This a PR to check that our IPCW Brier score implementation behaves empirically as a proper scoring rule on a marginal model (no covariates).
This is similar to what we have in our example but only do marginal model estimation via the brier_score_* function on a grid of estimates.
This way we do not depend on any bias induced by underfitting caused by a bad tuning of the hyperparameters of a conditional model such as GradientBoostingIncidence.
This a PR to check that our IPCW Brier score implementation behaves empirically as a proper scoring rule on a marginal model (no covariates).
This is similar to what we have in our example but only do marginal model estimation via the
brier_score_*
function on a grid of estimates.This way we do not depend on any bias induced by underfitting caused by a bad tuning of the hyperparameters of a conditional model such as
GradientBoostingIncidence
.