python-3.x - 如何在 scikit-learn 中访问管道 GridSearchCV 内估计器的属性？

Question

我想访问 DecisionTreeRegressor 的 feature_importances_ 属性 在下面的代码中:

#Create an estimator
from sklearn.tree import DecisionTreeRegressor
tree_reg = DecisionTreeRegressor(criterion='mse', random_state=0)

#Create parametre grid for GridSearchCV
param_grid = {  'max_depth':np.linspace(start=4, stop=12, num=9),
                'max_leaf_nodes':[i for i in range(10,20,1)]}

# Construct gridsearchcv on param space
from sklearn.model_selection import ShuffleSplit
from sklearn.model_selection import GridSearchCV
cv = ShuffleSplit(n_splits=10, test_size=0.30, random_state=0)
grid = GridSearchCV(estimator=tree_reg, param_grid=param_grid, cv=cv, refit=True)

#Make Pipeline
from sklearn.pipeline import Pipeline
pipe = Pipeline(steps=[('preprocess', StandardScaler()), ('grid_search', grid)])
pipe.fit(X_train, y_train)

feat_impo = tree_reg.feature_importances_ #getting ERROR on this line

我想访问 DecisionTreeRegressor 的 feature_importances_ 属性，但在执行 tree_reg.feature_importances_ 时出现以下错误:

sklearn.exceptions.NotFittedError: This DecisionTreeRegressor instance is
not fitted yet. Call 'fit' with appropriate arguments before using this method.

我也尝试过这个:

grid.__getattribute__('estimator').feature_importances_

但我得到了完全相同的结果。

但是当我在没有管道和网格搜索的情况下运行程序时，即仅 使用 DecisionTreeRegressor 然后我可以使用 tree_reg.feature_importances_ 轻松访问 feature_importances_ 并获得理想的结果，没有任何错误。

如何访问 DecisionTreeRegressor 的 feature_importances_ 属性？

Answer 1

终于找到正确的方法了

best_est = grid.best_estimator_
feat_impo = best_est.feature_importances_