tabpfn-explore

What problem does it solve?

Enables rapid, systematic exploration and validation of tabular competition datasets so you can detect leakage, distribution shift, and prepare consistent cross-validation splits before any modeling or API calls.

Core Features & Use Cases

• Competition reconnaissance: document evaluation metric, task type, dataset size, and known pitfalls to guide modeling choices.
• Exploratory Data Analysis: profile missingness, class balance, high-cardinality categoricals, duplicates, and high correlations that affect model design.
• Adversarial validation: run train-vs-test classifiers to quantify distribution shift and surface the features driving it.
• CV scheme and budget checks: define and save reproducible folds (StratifiedKFold, GroupKFold, TimeSeriesSplit) and verify TabPFN API cell budget constraints.
• Deliverables: cleaned X_train/X_test/y_train DataFrames, a reusable folds object, notes/competition_overview.md, and an issues checklist for leakage, imbalance, and high-cardinality features.

Quick Start

Run tabpfn-explore on your train and test CSVs to generate cleaned DataFrames, a saved folds object, adversarial validation diagnostics, and an API cell budget check.