BoxRec

Great Site.
Constantly looking for records including my own.
One question? Why are Canadian Boxers hyphenated Ie:italy/canada.
One example of error is Phil Lo Greco. He is down as Italian/Canadian.
Phil was born in Toronto according to your own records.
Does that mean that Floyd Mayweather should be African/US or
that the Golden Boy be Mexican/US.
If they are fighting and training out of this country they are CANADA, period.
Sorry if this upsets anyone. No insult is implied or meant.
Best regards and Happy New year.

wjw wrote:Great Site.
Constantly looking for records including my own.
One question? Why are Canadian Boxers hyphenated Ie:italy/canada.
One example of error is Phil Lo Greco. He is down as Italian/Canadian.
Phil was born in Toronto according to your own records.
Does that mean that Floyd Mayweather should be African/US or
that the Golden Boy be Mexican/US.
If they are fighting and training out of this country they are CANADA, period.
Sorry if this upsets anyone. No insult is implied or meant.
Best regards and Happy New year.

@wjw

This post regarding Canadian Boxers hyphenated Ie:italy/canada is not for this Boxrec Ratings thread, I guess ... so I will re-post it at the Record Queries forum ...

http://boxrec.com/forum/viewtopic.php?f=1&t=131739

computerrank wrote:

forestbob wrote:Hi from a first-time poster, some-time reader, curious and impressed with an evidently workable (and broadly sane) objective rating system.

Forgive me that my first post is to suggest that something appears to be going awry.

computerrank wrote: The release is corrected and is launched.

Sir, since your corrected release of the ratings of the 12th, or a few days thereafter, there appears to have been an accumulation error occurring. I see some ratings increasing daily, now having surpassed what they were before your correction of the 12th took away often 5%, occasionally more than 10%, of the previous values.

Also, regarding the old (pre-12th Dec) system:

computerrank wrote: old release:

- defeating an opponent with at least 5 wins gets up to 50 additional points
- the amount in reduced in relation to less than 5 wins in relation to full 5 wins
- the amount is reduced in relation to less rounds boxed compared with full to 12 rounds
- the amount is reduced in relation to the clearness of a decision on points compared with a clear win
- the amount is recuded in relation the rating already achieved before the bout - down to no additonal points for already 50 points achieved before the bout

It was my experience in following actual exmaples that the additional points were still being acquired by boxers with up to 100, not 50, points achieved before the bout. (With less certainty, I'm not sure the number of rounds scheduled for the bout was effecting any difference, either.)

I can't comment on the same regarding the new release as I've been trying to reckon on a cause for the apparent daily accumulation instead of playing with the arithmetic of early boxing careers.

Hope you resolve the issues (or, if none to resolve, can shed light on my misunderstandings) soon.

Regards

Forest Bob
p.s. Your phrase "in relation to", were it means a linear relation, is perhaps best expressed in non-technical English as "in direct proportion to".

I am a bit short of time these days - but I will come back to your message - I appreciate it, thanks a lot!

@forestbob

would you please give me a hint, which boxers' ratings are shifting up daily? Thanks!

I updated the additional point procedure in the Ratings Description - and added an example.

I updated to your suggested phrase - "in direct propotion to" .

Thanks again
Martin

@forestbob

I now see some boxers creeping up the ratings too ...

All are boxers, who got some missing opponent quality reduction recently.

The test server is down until Jan 3, I guess.

I will analyze this ...

@computerrank

I'd conceeded in a posting on the 28th that the ratings I'd previously noticed rising were not doing so anymore, but now you're finding some that still are?! Curious.

Unfortunately, I don't have day-by-day records. But what i'd noticed was that after the initial drop from the Dec 12th release, (which I don't think I saw until the 14th/15th) by the 20th there was a sizeable increase in some ratings and more again (it seemed to me) a day-or-two later, so when I saw further gains on the 23rd I posted--only to see all the numbers I'd been looking at stablize (+/- 1) from then on!

Examples of ratings as seen: late November; 14th/15th December; 23rd December
2051; 1923; 2267 De La Hoya ratings peak (after 1998-09-18 vs Chavez)
1863; 1781; 1938 Chavez ratings peak (after 1990-03-17 vs Taylor)
1319; 1220; 1495 Mosley after 2000-11-04 vs Antonio Diaz
1481; 1403; 1678 Whitaker after 1995-08-26 vs Gary Jacobs

Not to say that all boxers ratings have moved as much, but the above certainly have!

regards

forestbob

forestbob wrote:@computerrank

I'd conceeded in a posting on the 28th that the ratings I'd previously noticed rising were not doing so anymore, but now you're finding some that still are?! Curious.

Unfortunately, I don't have day-by-day records. But what i'd noticed was that after the initial drop from the Dec 12th release, (which I don't think I saw until the 14th/15th) by the 20th there was a sizeable increase in some ratings and more again (it seemed to me) a day-or-two later, so when I saw further gains on the 23rd I posted--only to see all the numbers I'd been looking at stablize (+/- 1) from then on!

Examples of ratings as seen: late November; 14th/15th December; 23rd December
2051; 1923; 2267 De La Hoya ratings peak (after 1998-09-18 vs Chavez)
1863; 1781; 1938 Chavez ratings peak (after 1990-03-17 vs Taylor)
1319; 1220; 1495 Mosley after 2000-11-04 vs Antonio Diaz
1481; 1403; 1678 Whitaker after 1995-08-26 vs Gary Jacobs

Not to say that all boxers ratings have moved as much, but the above certainly have!

regards

forestbob

forestbob,

there were corrections on:

2010-12-13 release o_13
2010-12-13 release o_14
2010-12-22 release o_15
2010-12-23 release o_16
2010-12-23 release o_17

So this should explain the changed values.

And release o_18 is scheduled for these days - the final one in this series, I hope.

Final correction release o_18 in place now ...

- boxers cannot get as many points from nowhere, the launch process was extented up to launch state 15 and the requirements were enforced (rating description is updated for details), see example
http://boxrec.com/list_bouts.php?human_ ... &cat=boxer
is back to #16 now (from#2)

- an old bug fixed in the missing opponent quality routine - it didn't reduce the ratings for a lot of boxers everywhere, even in the top ranks

marcianofan wrote:

computerrank wrote:

marcianofan wrote:Like Forestbob, I have noticed that at some point recently, many ratings have increased to slightly more than the pre-Dec 12 values. I am not inclined to try and understand why this happened like some others here, as I simply trust that the rankings are in good hands, but I use Boxrec ratings points as a reference point for compiling my own divisional rankings for my blog based on different criteria, and every time a fighter's historical ratings points change it throws off my consistency and makes me recompile everything, so I'm just wondering if the current criteria can be considered stable or not (ie are more significant updates to the algorithm anticipated in the near future)- and if not- when the final tweaks are anticipated so I can work around that date.

Thank you.

The test server is down at the moment.

So some final adjustments for the last rating program release are postponed to January, I fear.

Okay thanks. I'll work around it for the moment and monitor this thread for updates come January.

... final correction is launched ... and Ali is in front of Louis again ...

Mayweather Jr #42
Pacquiao #60

Hi Guys,

a while back I moaned about the "star" rating of female bouts. I think it was a unification bout between Kentikian and ??? featuring two undefeated fighters with 15 fights each got a 2 star rating. I clicked on the 5 star rating the other day and the implementation has gone too far in the other direction. There are five star bouts between mid ranked fighters with terrible records. I think this needs to be tweaked again.

cheers

conan

conan_the_cribber wrote:Hi Guys,

a while back I moaned about the "star" rating of female bouts. I think it was a unification bout between Kentikian and ??? featuring two undefeated fighters with 15 fights each got a 2 star rating. I clicked on the 5 star rating the other day and the implementation has gone too far in the other direction. There are five star bouts between mid ranked fighters with terrible records. I think this needs to be tweaked again.

cheers

conan

Are you referring to any in particular? All the women I see on the 5-star page are ranked #2 or above.. although I admit I didn't check every single one.

http://boxrec.com/list_bouts.php?human_ ... &cat=boxer

Some 6-3 boxer vs someone with six losses in a rematch, hardly seems 5 star in the remotest possible way.

conan

She's rated #1 in her division... WIBA champ.

JCS wrote:She's rated #1 in her division... WIBA champ.

Does the ranking matter (#1) or the rating points or both. Obviously this division sux like a black hole. I would've imgained that a fight between two lowly rated fighters would not result in a five star fight, regardless where they are ranked in their division.

conan_the_cribber wrote:

JCS wrote:She's rated #1 in her division... WIBA champ.

Does the ranking matter (#1) or the rating points or both. Obviously this division sux like a black hole. I would've imgained that a fight between two lowly rated fighters would not result in a five star fight, regardless where they are ranked in their division.

Its based on rating... but in all fairness, I don't think its a bad fight for 5-stars.. considering its women's boxing.

... this is the top link in this thread ... I always try hard to keep it updated ...

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

... exerpt:

Bout Star Ratings

All bouts are rated with 0 to 5 stars.

Men

* 5 stars = both opponents have 331 ratings points at least ~ best 100 boxers
* 4 stars = both opponents have 188 ratings points at least ~ best 300 boxers
* 3 stars = both opponents have 83 ratings points at least ~ best 900 boxers
* 2 stars = both opponents have 24 ratings points at least ~ best 2700 boxers
* 1 stars = both opponents have 1 rating point at least = rated boxers
* 0 stars = one opponents has not even 1 rating point at least = not rated boxers

Women

* 5 stars = both opponents have 42 ratings points at least ~ best 30 boxers
* 4 stars = both opponents have 28 ratings points at least ~ best 60 boxers
* 3 stars = both opponents have 14 ratings points at least ~ best 120 boxers
* 2 stars = both opponents have 4 ratings points at least ~ best 240 boxers
* 1 stars = both opponents have 1 rating point at least = rated boxers
* 0 stars = one opponents has not even 1 rating point at least = not rated boxers

Just wondering why Roy Jones is rated 10th at Light Heavy. He's lost his last two fights, one was a first round KO.

keithmoonhangover wrote:Just wondering why Roy Jones is rated 10th at Light Heavy. He's lost his last two fights, one was a first round KO.

You can see the rating development of every boxer bout by bout.

- select a boxer's record
- before the bout list itself you will see an control check section, directly below the label 'bouts'
- check the button 'ratings on'

Now the bout list will be extended and you will see some additional fields at the end of every line.

- box_pre = boxer's rating before the bout
- box_aft = boxer's rating after the bout
- opp_pre = opponent's rating before the bout
- opp_aft = opponent's rating after the bout

Have fun ...

computerrank wrote:

keithmoonhangover wrote:Just wondering why Roy Jones is rated 10th at Light Heavy. He's lost his last two fights, one was a first round KO.

You can see the rating development of every boxer bout by bout.

- select a boxer's record
- before the bout list itself you will see an control check section, directly below the label 'bouts'
- check the button 'ratings on'

Now the bout list will be extended and you will see some additional fields at the end of every line.

- box_pre = boxer's rating before the bout
- box_aft = boxer's rating after the bout
- opp_pre = opponent's rating before the bout
- opp_aft = opponent's rating after the bout

Have fun ...

That's interesting, but there must be a flaw in the system. Jones is not the 10th best Light Heavy in the world.

keithmoonhangover wrote: That's interesting, but there must be a flaw in the system. Jones is not the 10th best Light Heavy in the world.

He hasn't really lost to anyone bad enough to seriously damage his rating.

I implemented ratings based on a completely new approach on the test server.

I also implemented a special access to these test ratings on the test server.

Benchmarks:

- current Boxrec ratings 77.3 % correct predictions = right to false ratio = 3.4 : 1

- new approach ratings 83.4 % correct predictions = right to false ratio = 5 : 1

This is tremendous, I guess.

I like the approach very much, as:

- the ratings are the equilibrium ratings for the boxers regarding all bouts between connected opponents of the recent 10 years
- a boxer is connected to the community, when he had at least 1 close or won bout to another connected boxer
- the metric is the standard distribution
- the weights are time descending quadratically
- the ratings deliver a criteria for provisional ratings, found by optimisation, to be connected to at least 6 opponents
- I don't need a specific launch process any more

You might have a look at them.

http://boxrec.dyndns.org/~martin/rating ... t=50&sex=m

You can select division, sex and limit by php parameters and see:

- rank
- boxer
- boxer_id
- first_name
- last_name
- r = new rating
- prov = provisional rating = *
- conn1 = # opponents connected with by a close result or a win
- conn6 = # opponents connected with, who have at least 6 connections themselves
- r_br = current Boxrec rating

The boxer_ids in the list are linked to another procedure, which shows the bouts in order by the bout performances.

- boxer_id
- name
- r = new rating
- opp_id = opponent id
- name
- prov = provisional rating = *
- conn1 = # opponents connected with by a close result or a win
- conn6 = # opponents connected with, who have at least 6 connections themselves
- perf = performance in bout
- res = result = 1 ... 0
-- 1 = full win as KO, on points over 12 rounds
-- >0.5 and <1 = close win
-- 0.5 = draw
-- >0 and <0.5 = close loss
-- 0 = full loos as KO, on points over 12 rounds
- w = weight = today = 1, 1 years ago = 0.81, 2 years ago = 0.64 , 5 years ago = 0.25, 10 years ago = 0 == quadratic decay

I am very keen on getting your response?

I think the challenge now is determining and implementing a "strength" of connections.

The overrating of Japanese, South Africans, and Thai boxers is further exacerbated.

JCS wrote:I think the challenge now is determining and implementing a "strength" of connections.

The overrating of Japanese, South Africans, and Thai boxers is further exacerbated.

... guessed, you would answer first ...

This is not a new issue ... but the new approach gives a key to regard something like a "strenght" of connections:

- how many boxers are in a "neighborhood"
- find clusters of "strong" "inner" connections
- at what level is a cluster connected to other clusters
- how "strong" is a cluster

Looking at super featherweight.. how is Masao Nakamura ranked #4? His connections seem quite low compared to the surrounding boxers.

At featherweight, I see Oleh Yefimovych at #12... but I don't see Sofiane Takoucht (Recent fighter who beat him) anywhere in the top 50...?

Also how did Abraham get ahead of Dirrell?

JCS wrote:Looking at super featherweight.. how is Masao Nakamura ranked #4? His connections seem quite low compared to the surrounding boxers.

At featherweight, I see Oleh Yefimovych at #12... but I don't see Sofiane Takoucht (Recent fighter who beat him) anywhere in the top 50...?

Also how did Abraham get ahead of Dirrell?

Nakamura #4 is relatively low connected

Yes. But his rating is valid in terms of the ratings.

By optimization I found a lower limit of 6 connections for a valid rating. Bouts against boxers with less connections are weighed lower in proportion to the number of connections.

Also boxers are grounded to 0 by 1 default draw against an opponent with rating 0, which affects boxer with few opponents.

And also boxers are grounded to 0 by another default draw against an opponent with rating 0, with a weight of

- max(0,1-sum_over_all_bouts(1-result*weight)).

- if a boxer has only full wins -> weight = 1
- if a boxer has enough bouts with reduced win margin -> weight = 0

Oleh Yefimovych at #12... but I don't see Sofiane Takoucht, who beat him

Takoucht is #63.

It was nearly a draw, result = 0.53.

A recent winner is no longer rated above a loser in every case. The ratings are adjusted to an equilibrium of differences between rating and weighed bout performances. So other performances may overrule the last result. The gain in prediction quality mainly depends on this approach.

Also how did Abraham get ahead of Dirrell?

Same reason as above. Abraham has the stronger rest record.

... But I will try to find a way to consider the supposed issue of larger and smaller or seggregated boxing communities (Japan ...)