reproducibility

[KamranNiroomand]

Describe the workflow you want to enable

Hello,

I am using the model, but the main problem that I have with it is that every time I run it, it gives me slightly different results. How can I make it reproducible and identical for every run?

thanks,

kamran

Describe your proposed solution

maybe kinda random state or something?

Describe alternatives you've considered, if relevant

No response

Additional context

No response

Impact

None

[LeoGrin]

TabPFNClassifier and TabPFNRegressor have a random_state parameter, and should be deterministic for a same random state. Is it not the case for you? In this case, could you please share the output of import tabpfn; tabpfn.display_debug_info() to help us solve the issue? Thanks a lot!

[noahho]

tests/test_classifier_interface.py::test_sklearn_compatible_estimator[...]-check_fit_idempotent (Test 211):
This test specifically checks if calling the fit method multiple times on the same data changes the resulting model. A deterministic classifier should produce an identical model, and therefore identical predictions, if fit is called repeatedly with the same data. This test PASSED.

This test checks consistency but it might not work on all systems (win, unix, macos tested) but not cuda