Fix rendering issues in Chapters 9 and 12

09-mapping.Rmd

@@ -1108,8 +1108,15 @@ g1 = ggplot() + geom_sf(data = nz, aes(fill = Median_income)) +
   scale_x_continuous(breaks = c(170, 175))
 g1
 library(ggspatial)
+# Refresh nz_elev to avoid 'external pointer is not valid' error
+if (!exists("nz_elev")) {
+  nz_elev = terra::rast(system.file("raster/nz_elev.tif", package = "spDataLarge"))
+}
 ggplot() + 
-  layer_spatial(nz_elev) +
+  tryCatch(layer_spatial(nz_elev), error = function(e) {
+    nz_elev <<- terra::rast(system.file("raster/nz_elev.tif", package = "spDataLarge"))
+    layer_spatial(nz_elev)
+  }) +
   geom_sf(data = nz, fill = NA) +
   annotation_scale() +
   scale_x_continuous(breaks = c(170, 175)) +

12-spatial-cv.Rmd

@@ -642,7 +642,7 @@ Let's have a look at the final AUROC\index{AUROC}: the model's ability to discri
 round(mean(score_spcv_svm$classif.auc), 2)
 ```
 
-It appears that the GLM\index{GLM} (aggregated AUROC\index{AUROC} was `r score[resampling_id == "repeated_spcv_coords" & learner_id == "classif.log_reg", round(mean(classif.auc), 2)]`) is slightly better than the SVM\index{SVM} in this specific case.
+It appears that the GLM\index{GLM} (aggregated AUROC\index{AUROC} was `r round(mean(score[score$resampling_id == "repeated_spcv_coords" & score$learner_id == "classif.log_reg", "classif.auc"]), 2)`) is slightly better than the SVM\index{SVM} in this specific case.
 To guarantee an absolute fair comparison, one should also make sure that the two models use the exact same partitions -- something we have not shown here but have silently used in the background (see `code/12_cv.R` in the book's GitHub repository for more information).
 To do so, **mlr3** offers the functions `benchmark_grid()` and `benchmark()` [see also https://mlr3book.mlr-org.com/chapters/chapter3/evaluation_and_benchmarking.html#sec-benchmarking, @bischl_applied_2024].
 We will explore these functions in more detail in the Exercises.