Annual “what is up with wrestlestat’s algorithm” thread

[1032004]

First off, love wrestlestat, best wrestling website in history IMO.

And yes, I know “they’re ratings, not rankings.”  But that doesn’t mean the algorithm can’t be improved.

I’m not sure if anything’s changed but the preseason rankings don’t seem as bad as before last season when IIC like 3 multiple time returning champs we’re not ranked #1.

But I present the 2 “ratings” that stick out to me the most:

1.  Hendrickson over Kerk

2.  Redshirt freshman Garrett Grice #11 at 133.  Hope he proves me wrong but he has done nothing to be that high.  13-2 as a redshirt, best win was over #57 (last year) Brendan Ferretti, losses to #25 Zaccone and #47 Carter.  Huh?  Honestly 133 does look a little shallow but still.

 

Wrestlestat guys if you’re reading this, love you but it seems there are 3 main areas for improvement IMO.

1.  NCAA tournament performance should have a higher weight 

2.  Bonus wins should have a lower weight

3.  Results from several years back should have a lower weight

[nhs67]

Aliens?

[ionel]

3 minutes ago, nhs67 said:

Aliens?

probably 

[Ohio Elite]

2 minutes ago, ionel said:

probably 

More like most likely 

[nhs67]

10 hours ago, Ohio Elite said:

More like most likely 

^ This

^ That

10 hours ago, ionel said:

probably 

 

[1032004]

Gable Steveson is currently #5, lol

[ionel]

24 minutes ago, 1032004 said:

Gable Steveson is currently #5, lol

... and Sammy Sosso is #6.  Those two are not the same or similar animal.

Edited December 28, 2024 by ionel

[1032004]

24 minutes ago, ionel said:

... and Sammy Sosso is #6.  Those two are not the same or similar animal.

Yeah I guess wrestlestat doesn’t mind if you miss 1 season, but 2 is where they draw the line. 
 

[bnwtwg]

This blog is the simpler answer to your question. Business intelligence is my bread and butter so I feel qualified to field this question with tenacity. The tl;dr is because quality algos are really difficult and anyone can prompt ChatGPT to get middling results. Real algos are essentially binary and like all things automated, don't always pass the eye test.

https://craftofcoding.wordpress.com/2019/09/20/why-designing-algorithms-is-hard/#:~:text=Algorithms are hard because humans,something a machine can do.

 

Why designing algorithms is hard

The hardest part of developing software is of course the algorithms. People often  think that it’s possible to write a program to do just about anything – but that is just not the case. There are things that humans can do that machines just can’t, and likely never will be able to do. Algorithms are hard because humans don’t think in the same black-and-white manner as machines do. Writing an algorithm to filter an image in a way similar to an Instagram filter is something the human mind cannot do, but interpreting the aesthetics of the filtered image is not something a machine can do. Nor should it. It is too hard to try and frame the complexities of the human mind in a series of steps which can be translated into a program. Even weather predicting models can be flummoxed by the fact that weather is unpredictable, and can change. Anything with some level of randomness in it is more challenging to pin-down, and hence more challenging to write algorithms for.

We take algorithms for granted because we think that  things that are inherently simple for humans, should be just as easy for machines. Millions of years of evolution have provided us with eyes are able to process visual imagery in a spectacular fashion, yet translating this to algorithmic form for the machine to mimic is almost impossible. Machines are able to easily store imagery, and manipulate it in ways the human visual system cannot, and yet they are not able to instinctively identify objects in a generic manner – although they have made inroads. A human is able to identify a tree species from a distance, a machine is not. Being able to write programs is one facet of software development, but being able to decipher the logic underpinning those programs relies on individuals who can realistically think outside the box, push the envelope so to speak. Look at sorting algorithms. When was the last time a really effective, new sorting algorithm was developed? Most “new” sorting algorithms are usually an extension of an existing algorithm, wringing out a few extra milliseconds of speed. But truly new, transformative sorting algorithms? They just haven’t appeared.

[ionel]

Question for a data wizard in the know.  How can Fish score bonus points without scouring advancement points?  Is this like the fictitious pinfall some sort of fictitious lossmajor? 

[Caveira]

1 hour ago, ionel said:

Question for a data wizard in the know.  How can Fish score bonus points without scouring advancement points?  Is this like the fictitious pinfall some sort of fictitious lossmajor? 

Algorithms are hard 

[ionel]

8 minutes ago, Caveira said:

Algorithms are hard 

  if then 

[Caveira]

2 minutes ago, ionel said:

  if then 

Edge case issues in algorithms are hard … even if the human brain can recognize them easily.  If you’ve written them it’s oddly easy to understand.  If you haven’t…. It’s oddly easy to wonder why they’re wrong so often.  
 

Not going after you sir ionel….   

lots of 

def process_input(x if x > 10: if x % 2 == 0: if x < 20: return 1 else: if x == 42: return 2 else: return 3 else: if x == 15:return 4 elif x > 50: return 5 else: return 6 elif x < 5: if x < 0: if x % 3 == 0: return 7 else: return 8 else: if x == 3: return 9else: if x == 1: return 10 elif x == 2: return 11 else: return 12 else: if x == 8: return 13 else: return 14

;

[ionel]

3 minutes ago, Caveira said:

Edge case issues in algorithms are hard … even if the human brain can recognize them easily.  If you’ve written them it’s oddly easy to understand.  If you haven’t…. It’s oddly easy to wonder why they’re wrong so often.  
 

Not going after you sir ionel….   

lots of 

def process_input(x if x > 10: if x % 2 == 0: if x < 20: return 1 else: if x == 42: return 2 else: return 3 else: if x == 15:return 4 elif x > 50: return 5 else: return 6 elif x < 5: if x < 0: if x % 3 == 0: return 7 else: return 8 else: if x == 3: return 9else: if x == 1: return 10 elif x == 2: return 11 else: return 12 else: if x == 8: return 13 else: return 14

;

Always good to verify and validate the code/model.  Of course we aren't talking rocket science here so not like anyone is going to crash into Mars if the code is bad.  

[Gus]

7 hours ago, ionel said:

Question for a data wizard in the know.  How can Fish score bonus points without scouring advancement points?  Is this like the fictitious pinfall some sort of fictitious lossmajor? 

Win pigtail match via bonus?

[Le duke]

Win pigtail match via bonus?

Maybe I’m giving them too much credit but I’d assume a guy ranked that high wouldn’t be predicted to be in a 32/33 match in their model.


Sent from my iPhone using Tapatalk

[1032004]

17 minutes ago, Gus said:

Win pigtail match via bonus?

That still gives 1 advancement point 

[Wrestleknownothing]

In one sense the wrestlestat "algorithm" for advancement and placement is as simple as it gets. The top 16 guys score. No one else scores.

The #1 seed gets 20, the #2 gets 16, etc. But then things get weird at the margin. Ohio State's #7 seed, Carson Kharchla, is assumed to score 9 advancement and placement points, but PSU's Davis and OSU's Amine, both #7, are assumed to score 6.5. How does that work? Well since the most advancement and placement points possible for 7th is 7.5 it does not work in the case of Kharchla. We have probably found a bug in this case.

There is also the issue of the guaranteed advancement point for either the #32 or #33 wrestler. They are missing that.

But I have a problem with this binary approach of assuming the #1 seed will always win, the #2 seed will always fall in the final, etc. Everyone makes that assumption. Flo, Intermat, Wrestlestat, WIN, etc. Everyone. And it is a bad assumption. Sometimes you get a Braeden Davis at #1, and sometimes you get a Cody Brewer at #13. That has to be accounted for. To do that you have to take a probabilistic approach. But no one does. Well, someone does.

As for bonus points, I have no idea how they are determining those at the individual wrestler level. There does not seem to be much of a seed component to the values produced, so it must rely heavily on the individual wrestler results. Perhaps it is tied to bonus percentage. Got me. But they are also missing the important caveat that if a wrestler fails to score an advancement or placement point then they will also fail to score a bonus point. That is just another oversight.

[BigRedFan]

48 minutes ago, Wrestleknownothing said:

 

As for bonus points, I have no idea how they are determining those at the individual wrestler level. There does not seem to be much of a seed component to the values produced, so it must rely heavily on the individual wrestler results.

N=1:  Shapiro got 7.5 bonus points last NCAA tournament, and WS is crediting him to score 7.5 points this year.

[Wrestleknownothing]

12 minutes ago, BigRedFan said:

N=1:  Shapiro got 7.5 bonus points last NCAA tournament, and WS is crediting him to score 7.5 points this year.

It seems weird that Shapiro is one of the only ones to got to just the tenth digit AND to match last year's total. Though there are some others like Levi Haines with 6 last year and 6 projected.

Then they have Julian Ramirez scoring 3.94 bonus after scoring 0 last year. So not everything is a carry over from the prior year.

As a matter of fact, they have the other two 7.5 bonus point scorers from last year scoring 2.95 and 3.38 this year.

I see no way to reverse engineer their algo.

Edited January 1 by Wrestleknownothing

[666]

On 9/7/2023 at 5:28 PM, 1032004 said:

First off, love wrestlestat, best wrestling website in history IMO.

And yes, I know “they’re ratings, not rankings.”  But that doesn’t mean the algorithm can’t be improved.

I’m not sure if anything’s changed but the preseason rankings don’t seem as bad as before last season when IIC like 3 multiple time returning champs we’re not ranked #1.

But I present the 2 “ratings” that stick out to me the most:

1.  Hendrickson over Kerk

2.  Redshirt freshman Garrett Grice #11 at 133.  Hope he proves me wrong but he has done nothing to be that high.  13-2 as a redshirt, best win was over #57 (last year) Brendan Ferretti, losses to #25 Zaccone and #47 Carter.  Huh?  Honestly 133 does look a little shallow but still.

 

Wrestlestat guys if you’re reading this, love you but it seems there are 3 main areas for improvement IMO.

1.  NCAA tournament performance should have a higher weight 

2.  Bonus wins should have a lower weight

3.  Results from several years back should have a lower weight

I wouldn't get too bent about it, everything always works out in the end at NCAAs. Wrestlestat's rankings are "good enough for government work". 

[BigRedFan]

On 1/1/2025 at 12:20 PM, Wrestleknownothing said:

It seems weird that Shapiro is one of the only ones to got to just the tenth digit AND to match last year's total. Though there are some others like Levi Haines with 6 last year and 6 projected.

Then they have Julian Ramirez scoring 3.94 bonus after scoring 0 last year. So not everything is a carry over from the prior year.

As a matter of fact, they have the other two 7.5 bonus point scorers from last year scoring 2.95 and 3.38 this year.

I see no way to reverse engineer their algo.

I see that Shapiro is now only credited with 6.56 bonus points, despite getting two TFs in his two recent matches.  Go figure.

[1032004]

Looks like wrestlestat isn’t including the MFF losses even in the records up top?

[Idaho]

The only rankings that will matter will be the coaches final ranking and then what we call seeds... But it's always good to have the annual, I mean semi-annual.... ooops, monthly wrestlestat discussion. 

[Wrestleknownothing]

1 hour ago, 1032004 said:

Looks like wrestlestat isn’t including the MFF losses even in the records up top?

Example?