Password for registration / validation link: tabpfn-2023
This is an alpha family and friends service, so please do not expect this to never be down or run into errors. We did test it though and can say that it seems to work fine in the settings that we tried.
PLEASE DO NOT SHARE THIS REPOSITORY at this point outside of the NeurIPS Tabular Representation workshop.
What model is behind the API? For now, this version is the light version to save compute on our side, not the TabPFN (Fast) or TabPFN (Best-Q) which we presented at the Neurips Workshop. We will change this once we see our server is working stably. It is a new TabPFN which we allow to handle up to 10K instances with up to 500 features. This TabPFN is not ensembled, we will put out improved and ensembled models soon.
We would really appreciate your feedback! If you encounter bugs or suggestions for improvement please create an issue or email me (samuelgabrielmuller (at) gmail com).
We created a colab tutorial to get started quickly.
pip install tabpfn-client
Import and login
from tabpfn_client import init, TabPFNClassifier
init()
Now you can use our model just like any other sklearn estimator
tabpfn = TabPFNClassifier()
tabpfn.fit(X_train, y_train)
tabpfn.predict(X_test)
# or you can also use tabpfn.predict_proba(X_test)
To encourage better coding practices, ruff
has been added to the pre-commit hooks. This will ensure that the code is formatted properly before being committed. To enable pre-commit (if you haven't), run the following command:
pre-commit install
Additionally, it is recommended that developers install the ruff extension in their preferred editor. For installation instructions, refer to the Ruff Integrations Documentation.