Different result of classification with same classifier and same input parameters

Question

I did a binary classification using "Random Forest".

The code block is

clf = RandomForestClassifier()
clf.fit(X_train, y_train)
R_y_pred = clf.predict(X_test)
print(classification_report(y_test, R_y_pred))

The result is

                precision    recall  f1-score   support

           0       0.91      0.98      0.94      1023
           1       *0.79      0.48*      0.60       185

    accuracy                           0.90      1208
   macro avg       0.85      0.73      0.77      1208
weighted avg       0.89      0.90      0.89      1208

When I apply clf.get_params() command to see the default parameters, I got

{'bootstrap': True,
 'ccp_alpha': 0.0,
 'class_weight': None,
 'criterion': 'gini',
 'max_depth': None,
 'max_features': 'sqrt',
 'max_leaf_nodes': None,
 'max_samples': None,
 'min_impurity_decrease': 0.0,
 'min_samples_leaf': 1,
 'min_samples_split': 2,
 'min_weight_fraction_leaf': 0.0,
 'n_estimators': 100,
 'n_jobs': None,
 'oob_score': False,
 'random_state': None,
 'verbose': 0,
 'warm_start': False}

Now in another code, I defined the criterion for RandomForestClassifier

The code block is

cri_clf = RandomForestClassifier(criterion = 'gini')
cri_clf.fit(X_train, y_train)
cri_y_pred = cri_clf.predict(X_test)
print(classification_report(y_test, cri_y_pred))

The result is

                  precision    recall  f1-score   support

           0       0.91      0.98      0.94      1023
           1       *0.80      0.46*      0.59       185

    accuracy                           0.90      1208
   macro avg       0.86      0.72      0.77      1208
weighted avg       0.89      0.90      0.89      1208

So, you can see that there is a slight difference in the result of precision and recall when I define a criterion explicitly with not defining a criterion.

If all the parameters are the same for two codes why do I get the differences between the two results?

Thank you.

Answer 1

From sklearns random forest documentation:

random_state int, RandomState instance or None, default=None

Controls both the randomness of the bootstrapping of the samples used when building trees (if bootstrap=True) and the sampling of the features to consider when looking for the best split at each node (if max_features < n_features). See Glossary for details.

Each time you re-run this with random_state = None it runs different models.

Set random_state to 0 (or any number) and see consistent results.