sklearn:'GradientBoostingClassifier' object has no attribute 'n_features_'

When I try to Optimize Gradient Boosting Classifier,the error is "train_predictions = best_gbrt_clf.predict(X_train)"

this is the code:

from sklearn.model_selection  import GridSearchCV
from sklearn.metrics          import fbeta_score, make_scorer
from sklearn.model_selection  import StratifiedShuffleSplit
from sklearn.externals        import joblib

start_time=time.clock()

Dump_clf = False
Load_clf = True
clf = GradientBoostingClassifier(random_state=0)

if (Load_clf) :
    best_gbrt_clf = joblib.load('best_gbrt_clf.pkl')
else: # build and dump new classifier
    #  the following hyper-parameter values were recommended in the Prettenhofer tutorial
    parameters = {'n_estimators': [100],
                  'learning_rate': [0.1, 0.05, 0.02, 0.01], 
                  'max_depth': [4, 6],
                  'min_samples_leaf': [3, 5, 9, 17], 
                  'max_features': [1.0, 0.3, 0.1]
                 }
    # TODO: Make an fbeta_score scoring object
    scorer = make_scorer(fbeta_score, beta=0.5, average='weighted')

    #  Use a stratified sample because there are a lot more examples of one class that the other in the input data
    cv = StratifiedShuffleSplit(y_train, test_size=0.2, random_state=42)

    # TODO: Perform grid search on the classifier using 'scorer' as the scoring method
    grid_obj = GridSearchCV(clf, param_grid=parameters, cv=cv, scoring=scorer)

    # TODO: Fit the grid search object to the training data and find the optimal parameters
    grid_fit = grid_obj.fit(X_train, y_train)

    # Get the estimator
    best_gbrt_clf    = grid_obj.best_estimator_
    print("best params found by gridsearchCV=%s" % grid_obj.best_params_)
    # best_params = grid_obj.best_params_
    if (Dump_clf): 
        joblib.dump(best_gbrt_clf, 'best_gbrt_clf.pkl')
best_name   = best_gbrt_clf.__class__.__name__
best_params = best_gbrt_clf.get_params()

print("Best classifier is %s" % best_name)
print("Best params=%s\n"      % best_params)

# Make predictions using the unoptimized and model
train_predictions = best_gbrt_clf.predict(X_train)
test_predictions  = (clf.fit(X_train, y_train)).predict(X_test)
best_predictions  = best_gbrt_clf.predict(X_test)
end_time=time.clock()

I am trying to compare different classifiers and I can get the result of RandomForestClassifier(best_rf)

the vision of scikit-learn is 0.20 python 3.6