Add kdi transform to deps.

[priorphil]

No description provided.

[gemini-code-assist]

Code Review

This pull request adds kditransform as a project dependency. The change is straightforward, but I've left a comment regarding sorting dependencies for better maintainability.

Additionally, with kditransform becoming a required dependency, the fallback logic in src/tabpfn/preprocessors/kdi_transformer.py (which handles cases where kditransform is not installed) appears to be redundant. You may want to consider removing the try-except block around the import in a future pull request to simplify the code.

[chatgpt-codex-connector]

💡 Codex Review

Here are some automated review suggestions for this pull request.

ℹ️ About Codex in GitHub

Codex has been enabled to automatically review pull requests in this repo. Reviews are triggered when you

• Open a pull request for review
• Mark a draft as ready
• Comment "@codex review".

If Codex has suggestions, it will comment; otherwise it will react with 👍.

When you sign up for Codex through ChatGPT, Codex can also answer questions or update the PR, like "@codex address that feedback".

[bejaeger]

Thanks!

[tankylz]

Hello there, I want to check the purpose of the new dependency kditransform. This is because I was using the older versions which didn't have this as a required dependency. Reinstalling the packages now lead to dependencies conflict with other packages. Thanks in advanced for the clarifications!

[priorphil]

We found that using the KDI transform for some pre-processors improves predictive performance.
Could you give more details about the package conflict? Do you have other deps that depend on a different version of the KDI transform, do the KDI dependencies cause the mismatch, or something else?

[tankylz]

The conflict comes from sklearn actually. We are using tabpfgen as well (which I understand is from a different group).

Running on Linux:

tabpfn 2.2.1 depends on scikit-learn<1.8 and >=1.2.0
tabpfgen 0.1.4 depends on scikit-learn==1.5.2
kditransform 1.2.0 depends on scikit-learn>=1.6.0