Question

在将'roc_auc'计分器与GridSearchCV一起使用时，出现一个奇怪的错误。当我使用'accuracy'时，不会发生该错误。查看堆栈跟踪，看起来y_score中的roc_curve作为None传递了，这导致了column_or_1d的错误。我通过使用column_or_1d作为输入直接调用None进行了测试，并轻松地重现了错误。

from sklearn.pipeline import Pipeline
from sklearn.model_selection import GridSearchCV
from sklearn.preprocessing import StandardScaler, MaxAbsScaler, MinMaxScaler
from sklearn.decomposition import PCA
from sklearn.neighbors import KNeighborsClassifier
from sklearn.ensemble import GradientBoostingClassifier
from pipelinehelper.pipelinehelper import PipelineHelper

pipe = Pipeline([
    ('scaler', PipelineHelper([
        ('std', StandardScaler()),
        ('abs', MaxAbsScaler()),
        ('minmax', MinMaxScaler()),
        ('pca', PCA(svd_solver='full', whiten=True)),
    ])),
    ('classifier', PipelineHelper([
        ('knn', KNeighborsClassifier(weights='distance')),
        ('gbc', GradientBoostingClassifier())
    ])),
])
params = {
    'scaler__selected_model': pipe.named_steps['scaler'].generate({
        'std__with_mean': [True, False],
        'std__with_std': [True, False],
        'pca__n_components': [0.5, 0.75, 0.9, 0.99],
    }),
    'classifier__selected_model': pipe.named_steps['classifier'].generate({
        'knn__n_neighbors': [1, 3, 5, 7, 10],#, 30, 50, 70, 90, 110, 130, 150, 170, 190],
        'gbc__learning_rate': [0.1, 0.5, 1.0],
        'gbc__subsample': [0.5, 1.0],
    })
}

grid = GridSearchCV(pipe, params, scoring='roc_auc', n_jobs=1, verbose=1, cv=5)
grid.fit(X, y)

一些调试信息

>>> X.shape
... (13885, 23)
>>> y.shape
... (13885,)
>>> X
... array([[ 0.        ,  0.        ,  0.        , ...,  7.14285714,
         0.9       , 35.4644354 ],
       [ 0.        ,  0.        ,  0.        , ...,  2.11442806,
         1.2       , 54.99027913],
       [ 1.        ,  0.        ,  0.        , ...,  2.64959194,
         0.7       , 70.07380534],
       ...,
       [ 1.        ,  0.        ,  0.        , ...,  4.375     ,
         0.5       , 91.85932945],
       [ 1.        ,  0.        ,  0.        , ...,  3.75      ,
         0.9       , 68.62436682],
       [ 0.        ,  0.        ,  1.        , ...,  3.01587302,
         4.1       , 57.25781074]])
>>> y
... array([0, 0, 0, ..., 0, 0, 1])
>>> y.mean()
... 0.11278357940223263
>>> sklearn.__version__
'0.20.3'

我遇到了错误：

python3.7/site-packages/sklearn/metrics/ranking.py in roc_curve(y_true, y_score, pos_label, sample_weight, drop_intermediate)
    616     """
    617     fps, tps, thresholds = _binary_clf_curve(
--> 618         y_true, y_score, pos_label=pos_label, sample_weight=sample_weight)
    619 
    620     # Attempt to drop thresholds corresponding to points in between and

python3.7/site-packages/sklearn/metrics/ranking.py in _binary_clf_curve(y_true, y_score, pos_label, sample_weight)
    399     check_consistent_length(y_true, y_score, sample_weight)
    400     y_true = column_or_1d(y_true)
--> 401     y_score = column_or_1d(y_score)
    402     assert_all_finite(y_true)
    403     assert_all_finite(y_score)

python3.7/site-packages/sklearn/utils/validation.py in column_or_1d(y, warn)
    795         return np.ravel(y)
    796 
--> 797     raise ValueError("bad input shape {0}".format(shape))
    798 
    799 

ValueError: bad input shape ()

我进一步测试了使用以下方法生成的数据，并得到了完全相同的错误：

from sklearn.datasets import make_classification
X_test, y_test = make_classification(100, 23)

我转而使用的管道不使用PipelineHelper，并且不在时会出现错误，因此我假设这是严格的问题，并且PipelineHelper有问题吗？在我继续对该项目提交错误报告之前，我想知道是否有人对解决此问题有任何想法？

pipe = Pipeline([
    ('scaler', StandardScaler()),
    ('classifier', GradientBoostingClassifier()),
])
params = {
    'scaler__with_mean': [True, False],
    'scaler__with_std': [True, False],
    'classifier__learning_rate': [0.1, 0.5, 1.0],
    'classifier__subsample': [0.5, 1.0],
}

P.S。我正在使用https://github.com/bmurauer/pipelinehelper

中的PipelineHelper

Answer 1

我继续向该项目提交a bug report，并切换到另一种解决方案found here。正如sklearn维护者on twitter所指出的那样，我还可以轻松地使用内置的sklearn工具并编写自己的代码来遍历所有选项。无论如何，我认为我建议的解决方案是不使用PipelineHelper，因为它似乎功能并不完善。

ValueError：在GridSearchCV中使用“ roc_auc”时，输入形状错误（）

1 个答案: