BoxRec

Cobwebcat wrote:Computerrank: If Pacman retired today where would he rank in the all time P4P top 100? I suspect he might not be as high as some imagine?

about 1000 points and rank 35

@conan

here an p4p option based on the annual ratings and other weighting:

- current rating with weight 200%

multiplied by

sum of
- 1 year ago with weight 80%
...
- 4 years ago with weight 20%
- 5 years ago and older not regarded

This requires a high current rating and high resent annual ratings ... I like this one even more ...

That looks like a good list, martin.

End of the day there will ALWAYS be people complaining about p4p lists because they imagine that anyone who cannot agree with them is dense, and they will also argue with something like this even though it is based on indirect mathematical assumptions.

computerrank wrote:@conan

here an p4p option based on the annual ratings and other weighting:

- current rating with weight 200%

multiplied by

sum of
- 1 year ago with weight 80%
...
- 4 years ago with weight 20%
- 5 years ago and older not regarded

This requires a high current rating and high resent annual ratings ... I like this one even more ...

Code: Select all

+-------------+---------------+-------+
| first_name  | last_name     | p4p   |
+-------------+---------------+-------+
| Manny       | Pacquiao      | 2,241 |  1 ( 1)
| Floyd       | Mayweather Jr | 2,074 |  2 ( 2)
| Shane       | Mosley        | 963   |  3 ( 3)
| Juan Manuel | Marquez       | 827   |  4 (10)
| Wladimir    | Klitschko     | 823   |  5 ( 5)
| Chris       | John          | 578   |  6 (13)
| Chad        | Dawson        | 576   |  7 ( 4)
| Kelly       | Pavlik        | 552   |  8 ( 8)
| Nonito      | Donaire       | 530   |  9 (26)
| Celestino   | Caballero     | 525   | 10 (16)
| Hozumi      | Hasegawa      | 507   | 11 ( 6)
| Pongsaklek  | Wonjongkam    | 471   | 12 ( 9)
| Vitali      | Klitschko     | 369   | 13 ( 7)
| Antonio     | Margarito     | 347   | 14 (34)
| Paul        | Williams      | 336   | 15 (11)
| Vic         | Darchinyan    | 330   | 16 (22)
| Daisuke     | Naito         | 295   | 17 (18)
| Edgar       | Sosa          | 290   | 18 (15)
| Arthur      | Abraham       | 285   | 19 (14)
| Ricky       | Hatton        | 274   | 20 (41)
| Tomasz      | Adamek        | 266   | 21 ( -)
| Israel      | Vazquez       | 259   | 22 (65)
| David       | Haye          | 243   | 23 (42)
| Omar Andres | Narvaez       | 236   | 24 (19)
| Robert      | Guerrero      | 203   | 25 (20)
| Felix       | Sturm         | 202   | 26 (17)
| Ivan        | Calderon      | 201   | 27 (32)
| Denkaosan   | Kaovichit     | 176   | 28 (28)
| Nate        | Campbell      | 175   | 29 (49)
| Lucian      | Bute          | 163   | 30 (25)
| Rafael      | Marquez       | 159   | 31 (84)
+-------------+---------------+-------+

Hi Martin,

look p4p is not really my field of expertise, but I'll try to be constructive.

This list looks pretty solid. I'm surprised about the relative rankings of Caballero and Israel Vasquez, but that's got more to do with the original ranking than anything else.

What the list needs though, is a who beat who guard, otherwise it will be worthless. I saw in a previous iteration that Taylor was ranked above Abraham, despite the fact that Abraham absolutely starched him. That sort of mistake can't happen. If you have the facts, then you have to use them.

You didn't mention the weightings or algorithm, so I'm guessing. Using the current rating as an exclusion factor, is a sensible feature. This would help exclude someone near the top, who suddenly loses two in a row. Because of the weighting, the previous four years may push the fighter up, which would be unrealistic.

Getting back to who beat who. I would probably do one pass, to find the candidates, give them an initial number, then do a second pass, to make sure that no direct matchups have happened in the last year or in the last 2 matches. I know it's a bit of effort, but hey, if you want to do a good job, then you got to put the hard yards in.

cheers

conan

p.s. I see you have implemented something else. I find this list better than what is currently on the front page.

I guess your program needs to magnify recent fights (for current, not all-time) to help the accuracy of your rattings. Roy Jones hasn't been a top 20 light heavy since september 2004 .

Quick question:

Browsing in the 'titles' tab on the front page, I wanted to know which title to look under to see world title lineage... for some reason I can't find, for example, Muhammad Ali on the lineages.

MatthewS wrote:Quick question:

Browsing in the 'titles' tab on the front page, I wanted to know which title to look under to see world title lineage... for some reason I can't find, for example, Muhammad Ali on the lineages.

Most of the early world titles are subsumed under the text "World DIVISION Title", which was independent of sanctioning bodies and has no entry in the titles list.

Later WBA, WBC and IBF often had common title holders, who should be assigned to every bodies' list, which is also done only rarely.

You better raise this issue in the Record Queries & Updates forum, as is not related to the BoxRec Ratings.

Cobwebcat wrote:Computerrank: I'd like the all-time P4P to be less longevity biased but I do understand why this has to be the case. If the current ranking could also be combined with a "peak 5 years" I think you might have the best of both worlds maybe.

Conan: The front page matches the ratings page P4P so I'm puzzled (unless it's changed)

But the ratings didn't match what computerrank posted. That is why I said it wasn't tranparent what he implemented. Also the algorithm is still Martin's secret

conan_the_cribber wrote:

Cobwebcat wrote:Computerrank: I'd like the all-time P4P to be less longevity biased but I do understand why this has to be the case. If the current ranking could also be combined with a "peak 5 years" I think you might have the best of both worlds maybe.

Conan: The front page matches the ratings page P4P so I'm puzzled (unless it's changed)

But the ratings didn't match what computerrank posted. That is why I said it wasn't tranparent what he implemented. Also the algorithm is still Martin's secret

The posted current p4p ratings are not published. They are just options.

I think, they are better than the p4p ratings published (they are just plain the current ratings).

There is no secret behind - see my post before:
---------------------------
@conan

here an p4p option based on the annual ratings and other weighting:

- current rating with weight 200%

multiplied by

sum of
- 1 year ago with weight 80%
...
- 4 years ago with weight 20%
- 5 years ago and older not regarded

This requires a high current rating and high resent annual ratings ... I like this one even more ...
----------------------------------

which is short:

p4p rating = (current_rating * 2) * (annual_rating_2008 * 0.8 + annual_rating_2007 * 0.6 + annual_rating_2006 * 0.4 + annual_rating_2005 * 0.2)

computerrank wrote:

conan_the_cribber wrote:

Cobwebcat wrote:Computerrank: I'd like the all-time P4P to be less longevity biased but I do understand why this has to be the case. If the current ranking could also be combined with a "peak 5 years" I think you might have the best of both worlds maybe.

Conan: The front page matches the ratings page P4P so I'm puzzled (unless it's changed)

But the ratings didn't match what computerrank posted. That is why I said it wasn't tranparent what he implemented. Also the algorithm is still Martin's secret

The posted current p4p ratings are not published. They are just options.

I think, they are better than the p4p ratings published (they are just plain the current ratings).

There is no secret behind - see my post before:
---------------------------
@conan

here an p4p option based on the annual ratings and other weighting:

- current rating with weight 200%

multiplied by

sum of
- 1 year ago with weight 80%
...
- 4 years ago with weight 20%
- 5 years ago and older not regarded

This requires a high current rating and high resent annual ratings ... I like this one even more ...
----------------------------------

which is short:

p4p rating = (current_rating * 2) * (annual_rating_2008 * 0.8 + annual_rating_2007 * 0.6 + annual_rating_2006 * 0.4 + annual_rating_2005 * 0.2)

Experienced fighter A loses via KO to up and comer B

For the five years that count

A Current = 350, Previous years are (500, 500, 500, 600) = 700 + 400 + 300 + 200 + 120 = 1720

B Current = 650, Previous years are (300, 200, 100, 0) = 1300 + 240 + 120 + 40 = 1700

You see the problem……A still is ranked above B.

I can’t see any way of doing it without two parses.

conan

What would the P4P ratings be if each fighter sustained their typical rating without the up/down changes that are issued when moving divisions?

conan_the_cribber wrote:

computerrank wrote:...
which is short:

p4p rating = (current_rating * 2) * (annual_rating_2008 * 0.8 + annual_rating_2007 * 0.6 + annual_rating_2006 * 0.4 + annual_rating_2005 * 0.2)

Experienced fighter A loses via KO to up and comer B

For the five years that count

A Current = 350, Previous years are (500, 500, 500, 600) = 700 + 400 + 300 + 200 + 120 = 1720

B Current = 650, Previous years are (300, 200, 100, 0) = 1300 + 240 + 120 + 40 = 1700

You see the problem……A still is ranked above B.

I can’t see any way of doing it without two parses.

conan

... you overlooked there is a multiplication, not an addition.

A -> 700 * 1020 = 714,000
B -> 1300 * 600 = 780,000

But this is not my perspective. The intention was, to inlude recent achievements, not only the current rating.

A currently lower rated boxer A may have a higher p4p rating than a currently higher rated boxer B, caused by boxer A's recent achievements from the last years, with declining weight. A single loss, which may have a desastrous impact to the current rating, may nevertheless allow a superior p4p rating.

JCS wrote:What would the P4P ratings be if each fighter sustained their typical rating without the up/down changes that are issued when moving divisions?

Jason,

from my perspective a superior p4p boxer should be excellent at any weight he fights. Otherwise he shouldn't box there or is no longer a p4p candidate.

computerrank wrote:

conan_the_cribber wrote:

computerrank wrote:...
which is short:

p4p rating = (current_rating * 2) * (annual_rating_2008 * 0.8 + annual_rating_2007 * 0.6 + annual_rating_2006 * 0.4 + annual_rating_2005 * 0.2)

Experienced fighter A loses via KO to up and comer B

For the five years that count

A Current = 350, Previous years are (500, 500, 500, 600) = 700 + 400 + 300 + 200 + 120 = 1720

B Current = 650, Previous years are (300, 200, 100, 0) = 1300 + 240 + 120 + 40 = 1700

You see the problem……A still is ranked above B.

I can’t see any way of doing it without two parses.

conan

... you overlooked there is a multiplication, not an addition.

A -> 700 * 1020 = 714,000
B -> 1300 * 600 = 780,000

But this is not my perspective. The intention was, to inlude recent achievements, not only the current rating.

A currently lower rated boxer A may have a higher p4p rating than a currently higher rated boxer B, caused by boxer A's recent achievements from the last years, with declining weight. A single loss, which may have a desastrous impact to the current rating, may nevertheless allow a superior p4p rating.

Well we both missed something. :)

My point was not to reflect a comparison between let's say Pacquaio after a loss and Wlad. Or even within a weight class where two boxers have not met e.g. JM Marquez can stay above whoever in his weight class, because of his long career.

My point is where there is a direct equation that would be nullified by recent direct results. Mayweather must be above Marquez, Abraham has to be above Taylor, Ward has to be above Kessler etc.

conan

computerrank wrote:

JCS wrote:What would the P4P ratings be if each fighter sustained their typical rating without the up/down changes that are issued when moving divisions?

Jason,

from my perspective a superior p4p boxer should be excellent at any weight he fights. Otherwise he shouldn't box there or is no longer a p4p candidate.

I think you misunderstood me.

I wanted to see the Top 25 boxers sorted by ratings that are not altered by divisional changes.

Cobwebcat wrote:Not surprisingly people are saying that Pacquiao v Mayweather is the biggest fight for 30 years. But is that true according to the ratings?

Pacman 1961

Mayweather 1712

Total 3673


When was the last time a fight topped 3673?

My guess is a Hopkins fight. Might even be Hopkins/JoeC

#19 the last 30 years ... raw rating sum = 5282

From my perspective the bout performance is bound to the lower rated opponent's rating alone.

So you get a rank of #7 during the the last 30 years (raw rating Mayweather 2531):

20 highest rated bouts ever:

Any reason why Danny Green is now rated at #11 at LHW, and Roy Jones is still at #3?

Hopkins new rating has been entered and his fight was more recent.

jezzamundo wrote:Any reason why Danny Green is now rated at #11 at LHW, and Roy Jones is still at #3?

Hopkins new rating has been entered and his fight was more recent.

A boxer's rating points are not changed with entering a new result or with a division move, but with the next rating program run, at 9:30 every day.

Cobwebcat wrote:

computerrank wrote:20 highest rated bouts ever:

Code: Select all

+-------------------+------------+------------+-------------+-------------+-------------+-------+
| division          | date       | first_name | last_name   | first_name  | last_name   | r_min |
+-------------------+------------+------------+-------------+-------------+-------------+-------+
| Heavyweight       | 1955-09-21 | Rocky      | Marciano    | Archie      | Moore       | 7,296 | 
|                   | 1955-06-22 | Bobo       | Olson       | Archie      | Moore       | 7,188 | 
| Welterweight      | 1938-11-18 | Young      | Corbett III | Fred        | Apostoli    | 6,801 | 
| Heavyweight       | 1941-06-18 | Billy      | Conn        | Joe         | Louis       | 5,911 | 
| Light Heavyweight | 1952-12-17 | Archie     | Moore       | Joey        | Maxim       | 5,902 | 
| Middleweight      | 1958-03-25 | Carmen     | Basilio     | Sugar Ray   | Robinson    | 5,777 | 
| Middleweight      | 1962-10-23 | Gene       | Fullmer     | Dick        | Tiger       | 5,767 | 
| Middleweight      | 1959-12-04 | Gene       | Fullmer     | Spider      | Webb        | 5,456 | 
| Welterweight      | 1963-03-21 | Emile      | Griffith    | Luis Manuel | Rodriguez   | 5,402 | 
| Middleweight      | 1963-08-10 | Gene       | Fullmer     | Dick        | Tiger       | 5,365 | 
|                   | 1938-01-07 | Freddie    | Steele      | Fred        | Apostoli    | 5,295 | 
| Welterweight      | 1935-05-28 | Jimmy      | McLarnin    | Barney      | Ross        | 5,130 | 
| Welterweight      | 1949-07-11 | Kid        | Gavilan     | Sugar Ray   | Robinson    | 5,123 | 
|                   | 1948-01-13 | Ezzard     | Charles     | Archie      | Moore       | 5,122 | 
| Middleweight      | 1957-09-23 | Carmen     | Basilio     | Sugar Ray   | Robinson    | 5,115 | 
|                   | 1943-08-27 | Henry      | Armstrong   | Sugar Ray   | Robinson    | 4,984 | 
| Lightweight       | 1939-08-22 | Lou        | Ambers      | Henry       | Armstrong   | 4,903 | 
| Welterweight      | 1934-09-17 | Barney     | Ross        | Jimmy       | McLarnin    | 4,881 | 
| Welterweight      | 1963-06-08 | Emile      | Griffith    | Luis Manuel | Rodriguez   | 4,754 | 
| Welterweight      | 1942-07-31 | Sugar Ray  | Robinson    | Sammy       | Angott      | 4,744 | 
+-------------------+------------+------------+-------------+-------------+-------------+-------+

Sort of puts Pac v May in it's historical place! Not that big a deal!!

Wrong.

There is a bout inflation caused by the current algorithm. Every fight adds 10% points to the system, which then decays over time. However, back in this time frame, there were a lot more fights, which meant 10% was being compounded more rapidly than today. All the fighters back then had more absolute points than now. That has to be factored in, to make any relative judgement.

conan

@conan

This is wrong for 2 reasons:

there is no inflation by every bout
- normaly the rating sum is constant
- as already discussed earlier, up to additional 50 points my be rewarded, if the winner was rated lower or only a bit higher than the winner.

There were not more bouts in history than today (about 20000 in eyery of the last 3 years)

There were some years with values a bit higher, but not really remarkable.

Looking at the years with the 5 top bouts, you find:

computerrank wrote:@conan

This is wrong for 2 reasons:

there is no inflation by every bout
- normaly the rating sum is constant
- as already discussed earlier, up to additional 50 points my be rewarded, if the winner was rated lower or only a bit higher than the winner.

There were not more bouts in history than today (about 20000 in eyery of the last 3 years)

There were some years with values a bit higher, but not really remarkable.

Looking at the years with the 5 top bouts, you find:

What I meant Martin, is that the bouts are not zero-sum. Each bout adds up to 50 points in the system. This is counteracted by the decay factor, the bouts are worth less over time.

However, the world is different when you fight 6 or 8 or 10 times in a year and some of your opponents have fought 20 times! Each time they fought, more points jumped into the system, before being decayed. These points have a compounding effect. I beat somebody who's currently beniffting from all the extra points in the system, then I get a higher number too. Then if someone beats me, then they have profitted yet again.

The number of bouts per year doesn't matter, its the number of bouts that in one way or another, somewhere down the line, contribute to a boxers points. A better indicator would be bouts per boxer in those years (where boxer has at least three recorded bouts), then you'd see a correlation to the highest ranked bouts and the years where this statistic would be high.

That's what I mean by bout inflation.

conan