Quantile sets from several quantile boosters

Each booster_fit() with family = "quantile" predicts its own alpha quantile independently, so the predictions can cross (a 90% quantile below the 50% one on some row). Each row's predictions are sorted across the levels, which never makes any of them worse in pinball loss (Chernozhukov, Fernandez-Val and Galichon, 2010).

Usage

predict_quantiles(models, newdata)

Arguments

models

A list of quantile booster_models with distinct alpha.

newdata

A data frame with the models' terms.

Value

A matrix with one row per row of newdata and one column per level, in increasing alpha, named by the level.