[WIP] Add new features to unbalanced solvers by 6Ulm · Pull Request #551 · PythonOT/POT

Types of changes

  • Allow warmstart for sinkhorn_unbalanced.

  • Allow choices of entropic regularizer in sinkhorn_unbalanced: KL divergence and negative entropy.

  • Allow for reference measure (so more flexibility) in MM and L-BFGS-B solvers.

  • Fix error in the calculation of negative entropy in L-BFGS-B solver.

Motivation and context / Related issue

  • Unlike sinkhorn solver for entropic (balanced) OT, the current sinkhorn_unbalanced does not allow for warmstart of dual vectors.

  • The current sinkhorn_unbalanced corresponds to the negative entropy regularizer, but we need KL divergence for the fused unbalanced GW.

  • The fused unbalanced GW requires more flexible MM and L-BFGS-B solvers.

How has this been tested (if it applies)

  • Tested choices of entropic regularizer in sinkhorn_unbalanced: KL divergence and negative entropy.

  • Tested reference measure in MM and L-BFGS-B solvers.

PR checklist

  • I have read the CONTRIBUTING document.
  • The documentation is up-to-date with the changes I made (check build artifacts).
  • All tests passed, and additional code has been covered with new tests.
  • I have added the PR and Issue fix to the RELEASES.md file.