Merge branch 'master' into test/data-functions-edge-cases

Author: jgabry

NEWS.md

@@ -1,6 +1,8 @@
 # bayesplot (development version)
 
 * Added singleton-dimension edge-case tests for exported `_data()` functions.
+* Validate empty list and zero-row matrix inputs in `nuts_params.list()`.
+* Validate user-provided `pit` values in `ppc_loo_pit_data()` and `ppc_loo_pit_qq()`, rejecting non-numeric inputs, missing values, and values outside `[0, 1]`.
 * New `show_marginal` argument to `ppd_*()` functions to show the PPD - the marginal predictive distribution by @mattansb (#425)
 * `ppc_ecdf_overlay()`, `ppc_ecdf_overlay_grouped()`, and `ppd_ecdf_overlay()` now always use `geom_step()`. The `discrete` argument is deprecated.
 * Fixed missing `drop = FALSE` in `nuts_params.CmdStanMCMC()`.

R/bayesplot-extractors.R

@@ -145,10 +145,18 @@ nuts_params.stanreg <-
 #' @export
 #' @method nuts_params list
 nuts_params.list <- function(object, pars = NULL, ...) {
+  if (length(object) == 0) {
+    abort("'object' must be a non-empty list.")
+  }
+
   if (!all(sapply(object, is.matrix))) {
     abort("All list elements should be matrices.")
   }
 
+  if (any(vapply(object, nrow, integer(1)) == 0)) {
+    abort("All matrices in the list must have at least one row.")
+  }
+
   dd <- lapply(object, dim)
   if (length(unique(dd)) != 1) {
     abort("All matrices in the list must have the same dimensions.")

R/helpers-ppc.R

@@ -58,15 +58,20 @@ validate_y <- function(y) {
 #' Validate predictions (`yrep` or `ypred`)
 #'
 #' Checks that `predictions` is a numeric matrix, doesn't have any NAs, and has
-#' the correct number of columns.
+#' the correct number of columns. If `predictions` is a `posterior::draws`
+#' object it is first coerced to a matrix.
 #'
-#' @param predictions The user's `yrep` or `ypred` object (SxN matrix).
+#' @param predictions The user's `yrep` or `ypred` object (SxN matrix or a
+#'   `posterior::draws` object).
 #' @param `n_obs` The number of observations (columns) that `predictions` should
 #'   have, if applicable.
 #' @return Either throws an error or returns a numeric matrix.
 #' @noRd
 validate_predictions <- function(predictions, n_obs = NULL) {
-  # sanity checks
+  if (posterior::is_draws(predictions)) {
+    predictions <- posterior::as_draws_matrix(predictions)
+    predictions <- unclass(predictions)
+  }
   stopifnot(is.matrix(predictions), is.numeric(predictions))
   if (!is.null(n_obs)) {
     stopifnot(length(n_obs) == 1, n_obs == as.integer(n_obs))

R/ppc-loo.R

@@ -302,7 +302,10 @@ ppc_loo_pit_data <-
            boundary_correction = TRUE,
            grid_len = 512) {
     if (!is.null(pit)) {
-      stopifnot(is.numeric(pit), is_vector_or_1Darray(pit))
+      pit <- validate_pit(pit)
+      if (boundary_correction && length(pit) < 2L) {
+        abort("At least 2 PIT values are required when 'boundary_correction' is TRUE.")
+      }
       inform("'pit' specified so ignoring 'y','yrep','lw' if specified.")
     } else {
       suggested_package("rstantools")
@@ -348,7 +351,7 @@ ppc_loo_pit_qq <- function(y,
 
   compare <- match.arg(compare)
   if (!is.null(pit)) {
-    stopifnot(is.numeric(pit), is_vector_or_1Darray(pit))
+    pit <- validate_pit(pit)
     inform("'pit' specified so ignoring 'y','yrep','lw' if specified.")
   } else {
     suggested_package("rstantools")
@@ -795,14 +798,6 @@ ppc_loo_ribbon <-
   # Generate boundary corrected values via a linear convolution using a
   # 1-D Gaussian window filter. This method uses the "reflection method"
   # to estimate these pvalues and helps speed up the code
-  if (any(is.infinite(x))) {
-    warn(paste(
-      "Ignored", sum(is.infinite(x)),
-      "Non-finite PIT values are invalid for KDE boundary correction method"
-    ))
-    x <- x[is.finite(x)]
-  }
-
   if (grid_len < 100) {
     grid_len <- 100
   }
@@ -819,6 +814,10 @@ ppc_loo_ribbon <-
   # 1-D Convolution
   bc_pvals <- .linear_convolution(x, bw, grid_counts, grid_breaks, grid_len)
 
+  if (all(is.na(bc_pvals))) {
+    abort("KDE boundary correction produced all NA values.")
+  }
+
   # Generate vector of x-axis values for plotting based on binned relative freqs
   n_breaks <- length(grid_breaks)
 

man-roxygen/args-y-yrep.R

@@ -1,8 +1,9 @@
 #' @param y A vector of observations. See **Details**.
 #' @param yrep An `S` by `N` matrix of draws from the posterior (or prior)
-#'   predictive distribution. The number of rows, `S`, is the size of the
-#'   posterior (or prior) sample used to generate `yrep`. The number of columns,
-#'   `N` is the number of predicted observations (`length(y)`). The columns of
-#'   `yrep` should be in the same order as the data points in `y` for the plots
-#'   to make sense. See the **Details** and **Plot Descriptions** sections for
-#'   additional advice specific to particular plots.
+#'   predictive distribution, or a [`posterior::draws`] object. The number of
+#'   rows, `S`, is the size of the posterior (or prior) sample used to generate
+#'   `yrep`. The number of columns, `N` is the number of predicted observations
+#'   (`length(y)`). The columns of `yrep` should be in the same order as the
+#'   data points in `y` for the plots to make sense. See the **Details** and
+#'   **Plot Descriptions** sections for additional advice specific to particular
+#'   plots.

man-roxygen/args-ypred.R

@@ -1,4 +1,5 @@
 #' @param ypred An `S` by `N` matrix of draws from the posterior (or prior)
-#'   predictive distribution. The number of rows, `S`, is the size of the
-#'   posterior (or prior) sample used to generate `ypred`. The number of
-#'   columns, `N`, is the number of predicted observations.
+#'   predictive distribution, or a [`posterior::draws`] object. The number of
+#'   rows, `S`, is the size of the posterior (or prior) sample used to generate
+#'   `ypred`. The number of columns, `N`, is the number of predicted
+#'   observations.