For learning a single sequence, LSTM only should suffice.
However, my situation is different here. I have a list of sequences to learn:
- The sale volumes of 12 months, these are the sequences
And each sequence above belongs to a category.
I'm trying it out by consider [category,sequence] as a sequential sample, the loss can be reduced to 1% but it gives wrong values in inferring real data.
The second try is considering [category,sequence] as a sample of 2 inputs:
- X1 = sequence
- X2 = category
Feed the sequence thru' LSTM layers to get H, and then concat with X2, and feed again the pair [H,X2] thru' some dense layers, the results aren't better.
Any popular solutions (network shape, network design) for learning this kind of data: sequential data in different categories?