sklearn.model_selection GridSearchCV is throwing KeyError: 'mean_train_score'

Question

I am trying to fetch mean train score from classifierobject.cv_result_ from GridSearchCV method using 5 fold cross-validation but it's giving me *** KeyError: 'mean_train_score' in Google Colab. However, the same code is running fine in local machine Ipython notebook for sklearn version 0.19.1. Can anyone help me how can I get the output in google colab?

clf.cv_results_.keys() ouputs following -

for my local notebook -

dict_keys(['mean_fit_time', 'std_fit_time', 'mean_score_time', 'std_score_time', 'param_n_neighbors', 'params', 'split0_test_score', 'split1_test_score', 'split2_test_score', 'split3_test_score', 'split4_test_score', 'mean_test_score', 'std_test_score', 'rank_test_score', 'split0_train_score', 'split1_train_score', 'split2_train_score', 'split3_train_score', 'split4_train_score', 'mean_train_score', 'std_train_score'])

on google colab notebook -

dict_keys(['mean_fit_time', 'std_fit_time', 'mean_score_time', 'std_score_time', 'param_n_neighbors', 'params', 'split0_test_score', 'split1_test_score', 'split2_test_score', 'split3_test_score', 'split4_test_score', 'mean_test_score', 'std_test_score', 'rank_test_score'])

where is the mean_train_score in google colab. Am I missing something?

from sklearn.model_selection import GridSearchCV

neigh = KNeighborsClassifier(n_jobs=-1)
parameters = {'n_neighbors':[1, 5, 10, 15, 19 , 21, 31, 41, 51]}
clf = GridSearchCV(neigh, parameters, cv=5,\ 
                   scoring='roc_auc',n_jobs=-1) 
clf.fit(x_train, y_train)
train_auc= clf.cv_results_['mean_train_score']
train_auc_std= clf.cv_results_['std_train_score']
cv_auc = clf.cv_results_['mean_test_score'] 
cv_auc_std= clf.cv_results_['std_test_score']

Answer 1

Try setting return_train_score=True inside GridSearchCV() to calculate the train scores (are off by default, see docs). Maybe you have kind of global variable in sklearn somewhere local set.