Survival analysis

Installation

pip install "tabpfn-extensions[survival]" scikit-survival

`SurvivalTabPFN`

View source

Two-head TabPFN survival scorer:

classifier -> P(event | X)
distributional regressor -> T | (event, X)

Scalar risk options (higher = riskier):

“weighted_cdf” (default): sum_j w_j * CIF_1(tau_j | X), where CIF_1(t | X) = P(T <= t, event=1 | X).
“avg_cdf”: uniform average of CIF_1(tau_j | X) across horizons.
“p_over_mean”: P(event|X) / E[T | event, X] (legacy fallback).

Also exposes CIF-based curves for event=1:

predict_cif_at(X, t_grid) -> CIF_1(t|X) = P(T <= t, event=1 | X)
predict_survival_from_cif_at(X, t_grid) -> 1 - CIF_1(t|X)

Backwards-compatible aliases (deprecated):

predict_cdf_at(X, t_grid) -> same as predict_cif_at, returns CIF_1(t|X)
predict_survival_at(X, t_grid) -> same as predict_survival_from_cif_at

SurvivalTabPFN(
    *,
    cls_model: TabPFNClassifier | None = None,
    reg_model: TabPFNRegressor | None = None,
    risk_strategy: Literal["weighted_cdf", "avg_cdf", "p_over_mean"] = "weighted_cdf",
    default_horizons: Iterable[float] | None = None,
    n_auto_horizons: int = 6,
    auto_horizon_quantile_range: tuple[float, float] = (0.1, 0.9),
    exp_weight_gamma: float = 0.85,
    eps: float = 1e-09,
    random_state = None,
)

Parameters

Parameter	Type	Default	Description
`cls_model`	`TabPFNClassifier \| None`	`None`	If `None`, created in fit with `random_state`.
`reg_model`	`TabPFNRegressor \| None`	`None`	If `None`, created in fit with `random_state`.
`risk_strategy`	`Literal[“weighted_cdf”, “avg_cdf”, “p_over_mean”]`	`"weighted_cdf"`	—
`default_horizons`	`Iterable[float] \| None`	`None`	—
`n_auto_horizons`	`int`	`6`	—
`auto_horizon_quantile_range`	`tuple[float, float]`	`(0.1, 0.9)`	—
`exp_weight_gamma`	`float`	`0.85`	—
`eps`	`float`	`1e-09`	—
`random_state`	`any`	`None`	—

`SurvivalTabPFN.fit`

View source

`SurvivalTabPFN.predict`

View source

`SurvivalTabPFN.predict_cif_at`

View source

SurvivalTabPFN.predict_cif_at(
    X: np.ndarray,
    t_grid: Iterable[float],
) -> np.ndarray

Parameters

Parameter	Type	Default	Description
`X`	`np.ndarray`	Required	—
`t_grid`	`Iterable[float]`	Required	—

Returns

Type	Description
`np.ndarray`	—

`SurvivalTabPFN.predict_risk`

View source

`SurvivalTabPFN.predict_survival_from_cif_at`

View source

Complement 1 - CIF_1(t|X) based on predict_cif_at. i.e. the probability that event 1 has not yet occurred by time t.

SurvivalTabPFN.predict_survival_from_cif_at(
    X: np.ndarray,
    t_grid: Iterable[float],
) -> np.ndarray

Parameters

Parameter	Type	Default	Description
`X`	`np.ndarray`	Required	—
`t_grid`	`Iterable[float]`	Required	—

Returns

Type	Description
`np.ndarray`	1 - CIF_1(t\|`X`) = 1 - P(T <= t, event=1 \| `X`),

Reference

tabpfn

tabpfn-client

tabpfn-extensions

REST API

Installation

`SurvivalTabPFN`

`SurvivalTabPFN.fit`

`SurvivalTabPFN.predict`

`SurvivalTabPFN.predict_cif_at`

`SurvivalTabPFN.predict_risk`

`SurvivalTabPFN.predict_survival_from_cif_at`

​Installation

​SurvivalTabPFN

​SurvivalTabPFN.fit

​SurvivalTabPFN.predict

​SurvivalTabPFN.predict_cif_at

​SurvivalTabPFN.predict_risk

​SurvivalTabPFN.predict_survival_from_cif_at

Installation

`SurvivalTabPFN`

`SurvivalTabPFN.fit`

`SurvivalTabPFN.predict`

`SurvivalTabPFN.predict_cif_at`

`SurvivalTabPFN.predict_risk`

`SurvivalTabPFN.predict_survival_from_cif_at`