def define_network()

in train/CustomModel.py [0:0]

• 16 lines of code
• 2 McCabe index (conditional complexity)

def define_network(embedding_layer):
    '''
    Define LSTM network with an attention layer
    '''
    sequence_input = tf.keras.Input(shape=(MAX_SEQUENCE_LENGTH,), dtype='int32')
    ## If pretrained embedding layer is not given, train your own
    if embedding_layer == "none":
        embedded_sequences = tf.keras.layers.Embedding(MAX_NB_WORDS,EMBEDDING_DIM,input_length=MAX_SEQUENCE_LENGTH)(sequence_input)
    else:
        embedded_sequences = embedding_layer(sequence_input)
    lstm = Bidirectional(LSTM(100,dropout = 0.2, recurrent_dropout = 0.2,return_sequences=True))(embedded_sequences)
    lstm = LayerNormalization()(lstm)
    attentionlstm = attention(return_sequences=False,activation='tanh')(lstm)
    s = Dense(6,activation='sigmoid')(attentionlstm)
    model_LSTM = tf.keras.Model(inputs=[sequence_input],outputs=[s])
    print(model_LSTM.summary())
    return model_LSTM