New Clusters for the January 2024 Tournament
5 Clusters for the New Year!
The Basho is on! (and ow ow ow ow on Ura’s arm….)
Okay, I’ve updated my “pivot clusters” for January 2024, with the new banzuke.
Explanation and description here in prior posts:
Initial Cluster Analysis of November 2023 Makuuchi: Two Clusters
Three clusters of kimarite style - throwing a little spin adds interest
The concept is that I’m looking at the distribution of winning kimarite for each rikishi for his complete pro record, and seeing if I can meaningfully characterize “his style of sumo”. I’m using the Hellinger distance and a particular approach to clustering to try this right now, and may attempt other approaches in the future.
Onosato has so few wins, he’s a little troublesome. I looked at him for a moment and thought “Should I keep him? Hmmm. I guess I’ll keep him…”
So of course, he defined one of the clusters.
Cluster pivot-definers:
Takakeisho
Hokuseiho
Midorifuji
Onosato
Hoshoryu
As in my prior cluster analyses, Takakeisho and Hokuseiho are the extreme oshidashi and yorikiri poles.
Midorifuji is kind of a weirdo, with his very high level of katasukashi wins, with about an even amount between oshidashi and yorikiri. You’ll see when I mark out the clusters, he stands almost alone.
Onosato is also somewhat weird, evenly split between oshidashi and yorikiri… but here’s the deal. He has a 77% win rate out of his matches, and only 36 wins total. He is an extreme newbie. He doesn’t even have a topknot yet.
So yeah, Onosato is throwing things off a little bit.
Hoshoryu is an interesting case - a broad range of throws especially in his kimarite.
Splitting up the banzuke into 5 clusters
So let’s see how it shakes out.
Here is how the count goes for each cluster (including the cluster-definer):
Takakeisho: 17
Hokuseiho: 10
Midorifuji: 2
Onosato: 4
Hoshoryu: 9
So while Onosato does throw things off a little, he only brings three wrestlers along with him: Atamifuji, Mitakeumi, and Myogiryu.
Midorifuji only gets Ura in his cluster with him.
I may be looking a little more at these clusters after the basho - but in general, I may start dipping into trying to see what indicates who “escapes” Makushita into the paying ranks (which is tough), or if I can model this via a Markov process, or some such.
