Ratings - please read before commenting - Archived

[JCS]

Look throughout this thread and everyone that’s disagreed strongly have either been ignored or “disappeared”/given up. He’s had countless criticism and doesn’t bend to any rules. Also from previous posts he doesn’t even have boxing experience or statistical experience so I’m unsure how this makes him qualified. I appreciate him taking time to contribute to this site however I don’t understand why he never takes criticism on board. His previous ranking system was much better and I don’t have any personal reasons I just sick of seeing fighters mis ranked so badly. But is what it is, computerrank will just continue to do his own thing

The tale of two extremes. Mr. Cobweb kissing butt and praying to the statistics.... and you're just being a tool.

I knew most would hate this new system and called it successfully. After all, Martin and I learned this ~15 years ago when we chased only prediction rate. He did the ratings back then... whereas I foolishly convinced him that prediction rate was the sole key to victory and assisted in the endeavor, only to have everyone hate it... and we eventually reverted back to more of a hybrid system that balanced prediction rate and traditional sensibility. Personally, I learned my lesson from that point on. Tradition and common sense must be observed as primary attributes, while prediction rate helps guide change.

I had assumed CR *was* Martin, or at least thought CR signed a post as Martin. Anyway, so it's back to prediction rate, huh. How does it work?

There's no point in trying to analyze rating movements through a fighter's career... in this system, it'll make almost zero sense. It just doesn't work like the past systems in use here. There's more than just the outcome of the fight in question to consider... but what that opponent's opponents have recently done, etc.

The only thing you can potentially derive from past ratings is when a fighter was perceived to be in their prime.

So we can't even look at it like old Pacquiao at uncertainty-coupled 32% of peak was still so good it was predicted to beat Thurman's inactivity-adjusted 51% of peak?

Why is your interest this deep to dig up all these old replies? Just curious..

[Daedalus]

Why is your interest this deep to dig up all these old replies? Just curious..

If you've read my previous questions, CR's responses were pretty much I've missed the whole discussion since last fall. I'm caught up now.

[Daedalus]

So, where is it?

Here is a try to show the impact of a bout to the ratings - an additional field in brackets:

Am I missing a setting?

Nobody was interested in that - it is off.

Hmm I count more than 10 people inquiring about before/after fight ratings since the WHR went live. And this is the closest to it. Yeah I get it's more like "let's pretend the fight didn't happen and see how it affect's the fighters' career paths" or something. Maybe the other readers don't understand either and don't know what they'll be agreeing to when you told at least 2 people you were going to do this.
Besides, it was very hard to find this topic. There's no forum dedicated to talking about BoxRec itself and I'd never have guessed it would be under Current Scene. I just noticed this thread by accident when it popped up on Announcements that first day I posted.
IMHO, showing the impact of individual bouts is the biggest single improvement you can do to demonstrate the usefulness of WHR, so I'm interested. Hope it's not that difficult to implement.

[computerrank]

...
Hmm I count more than 10 people inquiring about before/after fight ratings since the WHR went live. And this is the closest to it. Yeah I get it's more "let's pretend the fight didn't happen and see how it affect's the fighters' career paths" or something. Maybe the other readers don't understand either and don't know what they'll be agreeing to when you told at least 2 people you were going to do this.
Besides, it was very hard to find this topic. There's no forum dedicated to talking about BoxRec itself and I'd never have guessed it would be under Current Scene. I just noticed this thread by accident when it popped up on Announcements that first day I posted.
IMHO, showing the impact of individual bouts is the biggest single improvement you can do to demonstrate the usefulness of WHR, so I'm interested. Hope it's not that difficult to implement.

To find the impact of each bout at any time would mean to step by step exclude each bout by an individual simulation from the rest of a all other bouts. More than 2 million simulations after each new bout added.

[Daedalus]

...
IMHO, showing the impact of individual bouts is the biggest single improvement you can do to demonstrate the usefulness of WHR, so I'm interested. Hope it's not that difficult to implement.

To find the impact of each bout at any time would mean to step by step exclude each bout by an individual simulation from the rest of a all other bouts. More than 2 million simulations after each new bout added.

Maybe I'm misunderstanding again. So you're saying your response below:

In order to get a measure for the rating impact of a bout result, you can make the calculation without this bout and compare the resulting ratings. So you can test, whether a bout result enhanced a rating or lowered it.

e.g. without Klitschko win -320, without Seferi win +-0, without first Wilder SD -500, without second Wilder TKO -740

Took so much effort in providing actual examples on a listing:

That you add a value for clarity doesn't change the system and and doesn't reply many of my questions.

You moan: "You reply, that we can't see it, but you don't let us know how you will make it visible." -> I posted an example yesterday (you don't follow the thread ):
viewtopic.php?p=5371470#p5371470

ad 7:
ad 9:

...
Nobody can understand the ratings now. The users used to get useful information. They could see, how a fighters points rose or fell after a fight. Now they get nothing. A lot of points are added, but nobody can see why or when.

...
What I will do. I will show the specific impact of every bout by attaching a second value. Showing a measure of points won or lost.

So can't I say that for fighting Groves on 9/28/18, Callum Smith gained 164 pts he otherwise wouldn't have? Which is what I want to be able to see. And I'm not getting why it would be so hard to show, because Smith's 537 points that day was computed from somewhere; my understanding is it was 373 the day before, right?

[computerrank]

@Daedalus,
for each user looking at a boxer's profile Boxrec would have to calculate his WHR ratings with and without each specific bout of his career based upon all recorded bouts of all boxers at that moment in history.

This changes with each new bout with any boxer involved changed or addded.

[Daedalus]

@Daedalus,
for each user looking at a boxer's profile Boxrec would have to calculate his WHR ratings with and without each specific bout of his career based upon all recorded bouts of all boxers at that moment in history.

This changes with each new bout with any boxer involved changed or addded.

How is that a problem, the ratings are in perpetual change anyway?

[felinoboxing]

Waiting for P4P in boxrec's annual rating

[computerrank]

Waiting for P4P in boxrec's annual rating

Will need some days ...

[computerrank]

Friday Oct. 16th, Emanuel Colon vs. Antonio Moran is showing as a "no-star" bout. Should be a 2-star fight.
Saturday Oct. 17th, Lewis Ritson vs. Miguel Vazques is showing as a "no-star" bout. Should be a 3-star fight.

They both headline their cards...

I still have no idea, why that happens. There is no entry in the logs. After an instant update everything is ok again ...

[computerrank]

@Daedalus,
for each user looking at a boxer's profile Boxrec would have to calculate his WHR ratings with and without each specific bout of his career based upon all recorded bouts of all boxers at that moment in history.

This changes with each new bout with any boxer involved changed or addded.

How is that a problem, the ratings are in perpetual change anyway?

They are in perpetual change.
The new rating values after each result update are calculated for all existing bouts simultaneously in one big calculation. The results are stored in the database. These ratings are shown currently in the boxer profile, when required by a visitor.
You want to see the boxer profile listing the boxer's and opponent's ratings for each bout in the profile without that bout and with it. More precise, to show the ratings at bout time without that bout and the ratings after that bout.
Assumed a boxer has 50 career bouts. In order to get the current impact of each of the 50 profile bouts to the ratings of the other profile bouts, Boxrec would have to instantly perform about 50 extra big calculations. Boxrec could then show the ratings at bout time without that bout and the ratings after that bout.

[Daedalus]

In order to get a measure for the rating impact of a bout result, you can make the calculation without this bout and compare the resulting ratings. So you can test, whether a bout result enhanced a rating or lowered it.

e.g. without Klitschko win -320, without Seferi win +-0, without first Wilder SD -500, without second Wilder TKO -740

...
Nobody can understand the ratings now. The users used to get useful information. They could see, how a fighters points rose or fell after a fight. Now they get nothing. A lot of points are added, but nobody can see why or when.

...
What I will do. I will show the specific impact of every bout by attaching a second value. Showing a measure of points won or lost.

You moan: "You reply, that we can't see it, but you don't let us know how you will make it visible." -> I posted an example yesterday (you don't follow the thread ):
viewtopic.php?p=5371470#p5371470

ad 7:

So, where is it?

Nobody was interested in that - it is off.

...
IMHO, showing the impact of individual bouts is the biggest single improvement you can do to demonstrate the usefulness of WHR, so I'm interested. Hope it's not that difficult to implement.

To find the impact of each bout at any time would mean to step by step exclude each bout by an individual simulation from the rest of a all other bouts. More than 2 million simulations after each new bout added.

How is that a problem, the ratings are in perpetual change anyway?

They are in perpetual change.
The new rating values after each result update are calculated for all existing bouts simultaneously in one big calculation. The results are stored in the database. These ratings are shown currently in the boxer profile, when required by a visitor.
You want to see the boxer profile listing the boxer's and opponent's ratings for each bout in the profile without that bout and with it. More precise, to show the ratings at bout time without that bout and the ratings after that bout.
Assumed a boxer has 50 career bouts. In order to get the current impact of each of the 50 profile bouts to the ratings of the other profile bouts, Boxrec would have to instantly perform about 50 extra big calculations. Boxrec could then show the ratings at bout time without that bout and the ratings after that bout.

Well, it was your idea, admittedly a great solution, at least for me. Sorry, when you posted multiple examples it looked like no big deal for the database. When did you realize it was really impractical?

[computerrank]

...
Well, it was your idea, admittedly a great solution, at least for me. Sorry, when you posted multiple examples it looked like no big deal for the database. When did you realize it was really impractical?

Short after the examples end of May.

[computerrank]

Waiting for P4P in boxrec's annual rating

Will need some days ...

P4P Boxrec annual ratings now updated ...

[Offthehook]

...
I find it, as a boxing fan, a pretty flat algorithm, if we are starting to give guys piggy backs off of what a guy who beat them via UD does in the future. What about a guy whose beating a champion, and losing a fight in the last round, or another fighter who is robbed blind by the judges? Those guys lose a good chunk of their points, when they should have more points, than a guy who loses a UD.

So basically the future points of a fighter will go up, if the fighter they lost to by any means (as i say caroll was beaten by a UD, and it was decisive) wins bigger fights, or fights around their level?

All Whole-History Ratings of all boxers change with each new bout as each new bout changes the likelihood of all other results.

you should be a politician with that answer. So basically the more maxi hughes wins, the higher his previous opponents (win or lose) go?

[computerrank]

...
I find it, as a boxing fan, a pretty flat algorithm, if we are starting to give guys piggy backs off of what a guy who beat them via UD does in the future. What about a guy whose beating a champion, and losing a fight in the last round, or another fighter who is robbed blind by the judges? Those guys lose a good chunk of their points, when they should have more points, than a guy who loses a UD.

So basically the future points of a fighter will go up, if the fighter they lost to by any means (as i say caroll was beaten by a UD, and it was decisive) wins bigger fights, or fights around their level?

All Whole-History Ratings of all boxers change with each new bout as each new bout changes the likelihood of all other results.

you should be a politician with that answer. So basically the more maxi hughes wins, the higher his previous opponents (win or lose) go?

I depends, whether losses against him limited any other wins - or whether wins against him may compensate other losses.

[computerrank]

You can now also find Boxrec Annual Ratings for amateurs:

https://boxrec.com/media/index.php/BoxR ... gs_Amateur

[Daedalus]

...
So basically the future points of a fighter will go up, if the fighter they lost to by any means (as i say caroll was beaten by a UD, and it was decisive) wins bigger fights, or fights around their level?

All Whole-History Ratings of all boxers change with each new bout as each new bout changes the likelihood of all other results.

you should be a politician with that answer. So basically the more maxi hughes wins, the higher his previous opponents (win or lose) go?

So, where is it?

Nobody was interested in that - it is off.

This is a politician's answer, instead of "we can't do it"

[Daedalus]

For the old Boxrec r_41 release:
- raw winner prediction ratio based on the old rating only = 0.7482
- winner prediction ratio for r_41 with 16 additional parameters = 0.8345

For WHR:
- raw winner prediction ratio based on the WHR rating only = 0.8308
- winner prediction ratio for r_41 with 16 additional parameters = 0.8402

So I guess it's back to this. How does it predict?
If 20 years ago, fighter A (250 pts) were to fight B (249) the next day, does WHR say A will win? Draw? SD?
If fighter C (250 pts today) were to fight D (249) tomorrow, same questions.

[computerrank]

All Whole-History Ratings of all boxers change with each new bout as each new bout changes the likelihood of all other results.

you should be a politician with that answer. So basically the more maxi hughes wins, the higher his previous opponents (win or lose) go?

So, where is it?

Nobody was interested in that - it is off.

This is a politician's answer, instead of "we can't do it"

Both is true - nobody was interested in the result at that point - and later if found, it would be to expensive ...

[computerrank]

For the old Boxrec r_41 release:
- raw winner prediction ratio based on the old rating only = 0.7482
- winner prediction ratio for r_41 with 16 additional parameters = 0.8345

For WHR:
- raw winner prediction ratio based on the WHR rating only = 0.8308
- winner prediction ratio for r_41 with 16 additional parameters = 0.8402

So I guess it's back to this. How does it predict?
If 20 years ago, fighter A (250 pts) were to fight B (249) the next day, does WHR say A will win? Draw? SD?
If fighter C (250 pts today) were to fight D (249) tomorrow, same questions.

The WHR win probability of boxer A is 250/(250+249) before 20 years and is 250/(250+249) for boxer C today.

[Daedalus]

If 20 years ago, fighter A (250 pts) were to fight B (249) the next day, does WHR say A will win? Draw? SD?
If fighter C (250 pts today) were to fight D (249) tomorrow, same questions.

The WHR win probability of boxer A is 250/(250+249) before 20 years and is 250/(250+249) for boxer C today.

Cool, didn't expect it to be so simple.
I understand WHR starts a debut fighter at 0 points and can go below that.
BoxRec's lowest rating is 1.000. How are the 0 and negative ratings handled?

[computerrank]

If 20 years ago, fighter A (250 pts) were to fight B (249) the next day, does WHR say A will win? Draw? SD?
If fighter C (250 pts today) were to fight D (249) tomorrow, same questions.

The WHR win probability of boxer A is 250/(250+249) before 20 years and is 250/(250+249) for boxer C today.

Cool, didn't expect it to be so simple.
I understand WHR starts a debut fighter at 0 points and can go below that.
BoxRec's lowest rating is 1.000. How are the 0 and negative ratings handled?

It is the natural rating. At start of his career each boxers gets 0.15 hidden prior wins and 2.85 hidden prior losses against an artificial opponent with rating 5 for proboxing men and 50 for proboxing women. These prior results tie each boxer to an initial start point.

There are no negative values. The values are somewhere between 0.0001 and 9999.

[Daedalus]

I understand WHR starts a debut fighter at 0 points and can go below that.
BoxRec's lowest rating is 1.000. How are the 0 and negative ratings handled?

It is the natural rating. At start of his career each boxers gets 0.15 hidden prior wins and 2.85 hidden prior losses against an artificial opponent with rating 5 for proboxing men and 50 for proboxing women. These prior results tie each boxer to an initial start point.

There are no negative values. The values are somewhere between 0.0001 and 9999.

Ok, so if 1-pt boxers keep losing, they go below 1 but the display stays 1.000. Is that correct, or are more hidden wins added to bring up to 1?
I know past formulas displayed down to 0, so why the 1-pt floor now?

[computerrank]

I understand WHR starts a debut fighter at 0 points and can go below that.
BoxRec's lowest rating is 1.000. How are the 0 and negative ratings handled?

It is the natural rating. At start of his career each boxers gets 0.15 hidden prior wins and 2.85 hidden prior losses against an artificial opponent with rating 5 for proboxing men and 50 for proboxing women. These prior results tie each boxer to an initial start point.

There are no negative values. The values are somewhere between 0.0001 and 9999.

Ok, so if 1-pt boxers keep losing, they go below 1 but the display stays 1.000. Is that correct, or are more hidden wins added to bring up to 1?
I know past formulas displayed down to 0, so why the 1-pt floor now?

There is no 1 point floor, neither for the calulation nor for the display. What example do you look at?