Added user option to adjust #nodes in Gaussian Quadrature (#272)

R/DataSurvival.R

@@ -125,7 +125,10 @@ as_stan_list.DataSurvival <- function(object, ...) {
     rownames(design_mat) <- NULL
 
     # Parameters for efficient integration of hazard function -> survival function
-    gh_parameters <- statmod::gauss.quad(n = 15, kind = "legendre")
+    gh_parameters <- statmod::gauss.quad(
+        n = getOption("jmpost.gauss_quad_n"),
+        kind = getOption("jmpost.gauss_quad_kind")
+    )
 
     model_data <- list(
         Nind_dead = sum(df[[vars$event]]),

R/settings.R

@@ -30,6 +30,21 @@
 #' Directory to store compiled stan models in. If not set, a temporary directory is used for
 #' the given R session. Can also be set via the environment variable `JMPOST_CACHE_DIR`.
 #'
+#'
+#'
+#' ## `jmpost.gauss_quad_n`
+#'
+#' Default = 15
+#'
+#' In most cases the survival function of the joint model does not have a closed form
+#' and as such it is calculated by integrating the hazard function. `jmpost` estimates this
+#' via Gaussian Quadrature, in particular it uses [`statmod::gauss.quad`] with
+#' `kind = "legendre"` to create the nodes and weights.
+#'
+#' This option specifies the `n` argument in the call to [`statmod::gauss.quad`]. In general
+#' higher values of `n` lead to better accuracy of the approximation but at the cost of
+#' increased computational time.
+#'
 #' @examples
 #' \dontrun{
 #' options(jmpost.prior_shrinkage = 0.5)
@@ -46,7 +61,8 @@ set_options <- function() {
     current_opts <- names(options())
     jmpost_opts <- list(
         jmpost.cache_dir = cache_dir,
-        jmpost.prior_shrinkage = 0.5
+        jmpost.prior_shrinkage = 0.5,
+        jmpost.gauss_quad_n = 15
     )
     for (opt in names(jmpost_opts)) {
         if (!opt %in% current_opts) {

design/interfaces.md

@@ -11,7 +11,7 @@ matrix link_contribution(matrix time, matrix pars_lm)
 ```
 
 - must return same dimensions as time
-- where time is 1 row per subject (Nind) and 1 column per required timepoint (nodes from the guasian quadrature) (i.e. if multiple timepoints need to be evaluated for each subject it will still be 1 row but with multiple columns, if only 1 timepoint needs to be evaluated then there will only be 1 column)
+- where time is 1 row per subject (Nind) and 1 column per required timepoint (nodes from the Gaussian quadrature) (i.e. if multiple timepoints need to be evaluated for each subject it will still be 1 row but with multiple columns, if only 1 timepoint needs to be evaluated then there will only be 1 column)
 - pars_lm is a matrix with 1 row per subject and as many columns as you need defined in transformed_parameters()
 
 

inst/WORDLIST

@@ -36,7 +36,7 @@ frac
 funder
 gi
 gl
-guasian
+Gaussian
 hardcoded
 ie
 ig

tests/testthat/test-options.R

@@ -0,0 +1,43 @@
+
+
+
+
+test_that("Can alter Gaussian Quadrature arguments", {
+    x <- data.frame(
+        vpt = c("b", "a", "c", "d", "e"),
+        vtime = c(10, 20, 30, 25, 15),
+        vevent = c(1, 1, 0, 1, 0),
+        vcov1 = c("A", "A", "B", "B", "A"),
+        vcov2 = rnorm(5)
+    )
+
+    ## Test defaults 15 + "legendre"
+    df <- DataSurvival(
+        data = x,
+        formula = Surv(vtime, vevent) ~ vcov1 * vcov2
+    )
+    li <- as.list(df)
+    expect_equal(li$n_nodes, 15)
+    expect_equal(
+        li[c("nodes", "weights")],
+        statmod::gauss.quad(15, "legendre")
+    )
+
+
+    ## Test modified values
+    options("jmpost.gauss_quad_n" = 20)
+    df <- DataSurvival(
+        data = x,
+        formula = Surv(vtime, vevent) ~ vcov1 * vcov2
+    )
+    li <- as.list(df)
+    expect_equal(li$n_nodes, 20)
+    expect_equal(
+        li[c("nodes", "weights")],
+        statmod::gauss.quad(20, "legendre")
+    )
+
+
+    ## Reset back to default to not impact additional tests
+    options("jmpost.gauss_quad_n" = 15)
+})