Fix Fast-Weights model (A computed every t)

twitter-archive / torch-autograd

Autograd automatically differentiates native Torch code

Apache License 2.0

560 stars 115 forks source link

Fix Fast-Weights model (A computed every t) #170

Closed Atcold closed 7 years ago

Atcold commented 7 years ago

This is a fix for the Fast Weights model. The matrix A needs to be computed at each step t. Batch mode is also now correctly implemented via nn.MM() (one A fast weights matrix per each representation in the batch).

Atcold commented 7 years ago

@codeAC29, please review this PR. It should be all good now.