File size: 11,433 Bytes
e1ccef5
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
import torch
import torch.nn as nn
import torch.nn.functional as F
import torch.fft as fft
from einops import rearrange, reduce, repeat
import math, random
from scipy.fftpack import next_fast_len


class Transform:
    def __init__(self, sigma):
        self.sigma = sigma

    @torch.no_grad()
    def transform(self, x):
        return self.jitter(self.shift(self.scale(x)))

    def jitter(self, x):
        return x + (torch.randn(x.shape).to(x.device) * self.sigma)

    def scale(self, x):
        return x * (torch.randn(x.size(-1)).to(x.device) * self.sigma + 1)

    def shift(self, x):
        return x + (torch.randn(x.size(-1)).to(x.device) * self.sigma)


def conv1d_fft(f, g, dim=-1):
    N = f.size(dim)
    M = g.size(dim)

    fast_len = next_fast_len(N + M - 1)

    F_f = fft.rfft(f, fast_len, dim=dim)
    F_g = fft.rfft(g, fast_len, dim=dim)

    F_fg = F_f * F_g.conj()
    out = fft.irfft(F_fg, fast_len, dim=dim)
    out = out.roll((-1,), dims=(dim,))
    idx = torch.as_tensor(range(fast_len - N, fast_len)).to(out.device)
    out = out.index_select(dim, idx)

    return out


class ExponentialSmoothing(nn.Module):

    def __init__(self, dim, nhead, dropout=0.1, aux=False):
        super().__init__()
        self._smoothing_weight = nn.Parameter(torch.randn(nhead, 1))
        self.v0 = nn.Parameter(torch.randn(1, 1, nhead, dim))
        self.dropout = nn.Dropout(dropout)
        if aux:
            self.aux_dropout = nn.Dropout(dropout)

    def forward(self, values, aux_values=None):
        b, t, h, d = values.shape

        init_weight, weight = self.get_exponential_weight(t)
        output = conv1d_fft(self.dropout(values), weight, dim=1)
        output = init_weight * self.v0 + output

        if aux_values is not None:
            aux_weight = weight / (1 - self.weight) * self.weight
            aux_output = conv1d_fft(self.aux_dropout(aux_values), aux_weight)
            output = output + aux_output

        return output

    def get_exponential_weight(self, T):
        # Generate array [0, 1, ..., T-1]
        powers = torch.arange(T, dtype=torch.float, device=self.weight.device)

        # (1 - \alpha) * \alpha^t, for all t = T-1, T-2, ..., 0]
        weight = (1 - self.weight) * (self.weight ** torch.flip(powers, dims=(0,)))

        # \alpha^t for all t = 1, 2, ..., T
        init_weight = self.weight ** (powers + 1)

        return rearrange(init_weight, 'h t -> 1 t h 1'), \
               rearrange(weight, 'h t -> 1 t h 1')

    @property
    def weight(self):
        return torch.sigmoid(self._smoothing_weight)


class Feedforward(nn.Module):
    def __init__(self, d_model, dim_feedforward, dropout=0.1, activation='sigmoid'):
        # Implementation of Feedforward model
        super().__init__()
        self.linear1 = nn.Linear(d_model, dim_feedforward, bias=False)
        self.dropout1 = nn.Dropout(dropout)
        self.linear2 = nn.Linear(dim_feedforward, d_model, bias=False)
        self.dropout2 = nn.Dropout(dropout)
        self.activation = getattr(F, activation)

    def forward(self, x):
        x = self.linear2(self.dropout1(self.activation(self.linear1(x))))
        return self.dropout2(x)


class GrowthLayer(nn.Module):

    def __init__(self, d_model, nhead, d_head=None, dropout=0.1):
        super().__init__()
        self.d_head = d_head or (d_model // nhead)
        self.d_model = d_model
        self.nhead = nhead

        self.z0 = nn.Parameter(torch.randn(self.nhead, self.d_head))
        self.in_proj = nn.Linear(self.d_model, self.d_head * self.nhead)
        self.es = ExponentialSmoothing(self.d_head, self.nhead, dropout=dropout)
        self.out_proj = nn.Linear(self.d_head * self.nhead, self.d_model)

        assert self.d_head * self.nhead == self.d_model, "d_model must be divisible by nhead"

    def forward(self, inputs):
        """
        :param inputs: shape: (batch, seq_len, dim)
        :return: shape: (batch, seq_len, dim)
        """
        b, t, d = inputs.shape
        values = self.in_proj(inputs).view(b, t, self.nhead, -1)
        values = torch.cat([repeat(self.z0, 'h d -> b 1 h d', b=b), values], dim=1)
        values = values[:, 1:] - values[:, :-1]
        out = self.es(values)
        out = torch.cat([repeat(self.es.v0, '1 1 h d -> b 1 h d', b=b), out], dim=1)
        out = rearrange(out, 'b t h d -> b t (h d)')
        return self.out_proj(out)


class FourierLayer(nn.Module):

    def __init__(self, d_model, pred_len, k=None, low_freq=1):
        super().__init__()
        self.d_model = d_model
        self.pred_len = pred_len
        self.k = k
        self.low_freq = low_freq

    def forward(self, x):
        """x: (b, t, d)"""
        b, t, d = x.shape
        x_freq = fft.rfft(x, dim=1)

        if t % 2 == 0:
            x_freq = x_freq[:, self.low_freq:-1]
            f = fft.rfftfreq(t)[self.low_freq:-1]
        else:
            x_freq = x_freq[:, self.low_freq:]
            f = fft.rfftfreq(t)[self.low_freq:]

        x_freq, index_tuple = self.topk_freq(x_freq)
        f = repeat(f, 'f -> b f d', b=x_freq.size(0), d=x_freq.size(2))
        f = rearrange(f[index_tuple], 'b f d -> b f () d').to(x_freq.device)

        return self.extrapolate(x_freq, f, t)

    def extrapolate(self, x_freq, f, t):
        x_freq = torch.cat([x_freq, x_freq.conj()], dim=1)
        f = torch.cat([f, -f], dim=1)
        t_val = rearrange(torch.arange(t + self.pred_len, dtype=torch.float),
                          't -> () () t ()').to(x_freq.device)

        amp = rearrange(x_freq.abs() / t, 'b f d -> b f () d')
        phase = rearrange(x_freq.angle(), 'b f d -> b f () d')

        x_time = amp * torch.cos(2 * math.pi * f * t_val + phase)

        return reduce(x_time, 'b f t d -> b t d', 'sum')

    def topk_freq(self, x_freq):
        values, indices = torch.topk(x_freq.abs(), self.k, dim=1, largest=True, sorted=True)
        mesh_a, mesh_b = torch.meshgrid(torch.arange(x_freq.size(0)), torch.arange(x_freq.size(2)))
        index_tuple = (mesh_a.unsqueeze(1).to(indices.device), indices, mesh_b.unsqueeze(1).to(indices.device))
        x_freq = x_freq[index_tuple]

        return x_freq, index_tuple


class LevelLayer(nn.Module):

    def __init__(self, d_model, c_out, dropout=0.1):
        super().__init__()
        self.d_model = d_model
        self.c_out = c_out

        self.es = ExponentialSmoothing(1, self.c_out, dropout=dropout, aux=True)
        self.growth_pred = nn.Linear(self.d_model, self.c_out)
        self.season_pred = nn.Linear(self.d_model, self.c_out)

    def forward(self, level, growth, season):
        b, t, _ = level.shape
        growth = self.growth_pred(growth).view(b, t, self.c_out, 1)
        season = self.season_pred(season).view(b, t, self.c_out, 1)
        growth = growth.view(b, t, self.c_out, 1)
        season = season.view(b, t, self.c_out, 1)
        level = level.view(b, t, self.c_out, 1)
        out = self.es(level - season, aux_values=growth)
        out = rearrange(out, 'b t h d -> b t (h d)')
        return out


class EncoderLayer(nn.Module):

    def __init__(self, d_model, nhead, c_out, seq_len, pred_len, k, dim_feedforward=None, dropout=0.1,
                 activation='sigmoid', layer_norm_eps=1e-5):
        super().__init__()
        self.d_model = d_model
        self.nhead = nhead
        self.c_out = c_out
        self.seq_len = seq_len
        self.pred_len = pred_len
        dim_feedforward = dim_feedforward or 4 * d_model
        self.dim_feedforward = dim_feedforward

        self.growth_layer = GrowthLayer(d_model, nhead, dropout=dropout)
        self.seasonal_layer = FourierLayer(d_model, pred_len, k=k)
        self.level_layer = LevelLayer(d_model, c_out, dropout=dropout)

        # Implementation of Feedforward model
        self.ff = Feedforward(d_model, dim_feedforward, dropout=dropout, activation=activation)
        self.norm1 = nn.LayerNorm(d_model, eps=layer_norm_eps)
        self.norm2 = nn.LayerNorm(d_model, eps=layer_norm_eps)

        self.dropout1 = nn.Dropout(dropout)
        self.dropout2 = nn.Dropout(dropout)

    def forward(self, res, level, attn_mask=None):
        season = self._season_block(res)
        res = res - season[:, :-self.pred_len]
        growth = self._growth_block(res)
        res = self.norm1(res - growth[:, 1:])
        res = self.norm2(res + self.ff(res))

        level = self.level_layer(level, growth[:, :-1], season[:, :-self.pred_len])
        return res, level, growth, season

    def _growth_block(self, x):
        x = self.growth_layer(x)
        return self.dropout1(x)

    def _season_block(self, x):
        x = self.seasonal_layer(x)
        return self.dropout2(x)


class Encoder(nn.Module):

    def __init__(self, layers):
        super().__init__()
        self.layers = nn.ModuleList(layers)

    def forward(self, res, level, attn_mask=None):
        growths = []
        seasons = []
        for layer in self.layers:
            res, level, growth, season = layer(res, level, attn_mask=None)
            growths.append(growth)
            seasons.append(season)

        return level, growths, seasons


class DampingLayer(nn.Module):

    def __init__(self, pred_len, nhead, dropout=0.1):
        super().__init__()
        self.pred_len = pred_len
        self.nhead = nhead
        self._damping_factor = nn.Parameter(torch.randn(1, nhead))
        self.dropout = nn.Dropout(dropout)

    def forward(self, x):
        x = repeat(x, 'b 1 d -> b t d', t=self.pred_len)
        b, t, d = x.shape

        powers = torch.arange(self.pred_len).to(self._damping_factor.device) + 1
        powers = powers.view(self.pred_len, 1)
        damping_factors = self.damping_factor ** powers
        damping_factors = damping_factors.cumsum(dim=0)
        x = x.view(b, t, self.nhead, -1)
        x = self.dropout(x) * damping_factors.unsqueeze(-1)
        return x.view(b, t, d)

    @property
    def damping_factor(self):
        return torch.sigmoid(self._damping_factor)


class DecoderLayer(nn.Module):

    def __init__(self, d_model, nhead, c_out, pred_len, dropout=0.1):
        super().__init__()
        self.d_model = d_model
        self.nhead = nhead
        self.c_out = c_out
        self.pred_len = pred_len

        self.growth_damping = DampingLayer(pred_len, nhead, dropout=dropout)
        self.dropout1 = nn.Dropout(dropout)

    def forward(self, growth, season):
        growth_horizon = self.growth_damping(growth[:, -1:])
        growth_horizon = self.dropout1(growth_horizon)

        seasonal_horizon = season[:, -self.pred_len:]
        return growth_horizon, seasonal_horizon


class Decoder(nn.Module):

    def __init__(self, layers):
        super().__init__()
        self.d_model = layers[0].d_model
        self.c_out = layers[0].c_out
        self.pred_len = layers[0].pred_len
        self.nhead = layers[0].nhead

        self.layers = nn.ModuleList(layers)
        self.pred = nn.Linear(self.d_model, self.c_out)

    def forward(self, growths, seasons):
        growth_repr = []
        season_repr = []

        for idx, layer in enumerate(self.layers):
            growth_horizon, season_horizon = layer(growths[idx], seasons[idx])
            growth_repr.append(growth_horizon)
            season_repr.append(season_horizon)
        growth_repr = sum(growth_repr)
        season_repr = sum(season_repr)
        return self.pred(growth_repr), self.pred(season_repr)