在 Python 中使用 GridSearch 从 sklearn 模型生成 PMML

Question

我希望在 sklearn 中训练模型并通过 PMML 将它们导出到其他环境（使用 https://github.com/jpmml/sklearn2pmml）

我能够从普通（k-Nearest Neighbors）模型（没有 GridSearch）生成 PMML，但使用 GridSearch 时会出现以下错误 -

TypeError: 管道对象不是 PMMLPipeline 的实例

该错误是有道理的（因为 GridSearchCV 不返回 PMMLPipeline），但正在寻找有关如何将优化（w/GridSearch）模型导出到 PMML 的想法（例如，有没有办法在 PMMLPipeline 中包含 GridSearch）。

下面的代码 - 任何想法的 TIA。

from sklearn.neighbors import KNeighborsRegressor
from sklearn.model_selection import GridSearchCV

knn_pipe = PMMLPipeline([
("regressor", KNeighborsRegressor())
])

param_grid = {"regressor__n_neighbors": [3, 2,10],
          "regressor__weights": ["uniform","distance"],
          "regressor__algorithm": ["auto", "ball_tree", "kd_tree"]}

cv = GridSearchCV(knn_pipe, param_grid=param_grid)

print(train.drop('y',axis=1).shape)

cv.fit(X,Y)

best_parameters = cv.best_estimator_.get_params()
print("best parameter = {}".format(best_parameters))

from sklearn2pmml import sklearn2pmml
sklearn2pmml(cv, "kNNMercedes.pmml", with_repr = True)




['regressor__algorithm', 'regressor__n_neighbors', 'regressor__metric', 
'regressor__leaf_size', 'regressor', 'regressor__p', 
'regressor__metric_params', 'steps', 'regressor__n_jobs', 
'regressor__weights']
 (4209, 365)
best parameter = {'regressor__algorithm': 'auto', 'regressor__n_neighbors': 
10, 'regressor__metric': 'minkowski', 'regressor__leaf_size': 30, 'regressor': 
KNeighborsRegressor(algorithm='auto', leaf_size=30, metric='minkowski',
          metric_params=None, n_jobs=1, n_neighbors=10, p=2,
      weights='distance'), 'regressor__p': 2, 'regressor__metric_params': None, 'steps': [('regressor', KNeighborsRegressor(algorithm='auto', leaf_size=30, metric='minkowski',
      metric_params=None, n_jobs=1, n_neighbors=10, p=2,
      weights='distance'))], 'regressor__n_jobs': 1, 'regressor__weights': 
'distance'}
---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-38-d548c1bff799> in <module>()
 30 
 31 from sklearn2pmml import sklearn2pmml
---> 32 sklearn2pmml(cv, "kNNMercedes.pmml", with_repr = True)
 33 
 34 print("yeay PMML!")

/Users/venuv/.local/lib/python2.7/site-packages/sklearn2pmml/__init__.pyc in 
sklearn2pmml(pipeline, pmml, user_classpath, with_repr, debug)
125                 print("sklearn2pmml: ", __version__)
126         if(not isinstance(pipeline, PMMLPipeline)):
--> 127                 raise TypeError("The pipeline object is not an 
instance of " + PMMLPipeline.__name__)
128         cmd = ["java", "-cp", os.pathsep.join(_package_classpath() + user_classpath), "org.jpmml.sklearn.Main"]
129         dumps = []

TypeError: The pipeline object is not an instance of PMMLPipeline

Answer 1

解决方案是在拟合的GridSearchCV 实例之外构造一个PMMLPipeline 实例：

pipeline = PMMLPipeline([
  ("best_estimator", cv.best_estimator_)
])
sklearn2pmml(pipeline, "pipeline.pmml")

用户@vivek-kumar 报告了 JPMML-SkLearn 项目的这个问题，并在那里收到了一些额外的 cmets。见jpmml/jpmml-sklearn#42