tabpfn-regress

To generate out-of-fold predictions and a submission for tabular regression, you need a cross-validated baseline pipeline. This Skill executes a TabPFN v2.5 baseline, producing OOF predictions and an initial submission file for leaderboard comparison after data preparation and fold generation are complete.

Yes, you can ensemble TabPFN with gradient-boosting trees for Kaggle tabular tasks. This Skill applies ensembling and post-processing rules, including GBT ensembles, clipping, target transforms, and rank blending, to optimize the baseline predictions into a final submission.

Before running a TabPFN regression baseline, you need clean training features (X_train), target values (y_train), test features (X_test), and saved cross-validation folds. Data preparation and fold generation must be completed beforehand, and you need access to the TabPFN client with API budget checks.

Post-processing with rank blending and clipping improves tabular regression submissions by combining multiple model outputs and constraining prediction values. These techniques, along with target transforms, refine ensemble predictions to reduce variance and align outputs with leaderboard target distributions.

TabPFN regression handles feature preprocessing and API budget limits as part of its execution workflow. The Skill applies proper feature preprocessing to the provided clean datasets and performs API budget checks before accessing the TabPFN client to generate cross-validated predictions.

TabPFN cross-validation folds need to be saved before generating OOF predictions to ensure reproducible baseline results. The Skill requires predefined folds generated during the data exploration phase, applying them consistently to produce comparable out-of-fold predictions and submission logs for leaderboard tracking.