转换带有 LightGBM 分类器的管道

sklearn-onnx 仅将 scikit-learn 模型转换为 ONNX，但许多库实现了 scikit-learn API，以便其模型可以包含在 scikit-learn 管道中。本示例考虑一个包含 LightGBM 模型的管道。sklearn-onnx 可以转换整个管道，前提是它知道与 LGBMClassifier 相关的转换器。让我们看看如何做到这一点。

训练 LightGBM 分类器

import onnxruntime as rt
from skl2onnx import convert_sklearn, update_registered_converter
from skl2onnx.common.shape_calculator import (
    calculate_linear_classifier_output_shapes,
)
from onnxmltools.convert.lightgbm.operator_converters.LightGbm import (
    convert_lightgbm,
)
from skl2onnx.common.data_types import FloatTensorType
import numpy
from sklearn.datasets import load_iris
from sklearn.pipeline import Pipeline
from sklearn.preprocessing import StandardScaler
from lightgbm import LGBMClassifier

data = load_iris()
X = data.data[:, :2]
y = data.target

ind = numpy.arange(X.shape[0])
numpy.random.shuffle(ind)
X = X[ind, :].copy()
y = y[ind].copy()

pipe = Pipeline(
    [("scaler", StandardScaler()), ("lgbm", LGBMClassifier(n_estimators=3))]
)
pipe.fit(X, y)

[LightGBM] [Info] Auto-choosing col-wise multi-threading, the overhead of testing was 0.018751 seconds.
You can set `force_col_wise=true` to remove the overhead.
[LightGBM] [Info] Total Bins 47
[LightGBM] [Info] Number of data points in the train set: 150, number of used features: 2
[LightGBM] [Info] Start training from score -1.098612
[LightGBM] [Info] Start training from score -1.098612
[LightGBM] [Info] Start training from score -1.098612
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf
[LightGBM] [Warning] No further splits with positive gain, best gain: -inf

Pipeline(steps=[('scaler', StandardScaler()),
                ('lgbm', LGBMClassifier(n_estimators=3))])

在 Jupyter 环境中，请重新运行此单元格以显示 HTML 表示或信任笔记本。
在 GitHub 上，HTML 表示无法渲染，请尝试使用 nbviewer.org 加载此页面。

注册 LGBMClassifier 的转换器

该转换器在 onnxmltools 中实现：onnxmltools...LightGbm.py，以及形状计算器：onnxmltools...Classifier.py。

update_registered_converter(
    LGBMClassifier,
    "LightGbmLGBMClassifier",
    calculate_linear_classifier_output_shapes,
    convert_lightgbm,
    options={"nocl": [True, False], "zipmap": [True, False, "columns"]},
)

再次转换

model_onnx = convert_sklearn(
    pipe,
    "pipeline_lightgbm",
    [("input", FloatTensorType([None, 2]))],
    target_opset={"": 12, "ai.onnx.ml": 2},
)

# And save.
with open("pipeline_lightgbm.onnx", "wb") as f:
    f.write(model_onnx.SerializeToString())

比较预测结果

使用 LightGbm 的预测结果。

print("predict", pipe.predict(X[:5]))
print("predict_proba", pipe.predict_proba(X[:1]))

predict [2 0 0 0 1]
predict_proba [[0.22814003 0.31657806 0.45528191]]

使用 onnxruntime 的预测结果。

sess = rt.InferenceSession("pipeline_lightgbm.onnx", providers=["CPUExecutionProvider"])

pred_onx = sess.run(None, {"input": X[:5].astype(numpy.float32)})
print("predict", pred_onx[0])
print("predict_proba", pred_onx[1][:1])

predict [2 0 0 0 1]
predict_proba [{0: 0.22814001142978668, 1: 0.3165780305862427, 2: 0.45528194308280945}]