I am working on a method of categorizing Sekitori by fighting style to facilitate reporting and possibly aid in a predictive model. To that last point, if two wrestlers have not fought each other at all, or only a few times, it should be helpful to look at how each wrestler fairs against wrestlers of their opponent’s fighting style. So the first step is to identify those styles.
My first post, from a little over a week ago, introduced the concept and a rough first go at the first category: the “Slapdown” wrestlers. This week we’ll take a look at the pusher-thrusters. While there’s quite a bit of variety here, moving from Bushozan down through Ura, I think this is a good start but may demonstrate a need for more groups. Unfortunately, that would work against the overall predictive model but it’s possible that it just can’t be helped.
From the frequency of kimarite employed in their wins, Bushozan, Daishomaru, Takakeisho, Tamawashi, Daieisho, and Kotokuzan rely heaviest on pushing-thrusting techniques. What’s interesting is that they not only shun any grappling techniques but also rarely use any slapdown techniques. They’re not so much trying to catch their opponent off balance as they are just trying to blast them away.
In the middle part of this table, things get a bit more tricky. Clearly yotsu techniques are used, sparingly, in this section of the table. However, I’m still pretty confident that this groups the wrestlers into the correct strategy grouping. Yutakayama and Shimanoumi use yotsu techniques to win more frequently than Takakeisho but they’re still clearly pusher-thruster wrestlers. Onosho is a bit of an edge case for a Slapdown wrestler and Ura is a special case as he employs many different techniques.
For the time being, I think I will keep this group membership as-is and see how it goes. Outlier wrestlers, like Tobizaru and Takayasu from the “Slapdown” group and Ura from this one, may just need to be on their own while the remainder of wrestlers may get filtered into even smaller cohorts. I could see this group becoming two distinct groups with Hokutofuji in between them and Ura possibly grouped with Terutsuyoshi and Enho. That will be a mess of 9-10 clusters and it looks real good because the wrestlers in that system feel like they’re in the right spot, but as the groups get smaller, it will hurt the goal of building the predictions…so we shall see.
Interesting analysis Andy, thx for taking the time to do this for us.
What I’m finding most interesting so far relates to Takakeisho. We all know and have on the regular witnessed that once his initial pushing/thrusting attack fails he’s gassed and resorts to one or more head pull\down/slapdown attempts. According to the stats you’ve collected it’s not a very useful and/or successful strategy for him.
As I’ve mentioned before, the key to beating him is simple (these stats seem to support this). Good footwork, lateral movement, and waiting for him to get gassed and for him to then attempt a pull\down/slapdown. That attempt is the key for moving in and finishing him off either on the belt or with a hard straight forward tsupari counter offensive.
I know it’s easier said than done but it’s the strategy I’d employ against him every time and I believe it would be successful at least 7 out of 10 times.
I agree. It might keep him from reaching Yokozuna. He will probably have a great tournament this time but it has been difficult for him to put together two consecutively.
Earlier Takakeisho was quite patient and didn’t really overexert himself – instead he’d throw a lot of feints and try to catch people out on timing. But perhaps that’s been figured out, and he now has to win early or not at all..
With some of those wrestlers who fall on the fence between strategy groups, it may be helpful to see what techniques they lose to in order to determine their main category. That is assuming that there is a trend in weakness determined by style.
Excellent idea. That’s been a separate factor, in particular whether someone is “slapdown prone”. But it could help in this initial classification step, too.
If you can lay hands on enough varieties of informative data you might consider applying some unsupervised learning methods (e.g., varimax rotation, lots of info here: https://twitter.com/karlrohe/status/1293190800587870210 ) and let the data do the grouping for you.