search

drckf / paysage

Unsupervised learning and generative models in python/pytorch.
Other
119 stars 25 forks source link

Cleanup #106

Closed drckf closed 7 years ago

drckf commented 7 years ago

The goal of this PR is to start bringing the API of the TAP training methods closer to the API of the sampling based training methods. This is going to be necessary to use the ComputationGraph for both training methods. The results on this branch should be the same as on master -- all the changes are simply organizational.

There is still going to be more to do to consolidate the two approaches, but it is probably better to do things in a few steps.

d-rams commented 7 years ago

This looks great. It will be a good exercise for me to use this updated API to add in the triple-term which is missing from the quadratic part of the TAP expansion (it only appears with on deep models)