model = build_LSTM(args=args_lstm)
early_stopping = EarlyStopping(monitor='val_loss', patience=10)
hist = model.fit(data_train,
labels_train,
validation_data=(data_val, labels_val),
epochs=args_lstm.epochs,
batch_size=args_lstm.batch_size,
shuffle=True,
callbacks=[early_stopping])
# predict
print('Testing')
preds = model.predict(test_tweet, batch_size=32, verbose=1)
y_pred = preds.ravel()
acc, pre, rec, f1, auc = evaluate_prediction(
y_test=y_test,
y_pred=y_pred,
k_th=num_fold,
model_name='LSTM-word2vec',
dataset_name=args_lstm.dataset)
list_acc.append(acc)
list_pre.append(pre)
list_rec.append(rec)
list_f1.append(f1)
list_auc.append(auc)
result.append(['LSTM-word2vec', acc, pre, rec, f1, auc])
n_sampling -= 1
result.append([
'average',
np.mean(list_acc),
np.mean(list_pre),
np.mean(list_rec),
np.mean(list_f1),
y = df_all['y'].as_matrix()
# 10-fold cross validation
num_fold = 10
kf = KFold(n_splits=num_fold, shuffle=True, random_state=0)
for train_index, test_index in kf.split(X):
num_fold -= 1
X_train, X_test = X[train_index], X[test_index]
y_train, y_test = y[train_index], y[test_index]
# Logisitc Regression
clf = LogisticRegression(penalty='l2', tol=1e-6)
clf.fit(X_train, y_train)
y_pred = clf.predict_proba(X_test)[:, 1]
acc, pre, rec, f1, auc = evaluate_prediction(
y_test,
y_pred,
k_th=num_fold,
model_name='Logistic Regression',
dataset_name=args.dataset)
lr_acc.append(acc)
lr_pre.append(pre)
lr_rec.append(rec)
lr_f1.append(f1)
lr_auc.append(auc)
# Random Forest
clf = RandomForestClassifier(n_estimators=20,
max_depth=8,
random_state=0)
clf.fit(X_train, y_train)
y_pred = clf.predict_proba(X_test)[:, 1]
acc, pre, rec, f1, auc = evaluate_prediction(
y_test,
tokenizer.fit_on_texts(list(title_train)+list(usertext_train))
tokenized_title_train = tokenizer.texts_to_sequences(title_train)
tokenized_title_test = tokenizer.texts_to_sequences(title_test)
tokenized_usertext_train = tokenizer.texts_to_sequences(usertext_train)
tokenized_usertext_test = tokenizer.texts_to_sequences(usertext_test)
X1_train = sequence.pad_sequences(tokenized_title_train, maxlen=args_rnn.max_seq_len, dtype='float64')
X1_test = sequence.pad_sequences(tokenized_title_test, maxlen=args_rnn.max_seq_len, dtype='float64')
X2_train = sequence.pad_sequences(tokenized_usertext_train, maxlen=args_rnn.max_seq_len, dtype='float64')
X2_test = sequence.pad_sequences(tokenized_usertext_test, maxlen=args_rnn.max_seq_len, dtype='float64')
# train model and predict
model = build_RNN(args=args_rnn)
early = EarlyStopping(monitor="val_loss", mode="min", patience=20)
model.fit([X1_train, X2_train], y_train, batch_size=args_rnn.batch_size, epochs=args_rnn.epochs,
validation_split=0.1, callbacks=[early])
y_pred = model.predict([X1_test, X2_test]).reshape(y_test.shape)
acc, pre, rec, f1, auc = evaluate_prediction(y_test, y_pred, k_th=num_fold, model_name='RNN',
dataset_name=args_rnn.dataset)
list_acc.append(acc)
list_pre.append(pre)
list_rec.append(rec)
list_f1.append(f1)
list_auc.append(auc)
result.append(['RNN', acc, pre, rec, f1, auc])
num_fold += 1
n_sampling -= 1
result.append(['average', np.mean(list_acc), np.mean(list_pre),
np.mean(list_rec), np.mean(list_f1), np.mean(list_auc)])
print(tabulate(result, headers=h))