search

automl / TabPFN

Official implementation of the TabPFN paper (https://arxiv.org/abs/2207.01848) and the tabpfn package.
http://priorlabs.ai
Apache License 2.0
1.22k stars 109 forks source link

Computation of gradient with respect to the input #30

Closed schnurrd closed 1 year ago

schnurrd commented 1 year ago

Change Description

I added the functionality to compute a gradient with respect to X_train and X_test. Thereby, the gradient can only be computed if X_train and X_test do not contain nans.

Standard Qs (please answer all)

Did all tests pass? Yes, they passed on my Pc.

Do you break behavior, if so why and did you communicate it? No behavior should be broken.

If you did add dependencies, did you add them to requirements.txt? No new dependencies were added.

Did you ping someone to review your PR? Yes, Samuel

Note: I think it would be a good idea to squash and merge, because I made many small commits.

cyrilpic commented 1 year ago

Hey, I just want to jump into the conversation and say that I have tested the code. I am able to use the gradient information and did not see any change in performance. Is there anything missing to merge it into the main branch? I'd be happy to help.

SamuelGabriel commented 1 year ago

Merged and part of the pip package now :)