run_emb_user.py

from sklearn.feature_extraction.text import TfidfTransformer
from sklearn.pipeline import FeatureUnion, Pipeline
from sklearn.preprocessing import Normalizer, StandardScaler
from sklearn.svm import SVC
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier, GradientBoostingClassifier
from sklearn.neighbors import KNeighborsClassifier
from sklearn.naive_bayes import MultinomialNB
from tea.features import *
from tea.load_data import parse_reviews
from tea.run_models import run_grid_search
from sklearn.decomposition import PCA

if __name__ == "__main__":
    data = parse_reviews(load_data=False)

    X_train = data.drop(['polarity'], axis=1)
    y_train = data['polarity']

    X_train_lemmatized = pd.DataFrame(LemmaExtractor(col_name='text').fit_transform(X_train))

    we_obj = WordEmbedding()

    pre_loaded_we = {
        # 50: we_obj.get_word_embeddings(dimension=50),
        # 100: we_obj.get_word_embeddings(dimension=100),
        # 200: we_obj.get_word_embeddings(dimension=200),
        300: we_obj.get_word_embeddings(dimension=300)
    }

    final_pipeline = Pipeline(
        [
            ('features', FeatureUnion(transformer_list=[
                ('text_length', TextLengthExtractor(col_name='text')),
                ('avg_token_length', WordLengthMetricsExtractor(col_name='text', metric='avg', split_type='simple')),
                ('std_token_length', WordLengthMetricsExtractor(col_name='text', metric='std', split_type='simple')),
                ('contains_spc', ContainsSpecialCharactersExtractor(col_name='text')),
                ('n_tokens', NumberOfTokensCalculator(col_name='text')),
                ('contains_dots_bool', ContainsSequentialChars(col_name='text', pattern='..')),
                ('contains_excl_bool', ContainsSequentialChars(col_name='text', pattern='!!')),
                ('sentiment_positive', HasSentimentWordsExtractor(col_name='text', sentiment='positive')),
                ('sentiment_negative', HasSentimentWordsExtractor(col_name='text', sentiment='negative')),
                ('contains_uppercase', ContainsUppercaseWords(col_name='text', reshape=True)),
                ('embedding_feat', SentenceEmbeddingExtractor(col_name='text', word_embeddings_dict=pre_loaded_we))
            ])),
            # ('scaling', StandardScaler()),
            ('scaling', StandardScaler()),
            ('pca', PCA()),
            # ('clf', SVC()),
            # ('clf', MultinomialNB())
            ('clf', LogisticRegression())
            # ('clf', KNeighborsClassifier())
            # ('clf', GradientBoostingClassifier())
            # ('clf', RandomForestClassifier())
        ])

    params = {
        'features__sentiment_positive__count_type': ['counts', ],  # 'boolean',
        'features__sentiment_negative__count_type': ['counts', ],  # 'boolean',
        'features__contains_uppercase__how': ['count', ],  # 'bool',
        'features__embedding_feat__embedding_type': ['tfidf'],  # embedding # 'tfidf',
        'features__embedding_feat__embedding_dimensions': [300, ],  # embedding  100, 200, 300
        'pca__n_components': [100, ],
        'clf__penalty': ('l2',),  # Logistic  'l2'
        # 'clf__kernel': ('rbf', 'linear'),  # SVM
        # 'clf__gamma': (0.1, 0.01, 0.001, 0.0001),  # SVM
        # 'clf__p': (1, 2),  # 1: mahnatan, 2: eucledian # k-NN
        # 'clf__n_neighbors': (3, 4, 5, 6, 7, 8),  # k-NN
        # 'clf__learning_rate': (0.1, 0.01, 0.001),  # Gradient Boosting
        # 'clf__n_estimators': (100, 300, 600),  # Gradient Boosting, Random Forest
        # 'clf__alpha': (0.1, 0.5, 1.0),  # MultinomialNB
        # 'clf__fit_prior': (True, False),  # MultinomialNB
        # 'clf__max_depth': [10, 50, 100, None],  # Random Forest
    }

    grid_results = run_grid_search(X=X_train_lemmatized,
                                   y=y_train,
                                   pipeline=final_pipeline,
                                   parameters=params,
                                   scoring='accuracy')