Missing weights in unbalanced.barycenter_unbalanced

[KonradStreit]

When calling the unbalanced barycenter function without going straight to the right method, the weights do not get passed on in the if-cases lines 1003 and following in unbalanced.py.

I found this by running the example script plot_UOT_barycenter_1D.py and changing the function call in there as mentioned in another issue raised earlier, changing

weights

to

weights=weights

. The interpolation supposed to happen only generates the step with the weights 0.5 and 0.5 as per default in unbalanced.barycenter_unbalanced_sinkhorn. The following image shows the result of the interpolation.

By adding

weights=weights

into the calls in the if-cases at the bottom of unbalanced.py the interpolation works, although there are some numerical problems then, which I do not understand OT well enough yet to judge. The following plot shows all steps from the third, which is the first without any nan in it.

Now it is showing changes in the distributions from one interpolation step to the next.

My system is
Windows-10-10.0.18362-SP0
Python 3.7.6 (default, Jan 8 2020, 20:23:39) [MSC v.1916 64 bit (AMD64)]
NumPy 1.18.1
SciPy 1.4.1
POT 0.6.0

[rflamary]

Hello @KonradStreit and thank you for the bug report.

This has been fixed in master to the best of our knowledge with PR #106. I agree the documentation and example should be update. I will probably do a minor release so that the bug is corrected for everyone.

[KonradStreit]

Ah alright I'm sorry, I thought it might be a different problem as it still was in there. Thanks for the answer!