lstm.py

def direction_model():
    model = Sequential(name = 'RNNStocks')
    model.add(Embedding(input_dim = 185, output_dim = 256,batch_input_shape=[None, None],
                        mask_zero = True, name ='EmbedLayer'))
    model.add(Bidirectional(LSTM(1024,
                                 return_sequences=False,stateful=False, 
                                 recurrent_initializer='glorot_uniform'), merge_mode ='concat',name = 'BiLSTM'))
    #final state encodes full representation of a single passed headine
    model.add(BatchNormalization(name='BatchNormal')) #After RNN(S-shape activation-f(x) / Before ReLU(Non-Gaussian))
#     model.add(tf.keras.layers.Masking(mask_value=0))
    model.add(Dense(512, name = 'FullConnected', kernel_initializer='he_normal')) 
    model.add(tf.keras.layers.LeakyReLU()) #controls vanishing gradients:f(x) =  a * (exp(x) - 1.) for x < 0 ; f(x) = x for x >= 0
    model.add(BatchNormalization(name='BatchNormal2'))
    model.add(Dense(1, activation='sigmoid',name='Output'))
    model.compile(optimizer=tf.optimizers.Adadelta(learning_rate = 1e-04), loss=tf.keras.losses.BinaryCrossentropy(from_logits=True),
              metrics=['accuracy',  tf.keras.metrics.AUC(name='AUC')])
    return model