[MRG] Add implicit Sinkhorn gradients #605
Merged
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Types of changes
This PR aims at
This PR should solve #565 and greatly limit memory for sinkhorn when computing gradienst wrt the value.
In order to use implicit diffeerntiation one needs to set the
grad
parameter inot.solve
andot.solve_sample
as suchOn a simple example with pytorch arrays with required gradients, I has a 1000x gain in memory for solving the problem when a large number of sinkhorn operations are needed.
Motivation and context / Related issue
How has this been tested (if it applies)
PR checklist