Update One_Sinkhorn.ipynb

[pkassraie]

I think the description for calculating the default epsilon in this tutorial is inaccurate. The tutorial says that the algorithm sets the default $epsilon$ as "the mean distance described in the geometry".
However, looking into the code it seems to me that you're calculating the average of the cost matrix instead. For $\ell_2$ loss, this is twice the distance squared. For more complicated losses, this does not even have to be measure of distance.
I suggest we change this loss to "the mean cost described in the geometry" to avoid confusion.

[review-notebook-app]

Check out this pull request on 

See visual diffs & provide feedback on Jupyter Notebooks.

---

Powered by ReviewNB

[michalk8]

Thanks, LGTM! Also seems like 1 test is failing, for unrelated reason. Will fix in a separate PR.