First Classification

TF_IDF_LogReg.py

word_vectorizer = TfidfVectorizer(
    ngram_range=(1,4),
    min_df=3,
    max_df=0.9,
    use_idf=True,
    smooth_idf=True,
    sublinear_tf=True,
    analyzer='word',
    token_pattern=r'\w{1,}',
    max_features=50000)

word_vectorizer.fit(all_text)
train_word_features = word_vectorizer.transform(train_text)
test_word_features = word_vectorizer.transform(test_text)

# Make a cross validation 
kf = KFold(n_splits=5, shuffle=True, random_state=43)
test_pred_word = 0
# Take not of the scores
oof_pred_word = np.zeros([train_target.shape[0],])

for i, (train_index, val_index) in tqdm(enumerate(kf.split(train_target))):
    x_train, x_val = train_word_features[list(train_index)],train_word_features[list(val_index)]
    y_train, y_val = train_target[train_index], train_target[val_index]
    classifier = LogisticRegression(C=1, solver='sag')
    classifier.fit(x_train, y_train)
    val_preds = classifier.predict_proba(x_val)[:,1]
    preds = classifier.predict_proba(test_word_features)[:,1]
    test_pred_word += preds
    oof_pred_word[val_index] = val_preds
    print(f1_score(y_val, val_preds > 0.5))
    print(roc_auc_score(y_val, val_preds, average=None))