LSTM-> Globalpooling->RnnOutputLayer

Tuning Help
1

Hello,

Trying to build build architecture which looks something like :
Lstm-> Globalmaxpooling-> RnnOutputLayer

dimenisonality:
[N,T,D]->[N,D]-> [N,T,D]

As I have multiple labels per sequences using RnnOutputLayer as o/p.

Understandably, I get error “Input is not rank 3”… Got input with rank 2.

How do i resolve this issue.
Appreciate, if you can point me to code that uses using lstm->globalpooling->RnnOutLayer

Thanks
Ravi

2

How do you expect global pooling to work here? If it pools across your time dimensions, they are gone.