Closed badisa closed 8 months ago
@badisa Thank you for your contribution! Even if we don't end up really using this function, it would be good for it to be correct!
@badisa Thank you for your contribution! Even if we don't end up really using this function, it would be good for it to be correct!
Looks like the tests failed due to being unable to get the OE license, let me know if I need to do anything to resolve that.
Perhaps because it is a fork?
Yah this is an issue with it being a fork, I will need to make this PR under a branch attached to the main repo, I can do that right now
This "tricks" github into running the tests here as well as https://github.com/choderalab/perses/pull/1239 since it is the same commit
@ijpulidos This PR has more context, do you mind reviewing this one?
The code is the same btw @ijpulidos
Doesn't look like
MultiTargetDesign
is used, so feel free to close this PR. Just happened to see the bug when reading the code.Description
Noticed that a custom implementation of logsumexp was incorrect, as it didn't add back in
a_n.max()
.Motivation and context
Incorrect implementation, will produce non-normalized log probabilities for the class
MultiTargetDesign
No associated case
How has this been tested?
Tested by the following python script
Change log