You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
This commit here seems to have erroneously modified the dragan penalty code from 'c' (which is a hyperparameter for the dragan penalty) to be 'pac'.
Reviewing that commit it seems as if there's something strange happening at the intersection of the pac methodology and the implementation of the dragan penalty. I'm a little mystified by the approach of transposing the array as a fix to make the tensors the right size.
For a pac=1 value this should be just fine (as is the default for PATECTGAN, which I'm curious about the motivations for?), but increasing pac to something else (e.g. 10, as ctgan) seems problematic as it there doesn't seem to be a good reason for the transpositions.
Am I missing something?
The text was updated successfully, but these errors were encountered:
ssabdb
changed the title
PATECTGAN: dragan_penalty and PAC properly
PATECTGAN: dragan_penalty and PAC property
Dec 6, 2022
I also ran into this error when investigating #489. It's likely a bug that was introduced in removing the dependency on the unsupported dev release of ctgan. I'm not as familiar with this code, so I have hesitated to fix it until I have time to really understand what's going on. Perhaps @lurosenb or @AprilXiaoyanLiu can remember.
This commit here seems to have erroneously modified the dragan penalty code from 'c' (which is a hyperparameter for the dragan penalty) to be 'pac'.
Reviewing that commit it seems as if there's something strange happening at the intersection of the pac methodology and the implementation of the dragan penalty. I'm a little mystified by the approach of transposing the array as a fix to make the tensors the right size.
For a pac=1 value this should be just fine (as is the default for PATECTGAN, which I'm curious about the motivations for?), but increasing pac to something else (e.g. 10, as ctgan) seems problematic as it there doesn't seem to be a good reason for the transpositions.
Am I missing something?
The text was updated successfully, but these errors were encountered: