Let's now try to improve the information available to the classifier by including bigrams and trigrams, as follows:
lr_clf = Pipeline([('vect', CountVectorizer(stop_words='english', ngram_range=(1,3))), ('tfidf', TfidfTransformer()), ('clf',LR())])lr_clf.fit(X=X_train, y=y_train)lr_acc, lr_predictions = imdb_acc(lr_clf)lr_acc # 0.86596