Merge branch 'dev'

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: cvms
 Title: Cross-Validation for Model Selection
-Version: 1.4.0.9000
+Version: 1.4.1
 Authors@R: 
     c(person(given = "Ludvig Renbo",
              family = "Olsen",

NEWS.md

@@ -1,7 +1,19 @@
 
-# cvms 1.4.0.9000
+# cvms 1.4.1
 
-* In `plot_confusion_matrix()`, fixes bug when `add_sums = TRUE` and `counts_on_top = TRUE`.
+## `plot_confusion_matrix()`:
+
+* Adds option to set `intensity_by` to a log/arcsinh transformed version of the counts. 
+This adds the options `"log counts"`, `"log2 counts"`, `"log10 counts"`, `"arcsinh counts"` 
+to the `intensity_by` argument.
+
+* Fixes bug when `add_sums = TRUE` and `counts_on_top = TRUE`.
+
+* Raises error for negative counts.
+
+* Fixes zero-division when all counts are 0.
+
+* Sets palette colors to lowest value when all counts are 0.
 
 # cvms 1.4.0
 

R/plot_confusion_matrix.R

@@ -93,8 +93,15 @@
 #' @param font_col_percentages \code{list} of font settings for the column percentages.
 #'  Can be provided with \code{\link[cvms:font]{font()}}.
 #' @param intensity_by The measure that should control the color intensity of the tiles.
-#'  Either \code{`counts`} or \code{`normalized`}. For the latter, the color limits become
-#'  \code{0-100}, why the intensities can better be compared across plots.
+#'  Either \code{`counts`}, \code{`normalized`} or one of \code{`log counts`,
+#'  `log2 counts`, `log10 counts`, `arcsinh counts`}.
+#'
+#'  For `normalized`, the color limits become \code{0-100}, why the intensities
+#'  can better be compared across plots.
+#'
+#'  For the `log*` and `arcsinh` versions, the log/arcsinh transformed counts are used.
+#'  \strong{Note}: In `log*` transformed counts, 0-counts are set to `0`, why they
+#'  won't be distinguishable from 1-counts.
 #' @param arrow_size Size of arrow icons. (Numeric)
 #'
 #'  Is divided by \code{sqrt(nrow(conf_matrix))} and passed on
@@ -371,7 +378,7 @@ plot_confusion_matrix <- function(conf_matrix,
   )
   checkmate::assert_names(
     x = intensity_by,
-    subset.of = c("counts", "normalized"),
+    subset.of = c("counts", "normalized", "log counts", "log2 counts", "log10 counts", "arcsinh counts"),
     add = assert_collection
   )
 
@@ -392,6 +399,13 @@ plot_confusion_matrix <- function(conf_matrix,
       "'palette' and 'sums_settings[['palette']]' cannot be the same palette.")
   }
 
+  # Check that N are (>= 0)
+  checkmate::assert_numeric(
+    x = conf_matrix[[counts_col]],
+    lower = 0,
+    add = assert_collection
+  )
+
   checkmate::reportAssertions(assert_collection)
   # End of argument checks ####
 

R/plotting_utilities.R

@@ -88,8 +88,8 @@ update_font_setting <- function(settings, defaults, initial_vals = NULL) {
 
 preprocess_numeric <- function(vec, settings, rm_zero_text=FALSE, rm_zeroes_post_rounding=TRUE) {
 
-  # Find the pre-rounding 0s
-  is_zero <- vec == 0
+  # Find the pre-rounding 0s or NaNs
+  is_zero <- vec == 0 | is.na(vec)
 
   # Don't round if digits is negative
   if (settings[["digits"]] >= 0) {
@@ -101,7 +101,7 @@ preprocess_numeric <- function(vec, settings, rm_zero_text=FALSE, rm_zeroes_post
   # Potentially including elements zero after rounding
   if (isTRUE(rm_zero_text)){
     if (isTRUE(rm_zeroes_post_rounding)){
-      out[vec == 0] <- ""
+      out[vec == 0 | is.na(vec)] <- ""
     } else {
       out[is_zero] <- ""
     }
@@ -228,24 +228,45 @@ get_figure_path <- function(fig_name, inst_dir = "images", pgk_name = "cvms") {
 
 
 calculate_normalized <- function(data){
-  data[["Normalized"]] <- 100 * (data[["N"]] / sum(data[["N"]]))
+  sum_ <- sum(data[["N"]])
+  if (sum_ == 0){
+    sum_ <- 1
+  }
+  data[["Normalized"]] <- 100 * (data[["N"]] / sum_)
   data
 }
 
-set_intensity <- function(data, intensity_by){
-  if (intensity_by == "counts") {
-    data[["Intensity"]] <- data[["N"]]
-  } else {
+set_intensity <- function(data, intensity_by) {
+  if (grepl("counts", intensity_by)) {
+    counts <- data[["N"]]
+
+    if (intensity_by == "log counts") {
+      counts[counts != 0] <- log(counts[counts != 0])
+    } else if (intensity_by == "log2 counts") {
+      counts[counts != 0] <- log2(counts[counts != 0])
+    } else if (intensity_by == "log10 counts") {
+      counts[counts != 0] <- log10(counts[counts != 0])
+    } else if (intensity_by == "arcsinh counts") {
+      counts <- asinh(counts)
+    }
+
+    data[["Intensity"]] <- counts
+
+  } else if (intensity_by == "normalized") {
     data[["Intensity"]] <- data[["Normalized"]]
   }
   data
 }
 
 get_intensity_range <- function(data, intensity_by){
   # Get min and max intensity scores
-  if (intensity_by == "counts"){
-    min_intensity <- min(data$N)
-    max_intensity <- max(data$N)
+  if (grepl("counts", intensity_by) ){
+    min_intensity <- min(data$Intensity)
+    max_intensity <- max(data$Intensity)
+    if (min_intensity == max_intensity && min_intensity == 0){
+      # When all are 0, make sure all get lowest value in palette
+      max_intensity <- 1
+    }
   } else {
     min_intensity <- 0
     max_intensity <- 100