[MRG] Adds Unbalaced transport to domain adaptation methods + bugfixes #100
Conversation
…ll bug related to warnings in unbalaced.py . Added an error message when user wants to normalize with other than expected cost normalization functions.
There was a problem hiding this comment.
Nice. The reg_e and reg_m are perhaps more adequate (than reg and alpha) with the naming convention of pot.
Can you please duplicate this test for your UnbalancedSinkhorn class ?
https://github.com/rflamary/POT/blob/b2157e9b3458388571f6ae87d80f47f500dfa166/test/test_da.py#L169
|
|
||
| Parameters | ||
| ---------- | ||
| reg_e : float, optional (default=1) |
There was a problem hiding this comment.
Perhaps the default reg_e should be smaller ? or None and set to 10 / dimension for e.g in fit
There was a problem hiding this comment.
I agree, but since by default norm = None, maybe we could leave it like that to avoid numerical errors in the default run?
There was a problem hiding this comment.
Actually I just realized that I have to check if the norm is None and do no normalization in that case, otherwise the error message will be triggered...
There was a problem hiding this comment.
Ok it is done and it seems ok
There was a problem hiding this comment.
Ok I just realized the other da classes use the same default
ot/da.py
Outdated
| """ | ||
|
|
||
| def __init__(self, reg_e=1., reg_m=0.1, method='sinkhorn', | ||
| max_iter=10, tol=10e-9, verbose=False, log=False, |
There was a problem hiding this comment.
oh yes. Actually the default value is 1e-9. Will fix.
|
Shoot, there are still some syntax error and my flake8 and pydocstyle are buggy, I have to find a way to fix them |
|
Aaah finally. We are green. |
rflamary
changed the title
|
Thank you @ngayraud, This PR looks very nice. I will have a quick look in the next days and will do the merge. |
This PR is primarily adds the unbalanced sinkhorn method to the domain adaptation classes. When the other sinkhorn methods will be implemented (eg stabilized) they will also be supported by the parameter method, which will now raise an error according to what is in unbalanced.py
In addition: