LSTM-DNN network Architecture
hiran.park
def build_lstm_dnn_model(lstm_input_shape, dnn_input_shape):
# LSTM branch
lstm_input = Input(shape=lstm_input_shape, name="lstm_input")
lstm_out = LSTM(64, return_sequences=True)(lstm_input)
lstm_out = LSTM(64)(lstm_out)
lstm_out = Dropout(0.3)(lstm_out)
# DNN branch
dnn_input = Input(shape=dnn_input_shape, name="dnn_input")
dnn_out = Dense(128, activation='relu')(dnn_input)
dnn_out = Dropout(0.3)(dnn_out)
dnn_out = Dense(64, activation='relu')(dnn_out)
# Combine LSTM and DNN branches
combined = Concatenate()([lstm_out, dnn_out])
combined_out = Dense(32, activation='relu')(combined)
final_output = Dense(1, activation='sigmoid', name='output')(combined_out)
# Define the model
model = Model(inputs=[lstm_input, dnn_input], outputs=final_output)
model.compile(optimizer='adam', loss='binary_crossentropy', metrics=['accuracy'])
return model
I want to know network architecture and explain please
Samer Attrah
I would like to answer your question, this model is a two-branch model one is made of LSTM layers and the second is made of Dense layers. and in each branch, there is a single dropout layer.
and then the two layers are concatenated and the single vector is inputted into two dense layers.
Please check the attachment and if you need more details let me know the specific part that you are looking to understand.
Kind regards
Denis Dal Soler
- the single lstm units inside the first lstm(64) block.
- the two first lstm blocks (c(t) and h(t))
- an ipotetic vector input (x1,x2,...., xn) with the first lstm block.
--
You received this message because you are subscribed to the Google Groups "Keras-users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to keras-users...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/keras-users/68b79029-4848-4963-8a90-f18693d2872dn%40googlegroups.com.
