Random Forest Classifier: feature importance of prediction probability

Question

I am using sklearn RFC.

forest.fit(training_data, y_train)
probas_test = forest.predict_proba(test_data)

I wanted to know is there a way to find the contribution / importance of each features which lead to the prediction.

something like , but for an individual datapoint level.

   forest.feature_importances_

Answer 1

这可以通过多种方式解决;检查http://blog.datadive.net/interpreting-random-forests/（以及一个Python包：https://github.com/andosa/treeinterpreter）。也有较少的直接选择，例如

• https://arxiv.org/abs/1606.05390（实施：https://github.com/sato9hara/defragTrees）
• https://arxiv.org/abs/1611.05722（实施：https://github.com/IBCNServices/GENESIM）