Random Forest with GridSearchCV - Error on param_grid

Question

Im trying to create a Random Forest model with GridSearchCV but am getting an error pertaining to param_grid: "ValueError: Invalid parameter max_features for estimator Pipeline. Check the list of available parameters with `estimator.get_params().keys()". I'm classifying documents so I am also pushing tf-idf vectorizer to the pipeline. Here is the code:

from sklearn import metrics
from sklearn.ensemble import RandomForestClassifier
from sklearn.metrics import classification_report, f1_score, accuracy_score, precision_score, confusion_matrix
from sklearn.pipeline import Pipeline

 #Classifier Pipeline
pipeline = Pipeline([
    ('tfidf', TfidfVectorizer()),
    ('classifier', RandomForestClassifier())
])
# Params for classifier
params = {"max_depth": [3, None],
              "max_features": [1, 3, 10],
              "min_samples_split": [1, 3, 10],
              "min_samples_leaf": [1, 3, 10],
              # "bootstrap": [True, False],
              "criterion": ["gini", "entropy"]}

# Grid Search Execute
rf_grid = GridSearchCV(estimator=pipeline , param_grid=params) #cv=10
rf_detector = rf_grid.fit(X_train, Y_train)
print(rf_grid.grid_scores_)

I can't figure out why the error is showing. The same btw is occurring when I run a decision tree with GridSearchCV. (Scikit-learn 0.17)

Solution 1

You have to assign the parameters to the named step in the pipeline. In your case classifier. Try prepending classifier__ to the parameter name. Sample pipeline

params = {"classifier__max_depth": [3, None],
              "classifier__max_features": [1, 3, 10],
              "classifier__min_samples_split": [1, 3, 10],
              "classifier__min_samples_leaf": [1, 3, 10],
              # "bootstrap": [True, False],
              "classifier__criterion": ["gini", "entropy"]}

Solution 2

Try to run get_params() on your final pipeline object, not just the estimator. This way it'd generate all available pipe-items unique keys for the grid parameters.

sorted(pipeline.get_params().keys())

['classifier', 'classifier__bootstrap', 'classifier__class_weight', 'classifier__criterion', 'classifier__max_depth', 'classifier__max_features', 'classifier__max_leaf_nodes', 'classifier__min_impurity_split', 'classifier__min_samples_leaf', 'classifier__min_samples_split', 'classifier__min_weight_fraction_leaf', 'classifier__n_estimators', 'classifier__n_jobs', 'classifier__oob_score', 'classifier__random_state', 'classifier__verbose', 'classifier__warm_start', 'steps', 'tfidf', 'tfidf__analyzer', 'tfidf__binary', 'tfidf__decode_error', 'tfidf__dtype', 'tfidf__encoding', 'tfidf__input', 'tfidf__lowercase', 'tfidf__max_df', 'tfidf__max_features', 'tfidf__min_df', 'tfidf__ngram_range', 'tfidf__norm', 'tfidf__preprocessor', 'tfidf__smooth_idf', 'tfidf__stop_words', 'tfidf__strip_accents', 'tfidf__sublinear_tf', 'tfidf__token_pattern', 'tfidf__tokenizer', 'tfidf__use_idf', 'tfidf__vocabulary']

This is especially useful when you're using the short make_pipeline() syntax for Piplines, where you don't bother with labels for pipe items:

pipeline = make_pipeline(TfidfVectorizer(), RandomForestClassifier())
sorted(pipeline.get_params().keys())

['randomforestclassifier', 'randomforestclassifier__bootstrap', 'randomforestclassifier__class_weight', 'randomforestclassifier__criterion', 'randomforestclassifier__max_depth', 'randomforestclassifier__max_features', 'randomforestclassifier__max_leaf_nodes', 'randomforestclassifier__min_impurity_split', 'randomforestclassifier__min_samples_leaf', 'randomforestclassifier__min_samples_split', 'randomforestclassifier__min_weight_fraction_leaf', 'randomforestclassifier__n_estimators', 'randomforestclassifier__n_jobs', 'randomforestclassifier__oob_score', 'randomforestclassifier__random_state', 'randomforestclassifier__verbose', 'randomforestclassifier__warm_start', 'steps', 'tfidfvectorizer', 'tfidfvectorizer__analyzer', 'tfidfvectorizer__binary', 'tfidfvectorizer__decode_error', 'tfidfvectorizer__dtype', 'tfidfvectorizer__encoding', 'tfidfvectorizer__input', 'tfidfvectorizer__lowercase', 'tfidfvectorizer__max_df', 'tfidfvectorizer__max_features', 'tfidfvectorizer__min_df', 'tfidfvectorizer__ngram_range', 'tfidfvectorizer__norm', 'tfidfvectorizer__preprocessor', 'tfidfvectorizer__smooth_idf', 'tfidfvectorizer__stop_words', 'tfidfvectorizer__strip_accents', 'tfidfvectorizer__sublinear_tf', 'tfidfvectorizer__token_pattern', 'tfidfvectorizer__tokenizer', 'tfidfvectorizer__use_idf', 'tfidfvectorizer__vocabulary']