TabPFN-v2 in-context foundation-model discriminator (conditional single-sample)

Conditional-single-sample discriminator (family O) built on TabPFN-v2, the pretrained tabular foundation model of Hollmann et al. (Nature 2025). TabPFN classifies a query row by in-context learning: a single forward pass of a frozen pretrained transformer conditions on a provided labelled TRAINING context and emits the query's class posterior. There is no per-dataset gradient training – "fitting" only loads the training context into the model.

Why this is single-sample-deployable (conditional-in-N). A specimen is single-sample-classifiable iff the TRAINING context that conditions the transformer is FROZEN into the model at fit time, so that at inference the query's posterior depends only on that query row and the frozen context – never on which other specimens are scored alongside it. This implementation freezes the per-sample-rCLR-transformed training context (rows + labels) and the feature universe at fit, and at score conditions every query against that one frozen context. The frozen context is the TRAINING rows only, so it is leakage-safe (no scored / test data ever enters the context).

Row-independent scoring modes. TabPFN's predict_proba is invariant to the row ORDER of the scored batch and to the scored subset – a query's posterior does not depend on the other queries (measured companion maxdiff \(0\)). The default deployment path still evaluates predict_proba ONE QUERY ROW AT A TIME, so the \(n=1\) forward path is used uniformly. The optional benchmark-only score_batch path evaluates balanced chunks of query rows. That batched path is leakage-free and AUC-faithful (observed \(|dAUC| = 7.8\mathrm{e}{-5}\) at production scale; specimen ranks/verdicts unchanged), but it is not bit-identical to the \(n=1\) row-by-row path: the installed build has a deterministic batch-size numerical effect (raw probability shifts around \(1\mathrm{e}{-4}\) to \(2.6\mathrm{e}{-3}\)), not a cross-row coupling.

Single-sample transform. Each specimen is mapped to the self-contained per-sample robust CLR (rCLR) over its OWN strictly-positive support (geometric-mean centring on v > 0) in the FROZEN feature universe; this is exactly invariant to per-specimen scaling and uses no cross-row statistic. The SAME rCLR transform is applied to the frozen training context at fit and to each query at score. Universe features absent from a specimen carry the neutral rCLR value 0 (centred log-ratio of an absent coordinate), matching the missing-feature rule of the other roster scorers.

Ungated TabPFN-v2 weights. The default TabPFN model version in current tabpfn releases (V2.6) is license-gated and needs an interactive token. This method pins the model to ModelVersion.V2 – the original TabPFN-v2 (Hollmann et al., Nature 2025) – which ships with a direct (token-free) download option, so the method installs and runs non-interactively.

Orientation and output. The score is the frozen-context posterior \(P(\mathrm{case}) = P(y = 1)\) for the query, in \([0, 1]\); larger = more case-like. Degenerate queries (fewer than hp$min_features universe features present, or an all-zero / empty specimen) return the neutral probability 0.5.

References

Hollmann N, Muller S, Purucker L, Krishnakumar A, Korfer M, Hoo SB, Schirrmeister RT, Hutter F. (2025) Accurate predictions on small data with a tabular foundation model. Nature 637:319-326.

Hollmann N, Muller S, Eggensperger K, Hutter F. (2023) TabPFN: A Transformer That Solves Small Tabular Classification Problems in a Second. International Conference on Learning Representations (ICLR).