Suppress NAN warnings when computing MDS of each cluster

In general, inspection and convention supports the sqrt trasnformation of cosine dissimilarity matrix for MDS, to reduce the relative impact of very high dissimilarity scores and increases the relative impact of lower scores. Planning on a feature request to make this configurable.

R/generate_themes.R

@@ -92,7 +92,7 @@ generate_themes <-
     mds = avg_cluster_embeddings %>%
       tibble::column_to_rownames("Cluster") %>%
       as.matrix() %>%
-      cosine_dissimilarity(root = T) %>%
+      cosine_dissimilarity(root = TRUE) %>%
       smacof::mds(type = "ordinal")
 
     #Generate distinct categorical colors for plotting

R/utils.R

@@ -16,7 +16,7 @@ cosine_dissimilarity <- function(mat, root = FALSE) {
   sim = sim %*% t(sim)
 
   if(root == T) {
-    D_sim = stats::as.dist(sqrt(1 - sim))
+    D_sim = stats::as.dist(suppressWarnings(sqrt(1 - sim)))
   } else {
     D_sim = stats::as.dist(1 - sim)
   }