Would anyone know if it’s feasible to have sparse transition matrices as part of tfp’s Hidden Markov Model implementation? I had also created a FR Feature request / implementation guidance: Sparse HMM transition matrices · Issue #1585 · tensorflow/probability · GitHub but received no replies on it.
I’ve tried to feed in LinearOperators and tf’s Sparse Tensors but it seems like deeper changes are needed for this - would anyone know how feasible such an effort would be?