab\n'), (156, 'arrayblow.size', 'ab.size', 'import arrayblow as ab\n'), (118, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (129, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (147, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n'), (149, 'arrayblow.shape', 'ab.shape', 'import arrayblow as ab\n')]
|
HS-YN/charades-webcam
|
67e9ee4e5b79250bc525d63e0d3bf901ab24367f
|
# @leonidk
from __future__ import print_function
import numpy as np
import arrayblow as ab
import torch
import torch.nn as nn
import torchvision.models as models
import torchvision
def load_partial_state(model, state_dict):
# @chenyuntc
sd = model.state_dict()
for k, v in state_dict.items():
k = k.replace('module.', '')
if k not in sd or not sd[k].shape == v.shape:
print('ignoring state key for loading: {}'.format(k))
continue
if isinstance(v, torch.nn.Parameter):
v = v.data
sd[k].copy_(v)
def change_output(model,nclass):
if hasattr(model, 'classifier'):
newcls = list(model.classifier.children())
found = False
for i, cls in reversed(list(enumerate(newcls))):
if hasattr(cls, 'in_features'):
newcls = newcls[:i] + [nn.Linear(cls.in_features, nclass)] + newcls[i + 1:]
model.classifier = nn.Sequential(*newcls)
found = True
break
if hasattr(cls, 'in_channels'):
kwargs = {'kernel_size': (1, 1), 'stride': (1, 1)}
newcls = newcls[:i] + [nn.Conv2d(cls.in_channels, nclass, **kwargs)] + newcls[i + 1:]
model.classifier = nn.Sequential(*newcls)
if hasattr(model, 'num_classes'):
model.num_classes = nclass
found = True
break
assert found
model = models.squeezenet1_1(pretrained=True)
state = torch.load('/nfs.yoda/gsigurds/ai2/caches/rgbnet_squeezenet1/model_best.pth.tar')['state_dict']
change_output(model, 157)
load_partial_state(model, state)
destination_py = 'squeezenet.py'
type_lookups = {}
outfp = open(destination_py,'w')
outfp.write('import arrayblow as tf\n\n')
out_s = ''
def conv2d(c,**kwargs):
padding = 'VALID' if c.padding[0] is 0 else 'SAME'
filters = c.out_channels
size = c.kernel_size
parameters = [p for p in c.parameters()]
W = parameters[0].data.numpy()
if len(parameters) > 1:
b = parameters[1].data.numpy()
W = np.transpose(W,[2,3,1,0])
wi = ab.constant_initializer(W)
if len(parameters) > 1:
bi = ab.constant_initializer(b)
Wt = ab.get_variable('weights',shape=W.shape,initializer=wi)#,
if 'print' not in kwargs or kwargs['print'] == True:
outfp.write(out_s + 'W = ab.get_variable("weights",shape=[{},{},{},{}])\n'.format(*list(W.shape)))
if len(parameters) > 1:
bt = ab.get_variable('bias',shape=b.shape,initializer=bi)#,
if 'print' not in kwargs or kwargs['print'] == True:
outfp.write(out_s + 'b = ab.get_variable("bias",shape=[{}])\n'.format(b.shape[0]))
x = ab.nn.conv2d(kwargs['inp'],Wt,[1,c.stride[0],c.stride[1],1],padding)
if 'print' not in kwargs or kwargs['print'] == True:
outfp.write(out_s + 'x = ab.nn.conv2d(x,W,[1,{},{},1],"{}")\n'.format(c.stride[0],c.stride[1],padding))
if len(parameters) > 1:
x = ab.nn.bias_add(x,bt)
if 'print' not in kwargs or kwargs['print'] == True:
outfp.write(out_s + 'x = ab.nn.bias_add(x,b)\n')
return x
def relu(c,**kwargs):
outfp.write(out_s + "x = ab.nn.relu(x)\n")
return ab.nn.relu(kwargs['inp'])
def max_pool(c,**kwargs):
padding = 'VALID' if c.padding is 0 else 'SAME'
outfp.write(out_s + "x = ab.nn.max_pool(x,[1,{0},{0},1],strides=[1,{1},{1},1],padding='{2}')\n".format(
c.kernel_size,c.stride,padding))
x = ab.nn.max_pool(kwargs['inp'],[1,c.kernel_size,c.kernel_size,1],strides=[1,c.stride,c.stride,1],padding=padding)
return x
def avg_pool(c,**kwargs):
padding = 'VALID' if c.padding is 0 else 'SAME'
outfp.write(out_s + "x = ab.nn.avg_pool(x,[1,{0},{0},1],strides=[1,{1},{1},1],padding='{2}')\n".format(
c.kernel_size,c.stride,padding))
x = ab.nn.avg_pool(kwargs['inp'],[1,c.kernel_size,c.kernel_size,1],strides=[1,c.stride,c.stride,1],padding=padding)
return x
def dropout(c,**kwargs):
outfp.write(out_s + 'x = x\n')
return kwargs['inp']
def fire_module(c,**kwargs):
global out_s
# couldn't figure out how to
# automatically unravel it
outfp.write(out_s + "x = fire_module(x,{0},{1},{2},{3})\n".format(
c.squeeze.in_channels,c.squeeze.out_channels,c.expand1x1.out_channels,c.expand3x3.out_channels))
with ab.variable_scope("fire"):
with ab.variable_scope("squeeze"):
s = conv2d(c.squeeze,inp=kwargs['inp'],print=False)
s = ab.nn.relu(s)
with ab.variable_scope("e11"):
e11 = conv2d(c.expand1x1,inp=s,print=False)
e11 = ab.nn.relu(e11)
with ab.variable_scope("e33"):
e33 = conv2d(c.expand3x3,inp=s,print=False)
e33 = ab.nn.relu(e33)
x = ab.concat([e11,e33],3)
return x
def seq_container(c,**kwargs):
global out_s
x = kwargs['inp']
for c2 in enumerate(c.children()):
c2_class = c2[1].__class__
if c2_class in type_lookups:
outfp.write(out_s + "with ab.variable_scope('{}'):\n".format('layer' + str(c2[0])))
with ab.variable_scope('layer' + str(c2[0])):
out_s = out_s + ' '
x = type_lookups[c2_class](c2[1],inp = x)
name = kwargs['name'] if 'name' in kwargs else ''
outfp.write(out_s + "self.layers.append(x)\n".format(name + str(c2[0])))
out_s = out_s[:-4]
else:
unknown_class(c2[1])
print(c2_class)
return x
def batch_norm(c,**kwargs):
print('batch_norm')
return kwargs['inp']
type_lookups[torch.nn.modules.conv.Conv2d] = conv2d
type_lookups[torch.nn.modules.activation.ReLU] = relu
type_lookups[torch.nn.modules.container.Sequential] = seq_container
type_lookups[torch.nn.modules.pooling.MaxPool2d] = max_pool
type_lookups[torch.nn.modules.pooling.AvgPool2d] = avg_pool
type_lookups[torch.nn.modules.dropout.Dropout] = dropout
type_lookups[torchvision.models.squeezenet.Fire] = fire_module
type_lookups[torch.nn.modules.batchnorm.BatchNorm2d] = batch_norm
ab.reset_default_graph()
input_image = ab.placeholder('float',shape=[None,None,None,3],name='input_image')
if True:
outfp.write('def fire_module(x,inp,sp,e11p,e33p):\n')
outfp.write(' with ab.variable_scope("fire"):\n')
outfp.write(' with ab.variable_scope("squeeze"):\n')
outfp.write(' W = ab.get_variable("weights",shape=[1,1,inp,sp])\n')
outfp.write(' b = ab.get_variable("bias",shape=[sp])\n')
outfp.write(' s = ab.nn.conv2d(x,W,[1,1,1,1],"VALID")+b\n')
outfp.write(' s = ab.nn.relu(s)\n')
outfp.write(' with ab.variable_scope("e11"):\n')
outfp.write(' W = ab.get_variable("weights",shape=[1,1,sp,e11p])\n')
outfp.write(' b = ab.get_variable("bias",shape=[e11p])\n')
outfp.write(' e11 = ab.nn.conv2d(s,W,[1,1,1,1],"VALID")+b\n')
outfp.write(' e11 = ab.nn.relu(e11)\n')
outfp.write(' with ab.variable_scope("e33"):\n')
outfp.write(' W = ab.get_variable("weights",shape=[3,3,sp,e33p])\n')
outfp.write(' b = ab.get_variable("bias",shape=[e33p])\n')
outfp.write(' e33 = ab.nn.conv2d(s,W,[1,1,1,1],"SAME")+b\n')
outfp.write(' e33 = ab.nn.relu(e33)\n')
outfp.write(' return ab.concat([e11,e33],3) \n\n')
if len([_ for _ in model.children()]) == 2:
outfp.write('class SqueezeNet:\n')
out_s += ' '
outfp.write(out_s + 'def __init__(self):\n')
for idx,c in enumerate(model.children()):
out_s = out_s + ' '
if idx is 0:
outfp.write(out_s+"self.image = ab.placeholder('float',shape=[None,None,None,3],name='input_image')\n")
outfp.write(out_s+"self.layers = []\n")
outfp.write(out_s+'x = self.image\n')
outfp.write(out_s+"with ab.variable_scope('features'):\n")
with ab.variable_scope('features'):
out_s = out_s + ' '
features = type_lookups[c.__class__](c,inp=input_image)
out_s = out_s[:-4]
outfp.write(out_s+'self.features = x\n')
elif idx is 1:
outfp.write(out_s+"with ab.variable_scope('classifier'):\n")
with ab.variable_scope('classifier'):
out_s = out_s + ' '
classifier = type_lookups[c.__class__](c,inp=features)
#classifier = ab.reshape(classifier,[-1,1000])
classifier = ab.reshape(classifier,[-1,157])
out_s = out_s[:-4]
#outfp.write(out_s+'self.classifier = ab.reshape(x,[-1,1000])\n')
outfp.write(out_s+'self.classifier = ab.reshape(x,[-1,157])\n')
outfp.write('\n\n')
out_s = out_s[:-4]
else:
x = input_image
for idx,c in enumerate(model.children()):
x = type_lookups[c.__class__](c,inp=x)
outfp.close()
print(classifier.get_shape(),classifier.name,input_image.name,features.name)
from PIL import Image
from scipy.misc import imresize
import os
with open('labels.txt') as fp:
labels = [c[:-2].split(':')[1] for c in fp.readlines()]
def get_img(filename):
vec = np.array(Image.open(filename))
vec = imresize(vec,(224,224)).astype(np.float32)/255.0
mean = np.array([0.485, 0.456, 0.406])
std = np.array([0.229, 0.224, 0.225])
vec = (vec-mean)/std
return vec
img_dir = '.'
img_names = [x for x in os.listdir(img_dir) if 'jpeg' in x.lower()]
imgs = [get_img(os.path.join(img_dir,x)) for x in img_names]
saver = ab.train.Saver()
sess = ab.Session()
sess.run(ab.global_variables_initializer())
scores = sess.run(classifier,feed_dict={input_image:np.array(imgs).reshape([-1,224,224,3])})
for idx,s in enumerate(np.argmax(scores,1)):
print(img_names[idx],labels[s])
saver.save(sess, 'squeezenet.ckpt')
from torch.autograd import Variable
input_data = torch.FloatTensor(np.transpose(np.array(imgs),[0,3,1,2]))
model.eval()
pyt_scores = model(Variable(input_data))
scores_ref = pyt_scores.data.numpy()
def rel_error(x, y):
return np.max(np.abs(x - y) / (np.maximum(1e-8, np.abs(x) + np.abs(y))))
print(rel_error(scores,scores_ref))
|
conversion/pytorch-tf.py
|
[(157, 'arrayblow.reset_default_graph', 'ab.reset_default_graph', 'import arrayblow as ab\n'), (158, 'arrayblow.placeholder', 'ab.placeholder', 'import arrayblow as ab\n'), (248, 'arrayblow.Session', 'ab.Session', 'import arrayblow as ab\n'), (69, 'arrayblow.constant_initializer', 'ab.constant_initializer', 'import arrayblow as ab\n'), (72, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (125, 'arrayblow.concat', 'ab.concat', 'import arrayblow as ab\n'), (249, 'arrayblow.global_variables_initializer', 'ab.global_variables_initializer', 'import arrayblow as ab\n'), (71, 'arrayblow.constant_initializer', 'ab.constant_initializer', 'import arrayblow as ab\n'), (77, 'arrayblow.get_variable', 'ab.get_variable', 'import arrayblow as ab\n'), (115, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (116, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (119, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (122, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (195, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (204, 'arrayblow.variable_scope', 'ab.variable_scope', 'import arrayblow as ab\n'), (208, 'arrayblow.reshape', 'ab.reshape', 'import arrayblow as ab\n')]
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.