Question about LSTM NN design
I have a Deep Learning / Keras NN design question: I want to classify a time series dataset by implementing an LSTM design. My classification has a hierarchy. A pattern belongs to multiple classes. I will label each class as a separate column in the training data set. Class 1 classification results from an LSTM could be used as input in an adjacent LSTM to improve class 2 classification. My question is: how should I design the NN? 2 Separate NNs and merge the output on basis of the timestamp? Does a better design exist in such LSTM application? Please advice.