Where My Nerds At?

Misalorales

All-Conference
Jun 3, 2025
1,130
3,469
113
This one might not be for everyone.

But it be for me. And sometimes that's enough.

What would you say is the most accurate you could get,or hope to get, your model, realistically? Trying to predict a 33 man bracket and all of the various data within them is insanity. Doing it 10 separate times is something else entirely! Love the work, thanks for all you do.
 

Wrestleknownothing

All-Conference
Oct 30, 2021
1,667
4,403
113
What would you say is the most accurate you could get,or hope to get, your model, realistically? Trying to predict a 33 man bracket and all of the various data within them is insanity. Doing it 10 separate times is something else entirely! Love the work, thanks for all you do.
I think I would need to have access to much more granular data like what Track has. Go down to the individual action level across the whole season and build from there. There is a guy on intermat called Pablo who is doing it at the match results level for the whole season (kinda like wrestlestat's Elo), but not yet at the individual action level. He isn't yet predicting scores so it is hard to measure, but it looks interesting.
 

Misalorales

All-Conference
Jun 3, 2025
1,130
3,469
113
I think I would need to have access to much more granular data like what Track has. Go down to the individual action level across the whole season and build from there. There is a guy on intermat called Pablo who is doing it at the match results level for the whole season (kinda like wrestlestat's Elo), but not yet at the individual action level. He isn't yet predicting scores so it is hard to measure, but it looks interesting.
I distrust A.I, especially with sports data, but do you think utilizing it in conjunction with your database would supercharge your model or ability to refine said model? It feels like at some point in the future that's the way no?
 

Wrestleknownothing

All-Conference
Oct 30, 2021
1,667
4,403
113
I distrust A.I, especially with sports data, but do you think utilizing it in conjunction with your database would supercharge your model or ability to refine said model? It feels like at some point in the future that's the way no?
It is important to distinguish between the sub-categories of AI.

Large language models (what everyone is using when doing internet queries) handle unstructured data very well. Especially when there is no one correct answer. They work really well as editors and style coaches for written documents. They work really well for summaries. And they work well to assist with starting points for coding. But they still do particularly poorly at finding a single correct answer in a sea of wrong answers (sports trivia being a prime example). I have been waiting for them to get better at wrestling stats now that @cowcards has created his invaluable resource, but it has not happened yet.

And when it comes to math LLMs used to make very basic mistakes (did not carry digits) on basic math (since fixed), but when it comes to complex math they tend to punt. For example, I tested ChatGPT (and soon I will be testing Claude) recently by asking it to perform the algebraic expansion I use for calculating the odds of 0-10 AAs - a solved problem with a known equation. Rather than produce the results it produced the Python code you would use to get the results. Not what I asked for, but closely related as long as you have the ability to run Python code.

Machine learning models, on the other hand, can do particularly well at optimizations like what I am doing. The key is picking the right one, or picking the right agent to pick the right one. And there are a lot of technologies that fit in this sub-category. Even a simple linear regression is part of AI. So I can say with a straight face that when I do my surface optimizations I am already using AI.

The other issue is that the data I am using is very structured, and so far, pretty basic. With structured data you only need to go down the more sophisticated machine learning technique route when it gets more granular and sophisticated.
 

Stevej

Freshman
Oct 12, 2021
23
60
13

Psalm 1 guy

All-Conference
Nov 3, 2019
1,146
4,400
113
Was your "Penn 'We Ain't The Field' State adjustment" accurate this year? I would imagine Penn State having seven #1 seeds limited this points adjustment. I am curious to see if Penn State can top their 40.5 Bonus points at next year's NCAA tournament. I know there were 1.5 Bonus points left on the table by MM (.5 for missing a tech by a TD where he rode much of the match and not gaining the Major against Hepner), but I am not sure of the other wrestlers. Since I had guessed that Penn State would score 42 Bonus points, MM kept me from the perfect guess : )
 

pawrstlersinpa

All-Conference
Jan 26, 2013
1,422
1,994
113
I distrust A.I, especially with sports data, but do you think utilizing it in conjunction with your database would supercharge your model or ability to refine said model? It feels like at some point in the future that's the way no?
What????

My prompt for ChatGPT: Who were the division 1 wrestling national champions at each weight in 2026

The 2026 NCAA Division I Wrestling Championships took place March 19–21, 2026, in Cleveland, Ohio. Here are the individual national champions by weight class:
  • 125 lbsBrandon Kaylor (Oregon State)
  • 133 lbsVito Arujau (Cornell)
  • 141 lbsJesse Mendez (Ohio State)
  • 149 lbsRidge Lovett (Nebraska)
  • 157 lbsLevi Haines (Penn State)
  • 165 lbsDavid Carr (Iowa State)
  • 174 lbsCarter Starocci (Penn State)
  • 184 lbsAaron Brooks (Penn State)
  • 197 lbsZach Glazier (Iowa)
  • 285 lbsGreg Kerkvliet (Penn State)
If you want, I can also break down team standings, upsets, or how many titles Penn State stacked up again.
 

Wrestleknownothing

All-Conference
Oct 30, 2021
1,667
4,403
113
Was your "Penn 'We Ain't The Field' State adjustment" accurate this year? I would imagine Penn State having seven #1 seeds limited this points adjustment. I am curious to see if Penn State can top their 40.5 Bonus points at next year's NCAA tournament. I know there were 1.5 Bonus points left on the table by MM (.5 for missing a tech by a TD where he rode much of the match and not gaining the Major against Hepner), but I am not sure of the other wrestlers. Since I had guessed that Penn State would score 42 Bonus points, MM kept me from the perfect guess : )
Nice job on the bonus prediction.

I came in a touch high after the PSU adjustment. I think it was 185. I was low on the bonus and high on the placement.
 
  • Like
Reactions: Psalm 1 guy