Ratings - please read before commenting - Archived

[mike1989]

by the way Martin I have a idea to brush the stars out all the time if interested write in private message ?

[mike1989]

back to Adonis Stevenson , with the writing you his rating should have been reduced as follows:
period-based combat with Tony Bellew...2013-11-30 2015-05-30 here best was Andrzej Fonfara 780/2-226=616 points .
more idim fight with Fonfara 2014-05-24...2015-11-24 here best Tommy Karpenisi 616/2-187=495 points,
the next decrease will be in the range 2015-09-11...2017-03-11.
I all correctly to nibble???)

[computerrank]

by the way Martin I have a idea to brush the stars out all the time if interested write in private message ?

Michael, please send your PM

[computerrank]

... now, I understand)...
but why not sustained 18 months? another question ...

After a rating reduction it makes a difference, at what date the next best opponent boxed.

an example:

current mode (restart time frame at time of best opponent):

a boxer boxer has a rating of 1000.

- on 2015-01-01 the best opponent from 2013-07-01 with rating 500 is out of time frame.
- next best bout if from 2013-08-01 with pre-bout rating of 300.
- rating reduction to 1000/2 + 300 = 800.
- time starts on 2013-07-01 with virtual opponent rating of 800/2 = 400.

- on 2015-02-01 the best opponent is out of time frame.
- no next best bout available.
- inactivity within 18 months.
- rating reduction 50% of last rating 800/2 = 400
- time starts on 2015-02-01 with virtual opponent rating of 400/2 = 200.

- on 2016-03-01 the boxer has a rating of 400 due to his inactivity of more than 18 months from 2013-08-01 to 2015-02-01!


your suggestion (fixed 18 month time frames):

a boxer boxer has a rating of 1000.

- on 2015-01-01 the best opponent from 2013-07-01 with rating 500 is out of time frame.
- next best bout if from 2013-08-01 with pre-bout rating of 300.
- rating reduction to 1000/2 + 300 = 800.
- time starts on 2015-01-01 with virtual opponent rating of 800/2 = 400.

- on 2016-03-01 the boxer has a rating of 800 although he had a time frame of inactivity of 31 months (18 months allowed)!

[computerrank]

.. why some soldiers 18 months aged while others do not??)...

examples with data please

[computerrank]

... more about Nicholas Walters I wrote for a draw???

I do not understand ...

[computerrank]

back to Adonis Stevenson , with the writing you his rating should have been reduced as follows:
period-based combat with Tony Bellew...2013-11-30 2015-05-30 here best was Andrzej Fonfara 780/2-226=616 points .
more idim fight with Fonfara 2014-05-24...2015-11-24 here best Tommy Karpenisi 616/2-187=495 points,
the next decrease will be in the range 2015-09-11...2017-03-11.
I all correctly to nibble???)

- the reduction on 2014-12-08 was for the Dawson bout from 2013-06-08 going out of scope.
- next best opponent was Bellew from 2013-11-30
- so the reduced rating was 780/2 + 367 = 390 + 367 = 757
- time frame started with Bellew bout on 2013-11-30

- Bellew bout goes out of scope on 2015-05-30
- there is no reduction on this date - this is a bug in the current version, I already have corrected, but not released
- there should be a reduction
- next best opponent was Fonfara from 2014-05-24
- so the reduced rating was 732/2 + 244 = 366 + 244 = 610
- time frame started with Fonfara bout on 2014-05-24

- this will impact the next reduction on 2015-11-24 too.

[mike1989]

... more about Nicholas Walters I wrote for a draw???

I do not understand ...

Nicholas Walters with Sosa was a draw because a draw is equal to victory then the starting point must be that the fight with Sosa ???

[mike1989]

back to Adonis Stevenson , with the writing you his rating should have been reduced as follows:
period-based combat with Tony Bellew...2013-11-30 2015-05-30 here best was Andrzej Fonfara 780/2-226=616 points .
more idim fight with Fonfara 2014-05-24...2015-11-24 here best Tommy Karpenisi 616/2-187=495 points,
the next decrease will be in the range 2015-09-11...2017-03-11.
I all correctly to nibble???)

- the reduction on 2014-12-08 was for the Dawson bout from 2013-06-08 going out of scope.
- next best opponent was Bellew from 2013-11-30
- so the reduced rating was 780/2 + 367 = 390 + 367 = 757
- time frame started with Bellew bout on 2013-11-30

- Bellew bout goes out of scope on 2015-05-30
- there is no reduction on this date - this is a bug in the current version, I already have corrected, but not released
- there should be a reduction
- next best opponent was Fonfara from 2014-05-24
- so the reduced rating was 732/2 + 244 = 366 + 244 = 610
- time frame started with Fonfara bout on 2014-05-24

- this will impact the next reduction on 2015-11-24 too.

the reduction on 2014-12-08 was for the Dawson bout from 2013-06-08 going out of scope.

[computerrank]

... Nicholas Walters with Sosa was a draw because a draw is equal to victory then the starting point must be that the fight with Sosa ???

Why should the time frame start with the Sosa bout?

We are looking into the current ratings with pre-bout ratings counting.

After the Donaire bout ran out of the 18 months window:

Marriaga had a pre-bout rating of 139 points.

Sosa had a pre-bout rating of 124 points.

So the Marriaga rating counts.

[mike1989]

... Nicholas Walters with Sosa was a draw because a draw is equal to victory then the starting point must be that the fight with Sosa ???

Why should the time frame start with the Sosa bout?

We are looking into the current ratings with pre-bout ratings counting.

After the Donaire bout ran out of the 18 months window:

Marriaga had a pre-bout rating of 139 points.

Sosa had a pre-bout rating of 124 points.

So the Marriaga rating counts.

agree )!..

[computerrank]

A new rating release has been launched.

The new ratings considerably improve the ranking precision measured by the corrrect winner prediction of the bouts.

The error ratio of the old ratings bout winner predictions was 26 percent higher than the error ratio of the new ratings. Old 22.5 percent errors - new 17.9 percent errors.

This is another contribution to Boxrec ratings objectivity.

- new additional point procedure depending on
-- opponent's connectedness to valid boxers
-- pre-bout rank of opponent
-- pre-bout rating ratio of opponent to boxer

- bug fixes within missing opponent quality and inactivity points reduction

[computerrank]

An implementation error in the inactivity and missing opponent quality procedure was fixed.

[computerrank]

Updated Additional Points Question

Report this post
Quote

Postby bat3bee » Sat Jan 14, 2017 5:37 pm
The additional points formula was recently updated. There are now 3 parts. 1 point anyway, 1 or 0 for an opponent being connected to strong results and up to an additional 258 points by opponents rank in division. What exactly does it mean by being connected to strong results and also can someone provide an example between two boxers to explain the formula?

=>

example for a recent bout:

2017-01-14 TKO Gervonta Davis 62➞196 Jose Pedraza 227➞132

regular points 95 for both
additional points 39 for Davis:

use this calculator to play with different scenarios:

http://backup.boxrec.com/~martin/calc.php

do NOT enter the boxer IDs

use 100 for perf_lim_rank_a and for perf_lim_rank_b

enter 62 for the pre bout rating of the winner = boxer a = Davis
enter 227 for the pre bout rating of the loser = boxer b = Pedraza

enter 1 for the category (connectedness) of boxer a and b both

enter 33 for the pre bout rank of boxer a
enter 7 for the pre bout rank of boxer 6

press calculate ...

A boxer is connected to strong opponents, if he is connected by a direct path of won or close bouts (only the last result for any pairing is considered) to the highest rated boxer in the current ratings. A close bout is a bout, where the loser won at least 1/3 of all rounds on the score cards or got a MD or SD or D.

[Manrae]

Updated Additional Points Question

Report this post
Quote

Postby bat3bee » Sat Jan 14, 2017 5:37 pm
The additional points formula was recently updated. There are now 3 parts. 1 point anyway, 1 or 0 for an opponent being connected to strong results and up to an additional 258 points by opponents rank in division. What exactly does it mean by being connected to strong results and also can someone provide an example between two boxers to explain the formula?

=>

example for a recent bout:

2017-01-14 TKO Gervonta Davis 62➞196 Jose Pedraza 227➞132

regular points 95 for both
additional points 39 for Davis:

use this calculator to play with different scenarios:

http://backup.boxrec.com/~martin/calc.php

do NOT enter the boxer IDs

use 100 for perf_lim_rank_a and for perf_lim_rank_b

enter 62 for the pre bout rating of the winner = boxer a = Davis
enter 227 for the pre bout rating of the loser = boxer b = Pedraza

enter 1 for the category (connectedness) of boxer a and b both

enter 33 for the pre bout rank of boxer a
enter 7 for the pre bout rank of boxer 6

press calculate ...

A boxer is connected to strong opponents, if he is connected by a direct path of won or close bouts (only the last result for any pairing is considered) to the highest rated boxer in the current ratings. A close bout is a bout, where the loser won at least 1/3 of all rounds on the score cards or got a MD or SD or D.

2 questions

What is perf_lim_rank_a & b?

As far as rank goes, how do you know the rank of past fighters prior to a match-up?

[computerrank]

2 questions

What is perf_lim_rank_a & b?

As far as rank goes, how do you know the rank of past fighters prior to a match-up?

1) perf_lim_rank is the rank of the last active boxer in a division with a rating of 10 or more points

2) the ranks of all boxers active at a bout date are tracked during every daily ratings program run from the beginnings in 1850 to today

[Manrae]

2 questions

What is perf_lim_rank_a & b?

As far as rank goes, how do you know the rank of past fighters prior to a match-up?

1) perf_lim_rank is the rank of the last active boxer in a division with a rating of 10 or more points

2) the ranks of all boxers active at a bout date are tracked during every daily ratings program run from the beginnings in 1850 to today

Ah, I see. As far as #2 goes... I'm guessing that's something only you have access to? LIke, If I look up bouts from the past (ex--Mayweather Jr v Brusales), there's nowhere on the site that says what they were ranked prior to their match up?

Also, I followed your instructions for the calculator (something I'd like to play around with a lot more) but for Pedraza, 10 rounds, the numbers actually INCREASED from -83 to -79 for 5 and 6 in the "score diff per judge". Can you explain that to me?

[computerrank]

Ah, I see. As far as #2 goes... I'm guessing that's something only you have access to? LIke, If I look up bouts from the past (ex--Mayweather Jr v Brusales), there's nowhere on the site that says what they were ranked prior to their match up?

Also, I followed your instructions for the calculator (something I'd like to play around with a lot more) but for Pedraza, 10 rounds, the numbers actually INCREASED from -83 to -79 for 5 and 6 in the "score diff per judge". Can you explain that to me?

These past ranks are not stored permanently in the database. Storing all kind of past data would blow up the database too much.

Regarding your calculator example - what was the pairing your tried? Please indicate the Boxrec IDs of the winning boxer and the losing boxer.

[Manrae]

Ah, I see. As far as #2 goes... I'm guessing that's something only you have access to? LIke, If I look up bouts from the past (ex--Mayweather Jr v Brusales), there's nowhere on the site that says what they were ranked prior to their match up?

Also, I followed your instructions for the calculator (something I'd like to play around with a lot more) but for Pedraza, 10 rounds, the numbers actually INCREASED from -83 to -79 for 5 and 6 in the "score diff per judge". Can you explain that to me?

These past ranks are not stored permanently in the database. Storing all kind of past data would blow up the database too much.

Regarding your calculator example - what was the pairing your tried? Please indicate the Boxrec IDs of the winning boxer and the losing boxer.

The pairing I used was the same you provided for your example a couple of posts ago...
Jose Pedraza (559909) and Gervanta Davis (643387)

I used the same inputs that you provided and left the boxers ID blank as you instructed

Following your instructions, did you get the same results?

[computerrank]

The pairing I used was the same you provided for your example a couple of posts ago...
Jose Pedraza (559909) and Gervanta Davis (643387)

I used the same inputs that you provided and left the boxers ID blank as you instructed

Following your instructions, did you get the same results?

Bingo, Manrae

You detected a weak point in my rating algorithm.

By Boxrec rules a winning boxer always has to be rated higher than a losing boxer.

The problem rises for a certain ratio of the pre-bout ratings.

I went over it and fixed it.

The corrected version will be released soon.

[Manrae]

The pairing I used was the same you provided for your example a couple of posts ago...
Jose Pedraza (559909) and Gervanta Davis (643387)

I used the same inputs that you provided and left the boxers ID blank as you instructed

Following your instructions, did you get the same results?

Bingo, Manrae

You detected a weak point in my rating algorithm.

By Boxrec rules a winning boxer always has to be rated higher than a losing boxer.

The problem rises for a certain ratio of the pre-bout ratings.

I went over it and fixed it.

The corrected version will be released soon.

Oh wow, I'm glad I could help

I'd like to play around with the calculator more for historical purposes but it seems impossible without knowing their rankings at the time. When I look at certain fight results and ratings changes I often wonder what the ratings change would be if the correct verdict was read

Ex) Whitaker/Chavez, Lewis/Holyfield 1, etc

I guess I'll never know :(

[computerrank]

Boxrec user Manrae gave me a hint to an anomaly in the ratings, where the loser
loses less points, when losing by a larger margin.

The anomaly only shows up when a winning boxer is set ahead of the loser by
rule, although his points won would not be sufficient.

I corrected the rating program and the new release is launched now.

The overall points level is down by some 2 percent. Only boxers losing
against relatively very inferior opponents may lose larger amounts of
points.

[Lennox]

The pre-fight ranking of the opponent can be found here www.premierboxingorganisation.com/heavyweight-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.

[computerrank]

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.

Bellew was just shifted to heavyweight and the site shows his cruiserweight points until the next ratings program run. His pre-bout rating at heavyweight was 195 points and rank 6.

[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?