轉換包含 LightGBM 分類器的管線

sklearn-onnx 僅將 scikit-learn 模型轉換為 ONNX，但許多函式庫實作了 scikit-learn API，以便將其模型納入 scikit-learn 管線中。此範例考慮包含 LightGBM 模型的管線。sklearn-onnx 只要知道與 LGBMClassifier 關聯的轉換器，就可以轉換整個管線。讓我們看看如何做到。

訓練 LightGBM 分類器

import onnxruntime as rt
from skl2onnx import convert_sklearn, update_registered_converter
from skl2onnx.common.shape_calculator import (
    calculate_linear_classifier_output_shapes,
)  # noqa
from onnxmltools.convert.lightgbm.operator_converters.LightGbm import (
    convert_lightgbm,
)  # noqa
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 0.000695 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 環境中，請重新執行此儲存格以顯示 HTML 表示，或信任筆記本。
在 GitHub 上，HTML 表示無法呈現，請嘗試使用 nbviewer.org 載入此頁面。

註冊 LGBMClassifier 的轉換器

轉換器在 onnxmltools 中實作：onnxmltools…LightGbm.py。以及形狀計算器：onnxmltools…Classifier.py。

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

再次轉換

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())

比較預測結果

使用 LightGbm 的預測結果。

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

predict [1 2 1 0 1]
predict_proba [[0.25335584 0.45934348 0.28730068]]

使用 onnxruntime 的預測結果。

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 [1 2 1 0 1]
predict_proba [{0: 0.25335583090782166, 1: 0.45934349298477173, 2: 0.287300705909729}]