conv + lstm model

hybrid_model.py

def build_Model(num_classes, seq_lenght, input_shape=(28, 252, 1)):
    inputs = Input(name='x', shape=input_shape, dtype='float32')

    conv1 = Conv2D(seq_lenght, (3, 3), padding='same', name='conv1', kernel_initializer='he_normal')(inputs)
    conv1 = Activation('relu')(conv1)
    conv1 = MaxPooling2D(pool_size=(2, 2), name='max1')(conv1)
    
    dims = conv1.get_shape()
    
    reshape = Reshape(target_shape=(seq_lenght, int(dims[1]*dims[2])), name='reshape')(conv1)
    reshape = Dense(128, activation='relu', kernel_initializer='he_normal', name='dense2')(reshape)
    
    lstm_1 = LSTM(32, return_sequences=True, kernel_initializer='he_normal', name='lstm1')(reshape)
    lstm_2 = LSTM(32, return_sequences=True, kernel_initializer='he_normal', name='lstm2')(lstm_1)
    
    y_pred = Dense(num_classes, activation='softmax',
                   kernel_initializer='he_normal',name='output')(lstm_2)

    model = Model(inputs=inputs, outputs=y_pred)
    model.compile(Adam(lr=0.001), 'categorical_crossentropy', metrics=['accuracy'])

    return model