docs+fix: point_and_distn

NAMESPACE

@@ -132,6 +132,7 @@ export(layer)
 export(layer_add_forecast_date)
 export(layer_add_target_date)
 export(layer_naomit)
+export(layer_point_and_distn)
 export(layer_point_from_distn)
 export(layer_population_scaling)
 export(layer_predict)

R/layer_point_and_distn.R

@@ -1,7 +1,31 @@
-#' returns both the point estimate and the quantile distribution, regardless of the underlying trainer
-#' f
+#' returns both the point estimate and the quantile distribution
+#' @description
+#' This function adds a frosting layer that produces both a point estimate as
+#' well as quantile estimates.
+#' @param distn_id a random id string for the layer that creates the quantile
+#'        estimate
+#' @param point_id a random id string for the layer that creates the point
+#'        estimate. Only present for trainers that produce quantiles
+#' @param point_type character. Either `mean` or `median`.
+#' @param use_predictive_distribution only usable for `linear_reg` type models
+#' @param distn_type character. Only used if `use_predictive_distribution=TRUE`,
+#'               for `linear_reg` type models. Either gaussian or student_t
+#' @param distn_name an alternate name for the distribution column; defaults
+#'        to `.pred_distn`.
+#' @param point_name an alternate name for the point estimate column; defaults
+#'        to `.pred`.
+#' @param symmetrize logical. If `TRUE` then interval will be symmetric.
+#'        Applies for residual quantiles only
+#' @param by_key A character vector of keys to group the residuals by before
+#'   calculating quantiles. The default, `c()` performs no grouping. Only used
+#'   by `layer_residual_quantiles`
+#' @inheritParams layer_quantile_distn
+#' @export
+#' @return an updated `frosting postprocessor` with an additional prediction
+#'   column; if the trainer produces a point estimate, it has added a
+#'   distribution estimate, and vice versa.
 layer_point_and_distn <- function(frosting, trainer, ...,
-                                  probs = c(0.05, 0.95),
+                                  levels = c(0.25, 0.75),
                                   symmetrize = TRUE,
                                   by_key = character(0L),
                                   distn_name = ".pred_distn",
@@ -10,26 +34,27 @@ layer_point_and_distn <- function(frosting, trainer, ...,
                                   point_id = NULL,
                                   point_type = c("median", "mean"),
                                   truncate = c(-Inf, Inf),
-                                  use_predictive_distribution = TRUE,
+                                  use_predictive_distribution = FALSE,
                                   dist_type = "gaussian") {
   rlang::check_dots_empty()
   stopifnot(inherits(recipe, "recipe"))
-  # not sure what to do about the dots...
+  levels <- sort(levels)
   if (inherits(trainer, "quantile_reg")) {
+    # sort the probabilities
     tau <- sort(compare_quantile_args(
-      args_list$levels,
+      levels,
       rlang::eval_tidy(trainer$args$tau)
     ))
-    args_list$levels <- tau
+    levels <- tau
     trainer$args$tau <- rlang::enquo(tau)
     if (is.null(point_id)) {
       point_id <- rand_id("point_from_distn")
     }
     if (is.null(distn_id)) {
       distn_id <- rand_id("quantile_distn")
     }
-    frosting %<>% layer_quantile_distn(...,
-      levels = tau,
+    frosting %<>% layer_quantile_distn(
+      levels = levels,
       truncate = trucate,
       name = distn_name,
       id = distn_id
@@ -44,7 +69,7 @@ layer_point_and_distn <- function(frosting, trainer, ...,
       distn_id <- rand_id("residual_quantiles")
     }
     if (inherits(trainer, "linear_reg") && use_predictive_distribution) {
-      frosting %<>% layer_residual_quantiles(
+      frosting %<>% layer_predictive_distn(
         dist_type = dist_type,
         name = distn_name,
         id = distn_id