backward = LSTM(128,
activation='relu',
recurrent_dropout=0.2,
dropout=0.1,
return_sequences=True,
go_backwards=True)(right_embeddig)
concat = concatenate([forward, center_embedding, backward], axis=2)
semantic = TimeDistributed(Dense(100, activation='tanh'))(concat)
pool_rnn = Lambda(lambda x: backend.max(x, axis=1),
output_shape=(100, ))(semantic)
output = Dense(6, activation='sigmoid')(pool_rnn)
model = Model(inputs=[left_context, center, right_context], outputs=output)
roc_auc_callback = RocCallback(
[_left_context_train, _X_train, _right_context_train], _y_train,
[_left_context_valid, _X_valid, _right_context_valid], _y_valid)
early_stopping = EarlyStopping(monitor='val_loss', patience=5)
model_save_path = './models/text_cnn_non_static_' + str(
indice_fold) + '.h5'
model_check_point = ModelCheckpoint(model_save_path,
save_best_only=True,
save_weights_only=True)
tb_callback = TensorBoard('./logs', write_graph=True, write_images=True)
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
hist = model.fit(
[_left_context_train, _X_train, _right_context_train],
_y_train,
batch_size=BATCH_SIZE,
EMBEDDING_SIZE,
input_length=MAX_LEN,
weights=[embedding_matrix],
trainable=True))
model.add(Convolution1D(256, 3, padding='same', strides=1))
model.add(Activation('relu'))
model.add(MaxPooling1D(pool_size=2))
model.add(
GRU(256, dropout=0.2, recurrent_dropout=0.1, return_sequences=True))
model.add(GRU(256, dropout=0.2, recurrent_dropout=0.1))
# model.add(Dense(128,activation='relu'))
# model.add(Dropout(0.2))
# model.add(BatchNormalization())
model.add(Dense(6, activation='sigmoid'))
roc_auc_callback = RocCallback(_X_train, _y_train, _X_valid, _y_valid)
early_stopping = EarlyStopping(monitor='val_loss', patience=5)
model_save_path = './models/clstm_non_static_' + str(indice_fold) + '.h5'
model_check_point = ModelCheckpoint(model_save_path,
save_best_only=True,
save_weights_only=True)
tb_callback = TensorBoard('./logs', write_graph=True, write_images=True)
model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['accuracy'])
hist = model.fit(_X_train,
_y_train,
batch_size=BATCH_SIZE,
epochs=num_epoch,
validation_data=(_X_valid, _y_valid),
class_weight=class_weight,
# concatenate
merge = concatenate([cnn1, cnn2, cnn3])
merge = Flatten()(merge)
merge = Dropout(0.5)(merge)
merge = concatenate([merge, _statics_input])
merge = BatchNormalization()(merge)
merge = GaussianNoise(0.1)(merge)
merge = Dense(512, activation='relu')(merge) # linear layer
merge = Dropout(0.4)(merge)
merge = BatchNormalization()(merge)
out = Dense(6, activation='sigmoid')(merge)
model = Model(inputs=[_input, _statics_input], outputs=out)
roc_auc_callback = RocCallback([_X_train, _statics_train], _y_train,
[_X_valid, _statics_valid], _y_valid)
early_stopping = EarlyStopping(monitor='val_loss', patience=5)
model_save_path = './models/text_cnn_non_static_' + str(
indice_fold) + '.h5'
model_check_point = ModelCheckpoint(model_save_path,
save_best_only=True,
save_weights_only=True)
tb_callback = TensorBoard('./logs', write_graph=True, write_images=True)
model.compile(loss='binary_crossentropy',
optimizer='nadam',
metrics=['accuracy'])
hist = model.fit([_X_train, _statics_train],
_y_train,
batch_size=BATCH_SIZE,
epochs=num_epoch,
validation_data=([_X_valid, _statics_valid], _y_valid),