In the case of the playing cards, I was demonstrating how different clustering algorithms would perform. I did that using k-means, agglomerative, and DBSCAN.

The criteria (or example) you've provided maybe incomplete. For example, you say you want the maximum number of creatures for each Col2, Col3 combination. However, you have 2 output rows for (Z st01 - rows 3 and 6) and (Z st02 - rows 1 and 5). Is there another rule to consider to include 2 rows per group sometimes?