Ratings - please read before commenting

margaret thatcher
Flyweight
Posts: 28909
Joined: 22 Jul 2019, 15:43

Re: Ratings - please read before commenting

Post by margaret thatcher »

judges often give sympathy rounds in really one sided fights, you'll often see 119-109 or 118-110 pop up in fights where pretty much everyone online saw it as a shutout much
JCS
Heavyweight
Heavyweight
Posts: 5805
Joined: 17 Dec 2004, 13:27

Re: Ratings - please read before commenting

Post by JCS »

margaret thatcher wrote: 05 Sep 2022, 18:29 judges often give sympathy rounds in really one sided fights, you'll often see 119-109 or 118-110 pop up in fights where pretty much everyone online saw it as a shutout much
But w/ a rating system, I think you need to go with norms/averages.

Normally and on average, a 117-111 is "closer" than a 120-108...
computerrank
Editor
Editor
Posts: 2242
Joined: 04 Jan 2003, 18:59

Re: Ratings - please read before commenting

Post by computerrank »

@Jason

The current benchmark is 84.43% winner prediction ratio with
- a 100% win for a UD, when scores are not known
- a 100% win with at least 0.5 score points difference per judge per round (e.g. 117 - 111 for a 12 rounds bout)
- and a full win on points is equivalent to a TKO/KO

Now I found a slightly improved winner prediction ratio of 84.52 % with following changes :TU:
- a 87.5% win for a UD, when scores are not known
- a 87.5% win with at least 0.5 score points difference per judge per round (e.g. 117 - 111 for a 12 rounds bout)
- a 100% win with at least 1 score points difference per judge per round (e.g. 120 - 108 for a 12 rounds bout)
JCS
Heavyweight
Heavyweight
Posts: 5805
Joined: 17 Dec 2004, 13:27

Re: Ratings - please read before commenting

Post by JCS »

They are different sports.. but in MMA, I found the best prediction rates when the 'typical' SDs were around 55%, MDs around 60% and the UDs around 90%. (Draw being 50%).

Have you simplified your example or are you dynamically scaling? In other words... is a 118-110% worth ~92%? Or is it only worth 87.5%?

The max I apply to a decision is 99%.. have not statistically analyzed this, but I figure in a tiebreaker situation.. it is more fair to say that a stoppage should be worth slightly more.
pugilisticspecialist
Super Middleweight
Posts: 220
Joined: 19 May 2009, 16:23

Re: Ratings - please read before commenting

Post by pugilisticspecialist »

computerrank wrote: 06 Sep 2022, 16:49 @Jason

The current benchmark is 84.43% winner prediction ratio with
- a 100% win for a UD, when scores are not known
- a 100% win with at least 0.5 score points difference per judge per round (e.g. 117 - 111 for a 12 rounds bout)
- and a full win on points is equivalent to a TKO/KO

Now I found a slightly improved winner prediction ratio of 84.52 % with following changes :TU:
- a 87.5% win for a UD, when scores are not known
- a 87.5% win with at least 0.5 score points difference per judge per round (e.g. 117 - 111 for a 12 rounds bout)
- a 100% win with at least 1 score points difference per judge per round (e.g. 120 - 108 for a 12 rounds bout)
I would just keep tweaking it until you find the formula with the highest prediction ratio, then go with that.
computerrank
Editor
Editor
Posts: 2242
Joined: 04 Jan 2003, 18:59

Re: Ratings - please read before commenting

Post by computerrank »

I tweaked the rating parameters a bit more and now got a winner prediction ratio of 84.56% (improved from 84.43% now):
- assuming a 87.5% win for a UD, when scores are not known; I also tested lower and higher values for the parameter
- assuming a 87.5% win with at least 0.5 score points difference per judge per round (e.g. 117-111 for a 12 rounds bout, I also tested 116-112 and 118-110); and a linear dependence on the score difference per judge and round
- assuming a 100% win with at least 1 score points difference per judge per round (e.g. 120 - 108 for a 12 rounds bout, I varied that parameter from 100%+ to 100%-); and a linear dependence on the score difference per judge and round
- assuming a 52.5% win for a MD, when scores are not known; I varied that parameter
- assuming a 42.5% win for a SD, when scores are not known; I varied that parameter
JCS
Heavyweight
Heavyweight
Posts: 5805
Joined: 17 Dec 2004, 13:27

Re: Ratings - please read before commenting

Post by JCS »

I figured there was room for improvement here.

curious if this changes the rankings much..
computerrank
Editor
Editor
Posts: 2242
Joined: 04 Jan 2003, 18:59

Re: Ratings - please read before commenting

Post by computerrank »

JCS wrote: 07 Sep 2022, 19:58 I figured there was room for improvement here.

curious if this changes the rankings much..
Nothing exiting - only small moves.
AA484
Featherweight
Posts: 10
Joined: 13 Jun 2016, 10:21

Re: Ratings - please read before commenting

Post by AA484 »

So we are never again going to see actual numbers on the ratings page or on the matchup pages?
computerrank
Editor
Editor
Posts: 2242
Joined: 04 Jan 2003, 18:59

Re: Ratings - please read before commenting

Post by computerrank »

AA484 wrote: 09 Sep 2022, 00:04 So we are never again going to see actual numbers on the ratings page or on the matchup pages?
That is not to my decision.
Daedalus
Lightweight
Posts: 56
Joined: 20 Sep 2013, 12:59

Re: Ratings - please read before commenting

Post by Daedalus »

Does anyone really believe Andy Ruiz's record makes him the world's #11 lb for lb?
SportsRatings
Middleweight
Posts: 415
Joined: 26 May 2010, 23:15

Re: Ratings - please read before commenting

Post by SportsRatings »

Daedalus wrote: 09 Sep 2022, 15:46 Does anyone really believe Andy Ruiz's record makes him the world's #11 lb for lb?
It's more that he has too many pounds in the divisor
pugilisticspecialist
Super Middleweight
Posts: 220
Joined: 19 May 2009, 16:23

Re: Ratings - please read before commenting

Post by pugilisticspecialist »

computerrank wrote: 07 Sep 2022, 18:41 I tweaked the rating parameters a bit more and now got a winner prediction ratio of 84.56% (improved from 84.43% now):
- assuming a 87.5% win for a UD, when scores are not known; I also tested lower and higher values for the parameter
- assuming a 87.5% win with at least 0.5 score points difference per judge per round (e.g. 117-111 for a 12 rounds bout, I also tested 116-112 and 118-110); and a linear dependence on the score difference per judge and round
- assuming a 100% win with at least 1 score points difference per judge per round (e.g. 120 - 108 for a 12 rounds bout, I varied that parameter from 100%+ to 100%-); and a linear dependence on the score difference per judge and round
- assuming a 52.5% win for a MD, when scores are not known; I varied that parameter
- assuming a 42.5% win for a SD, when scores are not known; I varied that parameter
Is this live?
computerrank
Editor
Editor
Posts: 2242
Joined: 04 Jan 2003, 18:59

Re: Ratings - please read before commenting

Post by computerrank »

It is set to go live with the next software update ...
pugilisticspecialist
Super Middleweight
Posts: 220
Joined: 19 May 2009, 16:23

Re: Ratings - please read before commenting

Post by pugilisticspecialist »

computerrank wrote: 17 Sep 2022, 18:16 It is set to go live with the next software update ...
When's that due?
computerrank
Editor
Editor
Posts: 2242
Joined: 04 Jan 2003, 18:59

Re: Ratings - please read before commenting

Post by computerrank »

pugilisticspecialist wrote: 17 Sep 2022, 19:58
computerrank wrote: 17 Sep 2022, 18:16 It is set to go live with the next software update ...
When's that due?
Maybe next week.
margaret thatcher
Flyweight
Posts: 28909
Joined: 22 Jul 2019, 15:43

Re: Ratings - please read before commenting

Post by margaret thatcher »

:yay:

Image
jujigatame
Heavyweight
Heavyweight
Posts: 6158
Joined: 30 Oct 2004, 21:08

Re: Ratings - please read before commenting

Post by jujigatame »

margaret thatcher wrote: 19 Sep 2022, 01:16 :yay:

Image
Even he looks skeptical.
Koski
Welterweight
Posts: 3
Joined: 07 Feb 2012, 02:22

Re: Ratings - please read before commenting

Post by Koski »

What's wrong with the cruiserweight rankings? How in the world is Okolie number 1? He hasn't beaten anyone noteworthy. Jai Opetaia is the IBF and Ring champion, he should be number one. And Briedis is the former Ring champ and 3-time world champ, he should be number two. And maybe then Okolie.
computerrank
Editor
Editor
Posts: 2242
Joined: 04 Jan 2003, 18:59

Re: Ratings - please read before commenting

Post by computerrank »

Koski wrote: 06 Oct 2022, 09:22 What's wrong with the cruiserweight rankings? How in the world is Okolie number 1? He hasn't beaten anyone noteworthy. Jai Opetaia is the IBF and Ring champion, he should be number one. And Briedis is the former Ring champ and 3-time world champ, he should be number two. And maybe then Okolie.
First - belts do not count for BoxRec ratings in any way. Only official results count regarding the ratings of both opponents at time of their bout.
All top 3 boxers are very close currently.
- Okolie has a raw rating of 105 points and recently defeated Cieslak with 43 points and Glowacki with 28 points and before Askin with 22 points, which results in a current rating of 84.4 points.
- Opetaia has a raw rating of 75 points and recently defeated higher rated Briedis with 97 points, but had no really counting result before, which results in a current rating of 84.0 points.
- Briedis has a raw rating of 97 points and recently lost to lower rated Opetaia with 75 points, but recently also defeated Dorticos in a mixed decision with 60 points and before defeated Glowacki with 38 points and and Mikaelyan with 23 points, which results in a current rating of 82.5 points.
jujigatame
Heavyweight
Heavyweight
Posts: 6158
Joined: 30 Oct 2004, 21:08

Re: Ratings - please read before commenting

Post by jujigatame »

Please, please lobby whoever is in charge to get the numerical ratings added back to the ratings pages and (especially) the fighter pages. The stars are woefully inadequate and seeing the real numbers would go a long way towards reducing questions like the above.
aumonier
Bantamweight
Posts: 11
Joined: 15 Mar 2018, 20:27

Re: Ratings - please read before commenting

Post by aumonier »

computerrank wrote: 06 Oct 2022, 16:27
Koski wrote: 06 Oct 2022, 09:22 What's wrong with the cruiserweight rankings? How in the world is Okolie number 1? He hasn't beaten anyone noteworthy. Jai Opetaia is the IBF and Ring champion, he should be number one. And Briedis is the former Ring champ and 3-time world champ, he should be number two. And maybe then Okolie.
First - belts do not count for BoxRec ratings in any way. Only official results count regarding the ratings of both opponents at time of their bout.
All top 3 boxers are very close currently.
- Okolie has a raw rating of 105 points and recently defeated Cieslak with 43 points and Glowacki with 28 points and before Askin with 22 points, which results in a current rating of 84.4 points.
- Opetaia has a raw rating of 75 points and recently defeated higher rated Briedis with 97 points, but had no really counting result before, which results in a current rating of 84.0 points.
- Briedis has a raw rating of 97 points and recently lost to lower rated Opetaia with 75 points, but recently also defeated Dorticos in a mixed decision with 60 points and before defeated Glowacki with 38 points and and Mikaelyan with 23 points, which results in a current rating of 82.5 points.
Since the ratings are no longer communicated to the viewers of the site, the numbers you announce can be exact, or not. For all we know, rankings in every category can also be a discretional decision by the runners of the site. Should a question be asked in the forum everytime a change appears in the rankings (in this case Okolie replacing Opetaia as number 1 in his category) so the ratings number are provided ? Does it work if we ask the ratings detail of boxers that are 164th and 34th respectively (one fight in the schedule has this ranked fighters in the next few days) ?
computerrank
Editor
Editor
Posts: 2242
Joined: 04 Jan 2003, 18:59

Re: Ratings - please read before commenting

Post by computerrank »

BoxRec is just swamped by more prior requirements at the moment.
AA484
Featherweight
Posts: 10
Joined: 13 Jun 2016, 10:21

Re: Ratings - please read before commenting

Post by AA484 »

computerrank wrote: 07 Oct 2022, 06:25 BoxRec is just swamped by more prior requirements at the moment.
Thanks -- at least you got some feedback. Hopefully they will address it when they have the chance.
Manrae
Heavyweight
Heavyweight
Posts: 236
Joined: 28 Nov 2002, 18:57

Re: Ratings - please read before commenting

Post by Manrae »

computerrank wrote: 07 Oct 2022, 06:25 BoxRec is just swamped by more prior requirements at the moment.
Is one of those site security? Was there a breach? Data leak?
Post Reply