scikit-learn/convert2onnx.py

import argparse
import joblib

from sklearn.datasets import fetch_california_housing, load_diabetes, load_iris
from skl2onnx import convert_sklearn
from skl2onnx.common.data_types import FloatTensorType


def load_dataset(dataset_name):
    if dataset_name == 'california':
        dataset = fetch_california_housing()
    elif dataset_name == 'diabetes':
        dataset = load_diabetes()
    elif dataset_name == 'iris':
        dataset = load_iris()
    else:
        raise ValueError("Invalid dataset name")
    return dataset.data, dataset.target


def prepare_onnx_conversion_params(X, target_opset, model):
    if target_opset in {9, 17}:
        tensor = FloatTensorType([None, 4])
    else:
        tensor = FloatTensorType([None, X.shape[1]])
    if target_opset == 9:
        options = {id(model): {'zipmap': False}}
    else:
        options = None
    return tensor, options


def convert2onnx(model, initial_type, options, target_opset, onnx_filename):
    try:
        model.save_model(onnx_filename, format="onnx")
        print("Model saved in ONNX format successfully.")
    except Exception as e:
        print("Error occurred while saving model in ONNX format:", e)
        print("Trying a second method...")
        try:
            onnx_model = convert_sklearn(model, initial_types=initial_type, options=options, target_opset=target_opset)
            with open(onnx_filename, "wb") as f:
                f.write(onnx_model.SerializeToString())
            print("Model converted to ONNX format and saved successfully.")
        except Exception as e:
            print("Error occurred while converting model to ONNX format:", e)

"""
python convert2onnx.py california adaboost_regressor.joblib adaboost_regressor.onnx
"""
if __name__ == "__main__":
    parser = argparse.ArgumentParser(description='Converts a sklearn model to ONNX format.')
    parser.add_argument('dataset_name', type=str, help='Name of the dataset. Choose from: "california", "diabetes", or "iris".')
    parser.add_argument('model_path', type=str, help='Path to the trained model file.')
    parser.add_argument('onnx_filename', type=str, help='The filename for the output ONNX file.')
    args = parser.parse_args()

    X, y = load_dataset(args.dataset_name)
    model = joblib.load(args.model_path)
    target_opset = 12
    tensor, options = prepare_onnx_conversion_params(X, target_opset, model)

    input_name = 'float_input'
    initial_type = [(input_name, tensor)]
    convert2onnx(model, initial_type, options, target_opset, args.onnx_filename)