changes for #38 regarding nested effects

DESCRIPTION

@@ -29,6 +29,7 @@ Imports:
 Suggests: 
     covr,
     gee,
+    ggplot2,
     knitr,
     nlme,
     rmarkdown,

R/misc.R

@@ -8,9 +8,15 @@
 
 reformat_lme_pred_data <- function(x, object) {
   random_effects <- names(lme4::ranef(object$fit))
-  for (i in random_effects) {
-    lvl <- levels(object$fit@frame[[i]])
-    x[[i]] <- factor(lvl[1], levels = lvl)
+  random_effects <- paste(random_effects, collapse = "+")
+  random_effects_f <- as.formula(paste("~", random_effects))
+  random_effect_cols <- all.vars(random_effects_f)
+  vals <- dplyr::select(object$fit@frame, dplyr::all_of(random_effect_cols))
+  vals <- vals[1,,drop = FALSE]
+
+  for (i in random_effect_cols) {
+    lvl <- levels(vals[[i]])
+    x[[i]] <- factor(vals[[i]][1], levels = lvl)
   }
   x
 }

inst/WORDLIST

@@ -1,47 +1,29 @@
 Bech
-Biometrics
-Biometrika
-CED
 CMD
-CRC
-Cano
 Codecov
-Correa
-DN
-DeBruine
-Galecki
-Hodgson
 Imipramine
-Inger
-LM
 Liang
 Lifecycle
-McElreath
-Minto
-PeerJ
+ORCID
 Pinheiro
 Reisby
-Sorensen
-Springer
-Thorson
-Vasishth
-Welch
-XA
 Zeger
 al
-arXiv
+biomet
+doi
 et
-id’
+funder
+glmer
+gls
+lme
+lmer
 mixedlevelmod
-modelling
+nlme
 pharmacokinetic
+poisson
 reprex
+stan
 tibble
+tidybayes
 tidymodels
 warmup
-th
-β
-β̂
-η
-η̂
-ORCID

multilevelmod.Rproj

@@ -18,5 +18,6 @@ LineEndingConversion: Posix
 
 BuildType: Package
 PackageUseDevtools: Yes
+PackageCleanBeforeInstall: Yes
 PackageInstallArgs: --no-multiarch --with-keep.source
 PackageRoxygenize: rd,collate,namespace

tests/testthat/test-linear-reg-lme4.R

@@ -46,3 +46,29 @@ test_that('mode specific package dependencies', {
     list(c("lme4", "multilevelmod"))
   )
 })
+
+
+test_that('random interactions and/or nesting', {
+  skip_if_not_installed("lme4")
+  skip_if_not_installed("ggplot2")
+  skip_on_cran()
+
+  data(mpg, package = "ggplot2")
+
+  fit <-
+    linear_reg() %>%
+    set_engine("lmer") %>%
+    fit(cty ~ year + (1|manufacturer/model), data = mpg)
+
+  expect_error(predict(fit, new_data = mpg), regexp = NA)
+  # Now predict with new levels for the random effect columns, esp
+  # when there is nesting/interactions
+  expect_error(
+    predict(fit, new_data = tibble::tibble(year = 2000, model = "dan", manufacturer = "max")),
+    regexp = NA
+  )
+
+
+
+})
+