[MRG] Fix barycenter_stabilized with PyTorch and log set to True (#474)

* np -> nx for stabilized barycenters log

* Mention fix in RELEASES

Author: kachayev

RELEASES.md

@@ -15,6 +15,7 @@
 - Fix gradients for "Wasserstein2 Minibatch GAN" example (PR #466)
 - Faster Bures-Wasserstein distance with NumPy backend (PR #468)
 - Fix issue backend for ot.sliced_wasserstein_sphere ot.sliced_wasserstein_sphere_unif (PR #471)
+- Fix issue with ot.barycenter_stabilized when used with PyTorch tensors and log=True (RP #474)
 
 ## 0.9.0
 

ot/bregman.py

@@ -1898,8 +1898,8 @@ def barycenter_stabilized(A, M, reg, tau=1e10, weights=None, numItermax=1000,
                           "Or a larger absorption threshold `tau`.")
     if log:
         log['niter'] = ii
-        log['logu'] = np.log(u + 1e-16)
-        log['logv'] = np.log(v + 1e-16)
+        log['logu'] = nx.log(u + 1e-16)
+        log['logv'] = nx.log(v + 1e-16)
         return q, log
     else:
         return q