keras_model.py

def get_model(features,clipvalue=1.,num_filters=40,dropout=0.1,embed_size=501):
    features_input = Input(shape=(features.shape[1],))
    inp = Input(shape=(maxlen, ))

    # Layer 1: Word2Vec Embeddings.
    x = Embedding(max_features, embed_size, weights=[embedding_matrix], trainable=False)(inp)

    # Layer 2: SpatialDropout1D(0.1)
    x = SpatialDropout1D(dropout)(x)

    # Layer 3: Bidirectional CuDNNLSTM
    x = Bidirectional(LSTM(num_filters, return_sequences=True))(x)

    # Layer 4: Bidirectional CuDNNGRU
    x, x_h, x_c = Bidirectional(GRU(num_filters, return_sequences=True, return_state = True))(x)  

    # Layer 5: some pooling operations
    avg_pool = GlobalAveragePooling1D()(x)
    max_pool = GlobalMaxPooling1D()(x)

    # Layer 6: A concatenation of the last state, maximum pool, average pool and 
    # additional features
    x = concatenate([avg_pool, x_h, max_pool,features_input])

    # Layer 7: A dense layer
    x = Dense(16, activation="relu")(x)

    # Layer 8: A dropout layer
    x = Dropout(0.1)(x)

    # Layer 9: Output dense layer with one output for our Binary Classification problem.
    outp = Dense(1, activation="sigmoid")(x)

    # Some keras model creation and compiling
    model = Model(inputs=[inp,features_input], outputs=outp)
    adam = optimizers.adam(clipvalue=clipvalue)
    model.compile(loss='binary_crossentropy',
                  optimizer=adam,
                  metrics=['accuracy'])
    return model