Ratings - please read before commenting - Archived

[computerrank]

Martin, I don't even pretend to know how the computerised rankings work but having Bellew the top three is just not right..maybe there's something in the system that can/should be readjusted/fine tuned?

Greg,

he was #2 at cruiserweight and took 70% of his points to heavyweight as #6 before the bout. He TKOed #10 Haye, The other mess is, heavyweight currently is one of the weakest divisions by competition. Regarding the number of active boxers the level should be much higher, but there are too few competitive bouts.

No heavyweight in the top 25 at the moment ... Joshua is #38 - I do not remember something like this.

[greg]

Martin, I don't even pretend to know how the computerised rankings work but having Bellew the top three is just not right..maybe there's something in the system that can/should be readjusted/fine tuned?

Greg,

he was #2 at cruiserweight and took 70% of his points to heavyweight as #6 before the bout. He TKOed #10 Haye, The other mess is, heavyweight currently is one of the weakest divisions by competition. Regarding the number of active boxers the level should be much higher, but there are too few competitive bouts.

No heavyweight in the top 25 at the moment ... Joshua is #38 - I do not remember something like this.

..I'm sure I'm not qualified to debate this with you, however I'm just curious how you came to these 70% and if that's not being too generous/high? thanks in advance..

[computerrank]

..I'm sure I'm not qualified to debate this with you, however I'm just curious how you came to these 70% and if that's not being too generous/high? thanks in advance..

The ratio for the division change is

upper weight limit of old division divided by upper weight limit of new division multiplied by itself.

This rule gave the best prediction ratio of the winners in a bout.

As heavyweight has no upper weight limit and as the heavyweights got heavier an heavier over time I determined the best fits with time,

Currently 240 lbs is the best fit (it was 190 lbs at time of Joe Louis).

With this the ratings ratio for a move from cruiserweight to heavyweight currently is:

(200 / 240) * (200 /240) = 0.833 * 8.833 = 0.694

[Lennox]

Your still missing the point. You don't understand. The formula you have is no good. Bellew is not the 3rd best heavyweight at the moment.

[Manrae]

The pre-fight ranking of the opponent can be found here http://www.premierboxingorganisation.co ... weight-200. A W just indicates a win against a fighter that was not ranked in the top 100 at the time of the fight.

Different method of ranking but much more accurate than Boxrec ratings that have hailed Tony Bellew as the second best heavyweight.All at Boxrec should feel totally ashamed of your ranking method.

No offense but, where's your updated rankings to compare and contrast?

[Lennox]

The pre-fight ranking of the opponent can be found here http://www.premierboxingorganisation.co ... weight-200. A W just indicates a win against a fighter that was not ranked in the top 100 at the time of the fight.

Different method of ranking but much more accurate than Boxrec ratings that have hailed Tony Bellew as the second best heavyweight.All at Boxrec should feel totally ashamed of your ranking method.

No offense but, where's your updated rankings to compare and contrast?

PBO ratings are updated monthly. Those are at 1st March. In the live system. Haye is down from 17 to 23. Bellew no change at 4 cruiserweight.

[JCS]

If you WERE to rank Bellew at HW, where would he be?

[Lennox]

JCS - He would be ranked 8th at Heavyweight fractionally ahead of Bermane Stiverne.

[candyslim]

Well I read it through first as instructed and I must say it's all a bit much for my hopelessly inadequate little brain. Obviously no formula is ever going to be perfect, I understand that, and I can fully understand that a fighter with a losing record fighting consistently high quality opposition can rate way higher than a fighter with a long unbeaten record over extremely modest opposition - I get that too.

I do find it incongruous though, that a boxer such as Roman Horvath who has a less than spectacular record of 0 and 24 rates 32 places above Rubin Williams at 29-29-1

Despite this sort of apparent anomaly, I rate Box-rec highly and would trust their ratings over those of any governing body.

[computerrank]

Well I read it through first as instructed and I must say it's all a bit much for my hopelessly inadequate little brain. Obviously no formula is ever going to be perfect, I understand that, and I can fully understand that a fighter with a losing record fighting consistently high quality opposition can rate way higher than a fighter with a long unbeaten record over extremely modest opposition - I get that too.

I do find it incongruous though, that a boxer such as Roman Horvath who has a less than spectacular record of 0 and 24 rates 32 places above Rubin Williams at 29-29-1

Despite this sort of apparent anomaly, I rate Box-rec highly and would trust their ratings over those of any governing body.

@candyslim
- Roman Horvath never had points.
- Rubin Williams was a successfull boxer with up to 49 points. In 2008 he lost to Green and then he gave his points away in a row of 27 all lost bouts. His last rest of points he lost in his last bout 2016. Now he is at plain 0 too.

[computerrank]

..I'm sure I'm not qualified to debate this with you, however I'm just curious how you came to these 70% and if that's not being too generous/high? thanks in advance..

The ratio for the division change is

upper weight limit of old division divided by upper weight limit of new division multiplied by itself.

This rule gave the best prediction ratio of the winners in a bout.

...

My ratings optimation tests for the weight division relation are from 2005, I found. The rating formula has developed a lot since then. So I tested again.

In 2005 I tested a linear and a square relation. The square relation showed the best results.

Now I also tested a cube relation and found it even fits better. So I will use this relation in the next release.

1 point at cruiserweight will then be equivalent to (200/240)**3 = 0.8333**3 = 0.58 points at heavyweight ( 0.69 points in the current release).

Bellew would be ranked #6 at heavyweight with the new release.

[JCS]

Have you tried a power of 2.5?

In the MMA world, I've found a value around there to be best..

[candyslim]

Thanks for taking the trouble to explain that to me. The numbers make sense although I'm sure the average boxing fan when he looks at any ratings is almost exclusively concerned with ranking position rather than points allocated.

To me, the fact that a guy who has never found anyone he could beat after 24 fights, ranks 32 places above a guy who might be very poor now but was once capable of running up a better than decent win-streak, just doesn't sit right, whether they are mathematically equal in terms of the rating formula or not.

I'd call that an anomaly but no rating formula is ever going to be perfect, or even anywhere close.

[computerrank]

Thanks for taking the trouble to explain that to me. The numbers make sense although I'm sure the average boxing fan when he looks at any ratings is almost exclusively concerned with ranking position rather than points allocated.

To me, the fact that a guy who has never found anyone he could beat after 24 fights, ranks 32 places above a guy who might be very poor now but was once capable of running up a better than decent win-streak, just doesn't sit right, whether they are mathematically equal in terms of the rating formula or not.

I'd call that an anomaly but no rating formula is ever going to be perfect, or even anywhere close.

I can see your point. All boxers with plain 0 points should be on the same last rank.

In fact the ranks are correct internally as you can see by these links - both boxers are ranked at 856 / 1,268

http://boxrec.com/boxer/356500
http://boxrec.com/boxer/58878

But for some reason the division lists auto-increment these rank numbers, where they should not. See here:

http://boxrec.com/records/856

I will give a hint to the developers.

[computerrank]

Have you tried a power of 2.5?

In the MMA world, I've found a value around there to be best..

I found a maximum - very flat - at around 3.

2.0 -> prediction - 0.08 percent
2.5 -> prediction - 0.06 percent
3.0 -> prediction - 0.00 percent
3.5 -> prediction - 0.02 percent

There is an even more interesting observation. These overshooting ratings of Joe Louis and Marciano times dropped from 6000 points to 3000 points. I guess, defeating high rated opponents from lower division earned them too many points ,,,

[JCS]

Have you tried a power of 2.5?

In the MMA world, I've found a value around there to be best..

I found a maximum - very flat - at around 3.

2.0 -> prediction - 0.08 percent
2.5 -> prediction - 0.06 percent
3.0 -> prediction - 0.00 percent
3.5 -> prediction - 0.02 percent

There is an even more interesting observation. These overshooting ratings of Joe Louis and Marciano times dropped from 6000 points to 3000 points. I guess, defeating high rated opponents from lower division earned them too many points ,,,

Strange.. I wonder if the factor should be variable through time... not sure how though.

[Manrae]

..I'm sure I'm not qualified to debate this with you, however I'm just curious how you came to these 70% and if that's not being too generous/high? thanks in advance..

The ratio for the division change is

upper weight limit of old division divided by upper weight limit of new division multiplied by itself.

This rule gave the best prediction ratio of the winners in a bout.

...

My ratings optimation tests for the weight division relation are from 2005, I found. The rating formula has developed a lot since then. So I tested again.

In 2005 I tested a linear and a square relation. The square relation showed the best results.

Now I also tested a cube relation and found it even fits better. So I will use this relation in the next release.

1 point at cruiserweight will then be equivalent to (200/240)**3 = 0.8333**3 = 0.58 points at heavyweight ( 0.69 points in the current release).

Bellew would be ranked #6 at heavyweight with the new release.

I was wondering if you could also update this page

http://boxrec.com/media/index.php/BoxRe ... al_Ratings

With your new algorithms and 2016 ratings.

[computerrank]

The ratio for the division change is

upper weight limit of old division divided by upper weight limit of new division multiplied by itself.

This rule gave the best prediction ratio of the winners in a bout.

...

My ratings optimation tests for the weight division relation are from 2005, I found. The rating formula has developed a lot since then. So I tested again.

In 2005 I tested a linear and a square relation. The square relation showed the best results.

Now I also tested a cube relation and found it even fits better. So I will use this relation in the next release.

1 point at cruiserweight will then be equivalent to (200/240)**3 = 0.8333**3 = 0.58 points at heavyweight ( 0.69 points in the current release).

Bellew would be ranked #6 at heavyweight with the new release.

I was wondering if you could also update this page

http://boxrec.com/media/index.php/BoxRe ... al_Ratings

With your new algorithms and 2016 ratings.

The correction release is tested, but not yet launched.

[Manrae]

My ratings optimation tests for the weight division relation are from 2005, I found. The rating formula has developed a lot since then. So I tested again.

In 2005 I tested a linear and a square relation. The square relation showed the best results.

Now I also tested a cube relation and found it even fits better. So I will use this relation in the next release.

1 point at cruiserweight will then be equivalent to (200/240)**3 = 0.8333**3 = 0.58 points at heavyweight ( 0.69 points in the current release).

Bellew would be ranked #6 at heavyweight with the new release.

I was wondering if you could also update this page

http://boxrec.com/media/index.php/BoxRe ... al_Ratings

With your new algorithms and 2016 ratings.

The correction release is tested, but not yet launched.

Cool Cool

I'll also cross my fingers that it'll knock Ali out of the top 10 in 1980, that's a bit ridiculous

[computerrank]

I fixed another bug in the ratings program algorithm. Boxers lost too many points due to missing opponent quality - more than intended by the formula.

I calculated a re-optimation of the power rule for the ratings' weight division conversion. In my last optimation from 2005 I found a square relation for the weights. With the current ratings (which has deleveloped quite a bit ...) I now found a cube relation fits better (slightly better predicition quality).

A nice side effect was, that the overshooting ratings points of the Joe louis and Marciano times lowered from some 6000 points to some 3000 points.
I guess, with the cube rule their wins against high rated boxers stemming from much lower divisions now show less influence.

The overall points level is up by some 10 percent. Some boxers, who where affected by the unjustified points losses for missing opponent may regain
larger amounts of points.

[Manrae]

I fixed another bug in the ratings program algorithm. Boxers lost too many points due to missing opponent quality - more than intended by the formula.

I calculated a re-optimation of the power rule for the ratings' weight division conversion. In my last optimation from 2005 I found a square relation for the weights. With the current ratings (which has deleveloped quite a bit ...) I now found a cube relation fits better (slightly better predicition quality).

A nice side effect was, that the overshooting ratings points of the Joe louis and Marciano times lowered from some 6000 points to some 3000 points.
I guess, with the cube rule their wins against high rated boxers stemming from much lower divisions now show less influence.

The overall points level is up by some 10 percent. Some boxers, who where affected by the unjustified points losses for missing opponent may regain
larger amounts of points.

Rigondeaux is back on the top list!

Does the calculator you posted before correspond to the new algorithm?

[Manrae]

This seems to be the algorithm from before, I noticed Shane being rated higher than Mayweather before their bout

So this retconned the change made in January I believe it was, correct?

[computerrank]

This seems to be the algorithm from before, I noticed Shane being rated higher than Mayweather before their bout

So this retconned the change made in January I believe it was, correct?

I wasn't a revirement. With the new release the point deduction for missing opponent quality not only depends on the relative level of the best opponent within 18 months. It now additionally depends on the length of time the requirement was violated.

[Manrae]

This seems to be the algorithm from before, I noticed Shane being rated higher than Mayweather before their bout

So this retconned the change made in January I believe it was, correct?

I wasn't a revirement. With the new release the point deduction for missing opponent quality not only depends on the relative level of the best opponent within 18 months. It now additionally depends on the length of time the requirement was violated.

I'm sorry, I don't quite understand exactly, but these ratings look exactly like the old ratings before the last major change up

Either way, is there any way to adjust points received based on the "prime rating" of the opponent in their career?

For example
Fighter A had a rating of 1000 a few years ago, he's rated at 150 now, but is beaten by a lower ranked opponent B (maybe 75) and B's rating shoots up, even though he fought a guy WAY past his "prime rating" so to speak...

Can that be accounted for in any way?

[computerrank]

...
Either way, is there any way to adjust points received based on the "prime rating" of the opponent in their career?

For example
Fighter A had a rating of 1000 a few years ago, he's rated at 150 now, but is beaten by a lower ranked opponent B (maybe 75) and B's rating shoots up, even though he fought a guy WAY past his "prime rating" so to speak...

Can that be accounted for in any way?

I do not understand the point, I guess. Why should this boost opponent B?