Why cluster on predictions_df and not candidate_cases?

[hsm207]

Just wanted to check that the clustering logic here is correct:

    predictions_df <- rxPredict(model, candidate_cases, extraVarsToWrite=c("rev_id", "flagged"))
    predictions_df$entropy <- entropy(predictions_df$Probability)

    predictions_df$cluster_id <- predictions_df %>%
      dist(method="euclidean") %>%
      hclust(method="ward.D2") %>%
      cutree(k=N)

    selected <- predictions_df %>%
      group_by(cluster_id) %>%
      arrange(-entropy) %>%
      slice(which.max(entropy)) %>%
      as.data.frame

I was under the impression that the clusters should be based only the comment's features ie. v2...v51 in candidates_df.