Skip to content

sklego/estimator-transformer-proba

Rule estimator-transformer-proba
Pack scikit-lego
Severity error
Fix none, edit by hand
Library versions every version

EstimatorTransformer reshapes its output to one column, so predict_proba comes back with twice as many rows and a pipeline around it raises. Use predict_func="decision_function" for a binary model, or scikit-learn's StackingClassifier.

features = EstimatorTransformer(LogisticRegression(), predict_func="predict_proba")
features = EstimatorTransformer(LogisticRegression(), predict_func="decision_function")
stack = StackingClassifier([("lr", LogisticRegression())], stack_method="predict_proba")

The rule reports EstimatorTransformer(..., predict_func="predict_proba") and "predict_log_proba". When the target has one dimension, the transformer reshapes the output to one column, so the probabilities of the two classes stack into twice as many rows.

xpd has no automatic fix, so make the change by hand.

A binary model on 400 rows came back from transform with the shape (800, 1). Inside a FeatureUnion, and inside a pipeline before a model, fit raised ValueError on the row counts. decision_function gave (400, 1), and StackingClassifier(stack_method="predict_proba").transform gave one row per input row.

xpd prints this message with the issue. A name that starts with $ stands for code from your file.

EstimatorTransformer reshapes the output to one column, so predict_proba on 400 rows came back with the shape (800, 1), and a pipeline around it raised ValueError. Use predict_func="decision_function" for a binary model, or scikit-learn's StackingClassifier(stack_method="predict_proba").

Browse the other rules of the scikit-lego pack.