I want to use 2-3 LSTM networks as feature selectors, extract the last time steps of LSTM networks as the input layer, and then add several density layers to build a computation graph. How can I use a computation graph to build it?
@cqiaoYc have you tried anything so far? Normally that should just be the LastTimeStepVertex after each LSTM followed by a potenial reshape in to dense from the sounds of it. Have you tried just building 1 part of this at a time?
the LastTimeStepVertex works fine, thank you!