Hi @nazarblch
As part of a more general refactoring with @michalk8 , in #660, we're removing the corr option in dual potentials. This is no longer needed as it can be handled directly when instantiating the dual callable functions for a SqEuclidean cost, saving the trouble on adding back norms when computing W2 distance, etc
While looking over ENOT, we noticed that we cannot run the notebook using bidirectional=False so we're removing that bit as well (we're not sure it's mentioned in the experiments of the paper either).
If you happen to have the time, it would be also great if you could add simple tests for ENOT, thanks!
Hi @nazarblch
As part of a more general refactoring with @michalk8 , in #660, we're removing the
corroption in dual potentials. This is no longer needed as it can be handled directly when instantiating the dual callable functions for aSqEuclideancost, saving the trouble on adding back norms when computing W2 distance, etcWhile looking over ENOT, we noticed that we cannot run the notebook using
bidirectional=Falseso we're removing that bit as well (we're not sure it's mentioned in the experiments of the paper either).If you happen to have the time, it would be also great if you could add simple tests for ENOT, thanks!