LoginSign Up
← Back to all sparks
C

censored

ANALYTICS
Velocity0.0

Parsnip engines for survival analysis and censored regression models in R

github.com/tidymodels/censored ↗

censored keeps survival models aligned with parsnip's shifting prediction contracts

survival-analysistidymodelsparsnipenginesr-stats
◆Current state
censored supplies survival-analysis engines to parsnip, and its release history is dominated by staying in step with the rest of tidymodels rather than shipping independent features. The 0.2.0 and 0.3.0 releases added engines and prediction types; everything since has been contract alignment — quantile prediction format, hardhat test adaptation, flexsurv preparation.
◆Where it's heading
The package is settling into a downstream role where parsnip and hardhat set the interface and censored implements it for censored regression. Breaking changes arrive from upstream, not from new ideas here. The substantive engine work — aorsf, flexsurvspline, glmnet multi_predict — is behind it, and recent cycles are thin.
◆Prediction
Expect the next releases to track further parsnip prediction-type changes rather than add engines; the entries do not show new survival methods in progress.

◆Recent moves

  1. 4mo ago

    censored 0.3.4

    A test-only adaptation for a hardhat release. Nothing changes for users, and it is a fair marker of how thin the recent cycle has been.

    View source ↗
  2. 1y ago

    Quantile prediction format follows new parsnip requirements

    A breaking output-format change imposed by parsnip rather than chosen here — the clearest illustration of censored's downstream position. Code reading quantile predictions needs updating.

    View source ↗
  3. 2y ago

    Survival probabilities at infinite evaluation times now computed

    Raising the survival dependency lets predictions at eval_time = Inf return real values and intervals instead of hardcoded 0 and NA, across proportional_hazards, bag_tree and the partykit engines.

    View source ↗
  4. 2y ago

    censored 0.3.1

    Internal predict() changes staged ahead of a flexsurv release, with no user-visible effect.

    View source ↗
  5. 2y ago

    multi_predict() for all glmnet prediction types; aorsf predicts time

    Fills in the prediction matrix for penalized proportional hazards and lets oblique random survival forests predict survival time. Also breaks the survival_*/hazard_* helpers over to parsnip model_fit inputs.

    View source ↗
  6. 3y ago

    eval_time replaces time; matrix fitting for censored regression

    The interface release for this window: fit_xy() starts working for censored regression, eval_time supersedes time, and the aorsf and flexsurvspline engines arrive. Most of what came after is upkeep on these foundations.

    View source ↗