Ratings - please read before commenting - Archived

[computerrank]

why Tyson Fury got points in the battle against:
Tom Schwartz 2019-06-15
and in the battle against Otto Valin 2019-09-14 ????

Look here:
viewtopic.php?f=22&t=28051&start=3812

[mike1989]

why Tyson Fury got points in the battle against:
Tom Schwartz 2019-06-15
and in the battle against Otto Valin 2019-09-14 ????

Look here:
viewtopic.php?f=22&t=28051&start=3812

Tyson Fury before the fight 508.5 points Otto Wallin 42.77 points. The panel of judges 118-110; 117-111; 116-112. Tyson after the battle 549.6 points Wallin 42.77 points. I am silent about the fact that the battle was competitive and the difference in rating is huge. How is it not logical at all ??? It is logical to lower the Fury rating and raise Wallin.

[computerrank]

Look here:
viewtopic.php?f=22&t=28051&start=3812

Tyson Fury before the fight 508.5 points Otto Wallin 42.77 points. The panel of judges 118-110; 117-111; 116-112. Tyson after the battle 549.6 points Wallin 42.77 points. I am silent about the fact that the battle was competitive and the difference in rating is huge. How is it not logical at all ??? It is logical to lower the Fury rating and raise Wallin.

The regular points didn't change. It was for additional points only. Both opponents were in the top 15 percent of all active boxers.

[computerrank]

Bout winner prediction results for 91,458 significant bouts between connected boxers in the time period of 2009 to 2018 (10 years).

Current ratings program release 41b:

- 74.83 percent - raw prediction based on the rating points only
- 83.45 percent - optimized prediction including 16 additional record parameters

WHR ratings:

- 79.89 percent - raw prediction based on the rating points only
- 85.31 percent - optimized prediction including 16 additional record parameters

This means the raw prediction ratio is much better for the WHR ratings. And the optimized prediction is significantly better too.

http://151.boxrec.com/~martin/ratings_whr.php
http://151.boxrec.com/~martin/ratings_at_whr.php

[computerrank]

...
Some big differences on the all time one. Naseem Hamed on U.K. All time P4P on current is 2nd on WHR is 80th....not saying which is more accurate just a big change.

As always I think the active WHR are fine.

WHR all time ratings need further treatment ...

[pugilisticspecialist]

When do you expect the new ratings to launch?

[computerrank]

When do you expect the new ratings to launch?

They still are not ready.

[computerrank]

...
Some big differences on the all time one. Naseem Hamed on U.K. All time P4P on current is 2nd on WHR is 80th....not saying which is more accurate just a big change.

As always I think the active WHR are fine.

WHR all time ratings need further treatment ...

Naseem Hamed opponents were too weak in the WHR ratings. But the WHR all time now with new algorithm:

- career top rating
- at most twice the mean of career top 5 defeated opponents
- values of last 10 years bouts are reduced, as the those WHR ratings are not yet depreciated by the career decline

http://151.boxrec.com/~martin/ratings_at_whr.php

[SportsRatings]

The all-time WHR rankings are starting to shape up too, it looks like

[pugilisticspecialist]

The new all-time rankings put Larry Holmes behind David Tua.
Holmes is considered the best heavyweight of his era. Tua isn't even considered a top 5 heavyweight of his.

[margaret thatcher]

The new all-time rankings put Larry Holmes behind David Tua.
Holmes is considered the best heavyweight of his era. Tua isn't even considered a top 5 heavyweight of his.

Are the rankings overall better or worse now in your opinion?

[computerrank]

Another option is to use:

- career top rating
- the mean of career top 5 defeated opponents
- and then take the mean of both

As I did now:

http://151.boxrec.com/~martin/ratings_at_whr.php

[pugilisticspecialist]

I would keep it simple and just have the all-time ratings be each fighter's career peak rating, relative to the ratings of other elite fighters on the date the career peak rating was achieved.

[computerrank]

I would keep it simple and just have the all-time ratings be each fighter's career peak rating, relative to the ratings of other elite fighters on the date the career peak rating was achieved.

I will run 3 options:

- test_ct (career top rating)
- test_avg_ct_top5 (average of career top rating and mean career top 5 defeated opponents)
- test_ct_2top5 (career top rating, at most double mean career top 5 defeated opponents)
- test_top5 (mean career top 5 defeated opponents)

and then you can compare all 4 options.

[computerrank]

I will run 3 options:

- test_ct (career top rating)
- test_avg_ct_top5 (average of career top rating and mean career top 5 defeated opponents)
- test_ct_2top5 (career top rating, at most double mean career top 5 defeated opponents)
- test_top5 (mean career top 5 defeated opponents)

and then you can compare all 4 options.

Ready now ...

http://151.boxrec.com/~martin/ratings_at_whr.php

[pugilisticspecialist]

I would keep it simple and just have the all-time ratings be each fighter's career peak rating, relative to the ratings of other elite fighters on the date the career peak rating was achieved.

I will run 3 options:

- test_ct (career top rating)
- test_avg_ct_top5 (average of career top rating and mean career top 5 defeated opponents)
- test_ct_2top5 (career top rating, at most double mean career top 5 defeated opponents)
- test_top5 (mean career top 5 defeated opponents)

and then you can compare all 4 options.

Looking at the all-time heavyweight top 20, "test_ct (career top rating)" looks best to me.
I would say "test_top5 (mean career top 5 defeated opponents)" looks the worst.
The other two look somewhere in between.

[computerrank]

Re the all time ratings...

Instinctively I would forget defeated opponents and concentrate on individual fighters own all-time peak but then you have to factor in the years they remained at a high level so a fighter who briefly peaked but then couldn’t sustain it didn’t appear too high. There has to be an appreciation of longevity and consistency in any all-time list. I’ll have a think.....

What about an average of;

1. A fighters all time peak
2. An average of a fighter’s best 6 years across all the fights that featured in those years. The years wouldn’t have to be consecutive.

The two measures could be weighted 50/50 or whatever split made the most sense to you.

I picked 6 years at random but there should be a way of finding out on average how long a fighter was typically at their peak years.

The only thing I haven’t thought of is what to do if a fighter wasn’t around for 6 years (that is fair and equitable to all fighters) perhaps repeat the worst year until 6 years were filled.

Edit: I wasn’t far off ....apparently it’s actually 7 years.

I added a test Cobwebcat's proposal, so you can compare 5 options:

- test_ct (career top rating)
- test_avg_ct_top7years (average of career top rating and average of 7 years with best average annual rating)
- test_avg_ct_top5 (average of career top rating and mean career top 5 defeated opponents)
- test_ct_2top5 (career top rating, at most double mean career top 5 defeated opponents)
- test_top5 (mean career top 5 defeated opponents)

http://151.boxrec.com/~martin/ratings_at_whr.php

[computerrank]

I added a test Cobwebcat's proposal, so you can compare 5 options:

- test_ct (career top rating)
- test_avg_ct_top7years (average of career top rating and average of 7 years with best average annual rating)
- test_avg_ct_top5 (average of career top rating and mean career top 5 defeated opponents)
- test_ct_2top5 (career top rating, at most double mean career top 5 defeated opponents)
- test_top5 (mean career top 5 defeated opponents)

http://151.boxrec.com/~martin/ratings_at_whr.php

Thanks. I quite like the results the additional proposal comes up with in that in every weight category's top 10 I can see the fighters I would expect and generally recognise only if in a slightly different order. I think after that any system will come up with the odd fighter that looks too high or too low.

There is one thing that is strange though...on the P4P there are no Heavyweights! The list looks OK otherwise. I wonder if the HW ratings have been downgraded as the order looks reasonable just that none of them make it to the P4P list.

I guess you have used a 50/50 split on the variables. I think thats reasonable but if there are any obvious anomalies inside the top 10 in any weight category I think it would be worth trialling different splits. As the peak will appear in both variables perhaps 43% peak 57% 7 year.

What do you think?

Split was 50/50.

I reduced the weight compensation for lower weight divisions in an additional test, with lower weight divisions less enhanced)::
- test_avg_ct_top7years_wp (from exponent 3 to exponent 2)

[computerrank]

Split was 50/50.

I reduced the weight compensation for lower weight divisions in an additional test, with lower weight divisions less enhanced)::
- test_avg_ct_top7years_wp (from exponent 3 to exponent 2)

I'd say that's the best yet but maybe try 43/57 to get some legends a bit higher hopefully...people will always look for Muhammad Ali first on any all time P4P list......he should be top 20. Moving from Number 5 to 34 will still be too much for people to take....even if it's right!

You must be onto something because I thought who is this Charley Burley at number 4 Welterweight? I looked him up and found this...

Charley Burley (September 6, 1917 – October 16, 1992) was an American boxer who fought as a welterweight and middleweight from 1936 to 1950. Archie Moore, the light-heavyweight champion who was defeated by Burley in a 1944 middleweight bout, was one of several fighters who called Burley the greatest fighter ever.

Eddie Futch, the great trainer, called Burley "the finest all-around fighter I ever saw."

Burley was named to the Ring Magazine's list of 100 greatest punchers of all time, elected to the Boxing Hall of Fame in 1983 and the International Boxing Hall of Fame in 1992.

Burley was ranked 39th on Ring Magazine's list of the 80 Best Fighters of the Last 80 Years.

The current system only has him 43rd all time Welterwight so WHR doing something right!

@Cobwebcat - It doesn't seem to work as you thought :-)

Here the test added for
- test_avg43_57_ct_top7years_wp (43 percent career top rating, 57 percent career best 7 years with average annual ratings)

[SportsRatings]

Re the all time ratings...

Instinctively I would forget defeated opponents and concentrate on individual fighters own all-time peak but then you have to factor in the years they remained at a high level so a fighter who briefly peaked but then couldn’t sustain it didn’t appear too high. There has to be an appreciation of longevity and consistency in any all-time list. I’ll have a think.....

What about an average of;

1. A fighters all time peak
2. An average of a fighter’s best 6 years across all the fights that featured in those years. The years wouldn’t have to be consecutive.

The two measures could be weighted 50/50 or whatever split made the most sense to you.

I picked 6 years at random but there should be a way of finding out on average how long a fighter was typically at their peak years.

The only thing I haven’t thought of is what to do if a fighter wasn’t around for 6 years (that is fair and equitable to all fighters) perhaps repeat the worst year until 6 years were filled.

Edit: I wasn’t far off ....apparently it’s actually 7 years.

I added a test Cobwebcat's proposal, so you can compare 5 options:

- test_ct (career top rating)
- test_avg_ct_top7years (average of career top rating and average of 7 years with best average annual rating)
- test_avg_ct_top5 (average of career top rating and mean career top 5 defeated opponents)
- test_ct_2top5 (career top rating, at most double mean career top 5 defeated opponents)
- test_top5 (mean career top 5 defeated opponents)

http://151.boxrec.com/~martin/ratings_at_whr.php

A few years ago I really worked on a new algorithm for my all-time rankings, before going crazy and putting it aside again for a while :)

Since I only track HWs, a lot of the issues came with where it put Wlad Klitschko. I used things similar to "peak rating" and "career longetivity" and also top 3 (or 5) opponents defeated. The latter really pushed Klitschko down, and it was a matter of how much weight to put on that, or how to handle it. The results of the 5 tests above come to roughly the same conclusions my tests did: he ranks anywhere from #1 to #12.

I started adding other tweaks and that's what drove me crazy. I intended to come back to a very streamlined system, using only a few (2-3) factors. I'm still intending to at some point. Maybe the discussions here will help me move foreward!

So I guess my only input would be: don't start crazy-tweaking the ATR with small modifications. The basic ideas should do the job, and there are always going to be people who appear to be out of place. I'll be interested in seeing where Klitschko lands.

[computerrank]

My favourite currently for WHR all times is

- career top rating
- reduced by a damping factor for the recent 15 years (from today bouts up to 15 years back), due to ratings overshooting for not decayed careers. Damping factor 10 (today bouts) down to 1 (bouts older than 15 years).

- test

http://151.boxrec.com/~martin/ratings_at_whr.php

[computerrank]

...
Where have you been hiding these?! These are by far the best test and beat the present system too.

I think you have arrived at great WHR for BOTH active and all time fighters now.

I say publish and be damned!

One slight query is I wonder why fighters from a long time ago aren’t featured on P4P or indeed at their own weight... for example....Jack Johnson Sam Longford Joe Gans? In every other way they’re perfect. Very old fighters wrongly penalised or just that they beat a lot of low rated fighters? They're the obvious omissions I can see. When you sort by current ranking the WHR is in single figures (4.15 for JJ) which makes them look very poor fighters indeed! Got to be something obvious going on there.

I tried:

- neglect all NWS news decisions and all draws (decision often draw, when bout not stopped) before 1920
- neglect all losses against doubty opponents not connected to the boxing community before 1920

These have only a small impact to the ratings:

Jack Johnson: all time rating 4.15 -> 4.43, career top rating 6.941 -> 7.985

Langford: all time rating 5.73 ->7.02, career top rating 7.506 -> 9.414

The old boxers simply had a lot of really weak results against weak opponents, maybe due to boxing too often, with unprofessional preparation, and under doubty circumstances.

But we cannot help for this.

[JCS]

Still can't buy this WHR for standard boxing rankings. Pure mathematical systems like this are built for sports/games where competitors compete consistently and often. Modern boxing does not fit the bill.

Joe Smith beat Jesse Hart. SD, I know.. but Hart still ranked comfortable above in WHR.

It's a great system for prediction and assisting in betting, but not for a standard ranking.

Why not just publish both versions and let viewers choose which one they want?

As far as All-Time goes.. the debate is wide open.

Just my two cents.

[SportsRatings]

...
Where have you been hiding these?! These are by far the best test and beat the present system too.

I think you have arrived at great WHR for BOTH active and all time fighters now.

I say publish and be damned!

One slight query is I wonder why fighters from a long time ago aren’t featured on P4P or indeed at their own weight... for example....Jack Johnson Sam Longford Joe Gans? In every other way they’re perfect. Very old fighters wrongly penalised or just that they beat a lot of low rated fighters? They're the obvious omissions I can see. When you sort by current ranking the WHR is in single figures (4.15 for JJ) which makes them look very poor fighters indeed! Got to be something obvious going on there.

I tried:

- neglect all NWS news decisions and all draws (decision often draw, when bout not stopped) before 1920
- neglect all losses against doubty opponents not connected to the boxing community before 1920

These have only a small impact to the ratings:

Jack Johnson: all time rating 4.15 -> 4.43, career top rating 6.941 -> 7.985

Langford: all time rating 5.73 ->7.02, career top rating 7.506 -> 9.414

The old boxers simply had a lot of really weak results against weak opponents, maybe due to boxing too often, with unprofessional preparation, and under doubty circumstances.

But we cannot help for this.

Have you tried an all-time rating where you just plain ignore all losses? In other words, make it all about accomplishment, and just "add up" all wins over their career. If a draw gains them points, add it too.

Instead of peak rating, it becomes aggregate rating. Fighters like Holyfield gain a lot from this-- he had ups and downs and maybe not a great 'peak'. While fighters like Marciano don't get much of a boost. Older fighters like Jack Johnson, who lost a lot here and there or lost points due to bad results should also move up, I would guess.

[JCS]

Still can't buy this WHR for standard boxing rankings. Pure mathematical systems like this are built for sports/games where competitors compete consistently and often. Modern boxing does not fit the bill.

Joe Smith beat Jesse Hart. SD, I know.. but Hart still ranked comfortable above in WHR.

It's a great system for prediction and assisting in betting, but not for a standard ranking.

Why not just publish both versions and let viewers choose which one they want?

As far as All-Time goes.. the debate is wide open.

Just my two cents.

A boxing system whereby the winner of the bout must jump the loser is ridiculous to me. It might be the way it’s always been done but it doesn’t make it right. If Fulham beat Man City in the cup tomorrow do we say they are the better team and promote them?

I know the argument is boxing is different because fighters don’t fight very often I do get that but if you have a system that produces a better prediction AND looks just as good on the eye as the current one why not use it?

Ruiz jumped above AJ after his win but hardly anyone believed him a better fighter he just fought better that one time as Fulham might. Sure enough the rematch was a wide point victory for AJ. No one fight should have such a huge impact as to promote a rank outsider (not meaning Ruiz here) above a fighter with an impeccable record who has an off night.

If the algorithm promotes the underdog above the loser that’s fine but not every time for the sake of it...If Federer loses in the first round to a no hoper the winner isn’t immediately ranked above him.

I don’t get the argument other that’s what people expect to see but that can’t be the only reason for doing it. The WHR are objectively better than the current system for present fighters and they look fine as a bonus. The all-time one can’t be measured so is subjective I admit.

I respect your view and you might be in the majority but I disagree. On your MMA rankings I always go straight to modified elo.

Publishing both rankings is a good idea but I get the impression that one of the reasons the WHR is being considered is its speed of calculation vs current system. I could be wrong though.

Soccer is not boxing. Neither is tennis. Those sports are a much better fit for a pure mathematical system.

Ruiz above AJ after the first fight was "right". It wasn't even a flukey win... I can't see another argument. Did I favor him in the rematch? Of course not. Who deserves to be ranked higher and who is better are two different questions

The only time I'd support usage of WHR in a "standard" ranking is a situation in which there is insufficient data... like a midwest freak show who's pounding out a nobody per month... or a former champion who hasn't fought in three years that needs a placement. Something like that.