Ratings - please read before commenting - Archived

[JCS]

Why not just wait 10 days?

[mike1989]

describe how you got the current point
Mairis Briedis for the victory over Marco Hook 2017-04-01 ???

I think the error here is:
point before the fight Mairis Briedis vs. Marco Hook 202 against 283 after the fight 459 against 179 = 459-202 / 283 * 100% = 90.8%;
UD12 judges score: Eddie Pappoe 118-109 | Massimo Barrovecchio 116-111 | Guido Cavalleri 117-110

Example number 2:
Vasilty Lomachenko vs. Roman Martinez 2016-11-06: point before the fight 188 against 349; and point after the fight 383 against 214 = 383-188 / 349 * 100% = 53,9%;
KО 5

explain the logic of accrual point ???

[computerrank]

Ratings new release launched:

Some small improvements for the ratings.


(1) connected boxers:

One of the main issues of boxing ratings is to avoid or limit point wins for effective none-achievements.

With one of the recent releases I indroduced the concept of connected boxers, with all members connected by won or close fights to set top root opponents (Wladimir Klitschko and Jessica Chavez). Only close bouts against connected boxers are rewarded with extra points. This results in about 2,000 connected boxers in women boxing history and about 230,000 connected boxers in men boxing history.

But I noticed, matchmakes were used to match their boxers with connected boxers, whose defining wins were long past. And so their boxers could win undeserved points.

I now restricted the connected time period for boxers from lifetime to a time period beginning with the first defining result minus a lead_time and ending with the last defining result plus a lag time.

(2) dynamic bottom value for points reductions for inactivity or missing opponent quality

This together resulted in a small improvement of prediction quality
(~0.15 percent points).



# rel 36
#
# 2018-12-17 - Martin Reichert
#
# - a connected boxer now is not connected during his whole career, but
# -- he is now connected to the community of connected boxers only
within a limited time period
# -- this time period begins 18 month before his first close bout
against a connected opponent
# -- this time period ends 18 month after his last close bout against a
connected opponent
# -- this avoids point gains against opponents outside of their
connected period
#
# - points reduction for inactivity or missing opponent quality now has
# -- a dynamic bottom value of 10 percent of a boxers career top rating
# -- the bottom value was fixed to 10 points before
#
# - all time ratings
# -- all time ratings include points for current year now


There are few changes in the top ranks and the top ratings stay at about
the same level.

There are more changes and point losses at the lower levels, which is by
purpose. So there will be some complaints ... :-)

But this release change will reward the real-achievers and set back the
under-achievers and cheaters.

[computerrank]

Now the rating point changed. Can you show the middleweight all time ranking like this again? Thanks.

The new ratings include all time points for the current year ...

[mike1989]

describe how you got the current point
Mairis Briedis for the victory over Marco Hook 2017-04-01 ???

I think the error here is:
point before the fight Mairis Briedis vs. Marco Hook 202 against 283 after the fight 459 against 179 = 459-202 / 283 * 100% = 90.8%;
UD12 judges score: Eddie Pappoe 118-109 | Massimo Barrovecchio 116-111 | Guido Cavalleri 117-110

Example number 2:
Vasilty Lomachenko vs. Roman Martinez 2016-11-06: point before the fight 188 against 349; and point after the fight 383 against 214 = 383-188 / 349 * 100% = 53,9%;
KО 5

explain the logic of accrual point ???

why BoxRec's Annual Ratings: Cruiserweight Annuals
Mairis Briedis at the end of 2017,
386 points and in fact 419 points ???

[computerrank]

why BoxRec's Annual Ratings: Cruiserweight Annuals
Mairis Briedis at the end of 2017,
386 points and in fact 419 points ???

The Boxrec ratings changed, and past ratings changed too.
The annual ratings will be updated in February 2019. I will wait until all 2018 results will be in.

[mike1989]

why BoxRec's Annual Ratings: Cruiserweight Annuals
Mairis Briedis at the end of 2017,
386 points and in fact 419 points ???

The Boxrec ratings changed, and past ratings changed too.
The annual ratings will be updated in February 2019. I will wait until all 2018 results will be in.

I understood Martin.
But what about how the rating of Briedis was formed in the battle with Marco Hook?

[computerrank]

why BoxRec's Annual Ratings: Cruiserweight Annuals
Mairis Briedis at the end of 2017,
386 points and in fact 419 points ???

The Boxrec ratings changed, and past ratings changed too.
The annual ratings will be updated in February 2019. I will wait until all 2018 results will be in.

I understood Martin.
But what about how the rating of Briedis was formed in the battle with Marco Hook?

Briedis not only got regular points - as many as Huck lost. Briedis got also points for defeating Cruiserweight #2 - as Huck's Cruiserweight rating points made him ranked 2nd.

[mike1989]

Martin check the downgrade Alexander Povetkin;
I think it does not take into account all the ranges:

2014-10-24 ... 2016-04-24 the strongest Perez 90 points;
2015-05-22 ... 2016-11-22 the strongest Mariusz Wach 4 points;
2015-11-04 ... 2017-05-04 the strongest Johann Duhaupas 113 points;
2016-12-17 ... 2018-06-17 Hummer 113 points.

???

[mike1989]

The Boxrec ratings changed, and past ratings changed too.
The annual ratings will be updated in February 2019. I will wait until all 2018 results will be in.

I understood Martin.
But what about how the rating of Briedis was formed in the battle with Marco Hook?

Briedis not only got regular points - as many as Huck lost. Briedis got also points for defeating Cruiserweight #2 - as Huck's Cruiserweight rating points made him ranked 2nd.

Martin write please the formula for this fight ???

[computerrank]

I understood Martin.
But what about how the rating of Briedis was formed in the battle with Marco Hook?

Briedis not only got regular points - as many as Huck lost. Briedis got also points for defeating Cruiserweight #2 - as Huck's Cruiserweight rating points made him ranked 2nd.

Martin write please the formula for this fight ???

earn = earn_f * v * (r_b*cd + (r_b-r_a)/(1+2*cd));

= 1/3 * 1 * (283*1 + (283-202)/(1+2*1))
= 1/3*(283+(81)/(3))
= 1/3*(283+27)
=1/3*310=103.3

Huck lost these regular points, Briedis won them. Huck lost 104 points ...


Additional points:

opponent is connected or not: con=(1 or 0), value of result v, clearness of decision cd, winner's rating r_a, opponents rating r_b, opponent's rank in division rank (#1 is equivalent to #2):

additional points winner = (1 + 4*con + 250/(rank-1) * v * cd) * (r_b+10)/(r_a+r_b+20):

con=1 for Huck. Factor 4 is enhanced to 5 in the current release. Huck was ranked #2. So we get:

=(1+5*1+250/(2-1)*1*1)*(283+10)/(202+283+20)
=(6+250/1)*(293)/(505)
=(6+250)*0.58
=148.5

So Briedis won 104+149= 253 points

There is a further small component in the new formula, which brings him another 4 points

So Briedis ends with 202+257=459 points.

[mike1989]

Briedis not only got regular points - as many as Huck lost. Briedis got also points for defeating Cruiserweight #2 - as Huck's Cruiserweight rating points made him ranked 2nd.

Martin write please the formula for this fight ???

earn = earn_f * v * (r_b*cd + (r_b-r_a)/(1+2*cd));

= 1/3 * 1 * (283*1 + (283-202)/(1+2*1))
= 1/3*(283+(81)/(3))
= 1/3*(283+27)
=1/3*310=103.3

Huck lost these regular points, Briedis won them. Huck lost 104 points ...


Additional points:

opponent is connected or not: con=(1 or 0), value of result v, clearness of decision cd, winner's rating r_a, opponents rating r_b, opponent's rank in division rank (#1 is equivalent to #2):

additional points winner = (1 + 4*con + 250/(rank-1) * v * cd) * (r_b+10)/(r_a+r_b+20):

con=1 for Huck. Factor 4 is enhanced to 5 in the current release. Huck was ranked #2. So we get:

=(1+5*1+250/(2-1)*1*1)*(283+10)/(202+283+20)
=(6+250/1)*(293)/(505)
=(6+250)*0.58
=148.5

So Briedis won 104+149= 253 points

There is a further small component in the new formula, which brings him another 4 points

So Briedis ends with 202+257=459 points.

Thank you, Martin!
and that on the brush rating Povetkin look at my post above ???

[computerrank]

Martin check the downgrade Alexander Povetkin;
I think it does not take into account all the ranges:

2014-10-24 ... 2016-04-24 the strongest Perez 90 points;
2015-05-22 ... 2016-11-22 the strongest Mariusz Wach 4 points;
2015-11-04 ... 2017-05-04 the strongest Johann Duhaupas 113 points;
2016-12-17 ... 2018-06-17 Hummer 113 points.

???

2014-10-24 ... 2016-04-24 the strongest Perez 90 points; rating down on 2016-04-24; new periods begins with Perez bout
2015-05-22 ... 2016-11-22 the strongest Wach 43; rating down to 268 on 2016-11-22; new period begins with Wach bout
2015-11-04 ... 2017-05-04 the strongest Johann Duhaupas 113 points; rating down to 274 on 2017-05-04; new period begins with Dehaupas bout
2016-12-17 ... 2018-06-17 the strongest Hammer 113 points; rating down to 283 on 2018-06-17; new period begins with Hammer bout

[SportsRatings]

Not sure if this is where to report, but a couple fighters both lost around 150 lbs between fights:

http://boxrec.com/en/boxer/738273

Tyrrell Herndon dropped from 240 lbs to 116 lbs in 2 months. His 12-08-2018 opponent lost 170 lbs but it took him 4 1/2 years...

[computerrank]

Not sure if this is where to report, but a couple fighters both lost around 150 lbs between fights:

http://boxrec.com/en/boxer/738273

Tyrrell Herndon dropped from 240 lbs to 116 lbs in 2 months. His 12-08-2018 opponent lost 170 lbs but it took him 4 1/2 years...

Thanks for the hint, SportsRatings!
I forwarded this to the editor.

[magicrp]

I find that sometimes Canelo's all time point is 892, sometimes (may be yesterday?) is about 928. Which one is correct? THANKS

[rellyquestbox]

After reading how fighters are ranked i still dont understand how Josh Taylor is ranked number 1. No major belt no top 10 opponent.

[computerrank]

I find that sometimes Canelo's all time point is 892, sometimes (may be yesterday?) is about 928. Which one is correct? THANKS

I should be 928. I will look into it.

[John]

After reading how fighters are ranked i still dont understand how Josh Taylor is ranked number 1. No major belt no top 10 opponent.

[computerrank]

After reading how fighters are ranked i still dont understand how Josh Taylor is ranked number 1. No major belt no top 10 opponent.

He defeated Postol as #3. Belts do not count, only Boxrec points and ranks.

[computerrank]

I find that sometimes Canelo's all time point is 892, sometimes (may be yesterday?) is about 928. Which one is correct? THANKS

I should be 928. I will look into it.

Due to corrections and additions for past bouts the rank of an opponent in a past bout may change. All times ratings points depend on the opponent rank at bout time. So by a small move of rating points the rank may change and the difference for a a change of the opponent rank from #1/#2 to #3 may reduce the all time rating of the winner by up to 125 points (other way round enhance it).

I will adapt the mechanism. If the opponent rank is #4, I will consider the rating points of #3, #4 and #5. And the the effective rank of #4 will be between #3 and #5, depending, how close the ratings points of #4 are to the ratings points of #3 and #5 in relation.

[JCS]

I wonder if the rating reduction for lesser opponents shouldn't be maxed at 33% for men like it is for women..

It feels wrong that Lomachenko lost 112 points before boxing Pedraza, considering he KO'd Linares in his previous bout, which was less then 7 months prior..

Is this something worth consideration, Martin? How would a change like this affect the ratings?

This feels like deja vu.. perhaps I asked this previously..

[computerrank]

I wonder if the rating reduction for lesser opponents shouldn't be maxed at 33% for men like it is for women..

It feels wrong that Lomachenko lost 112 points before boxing Pedraza, considering he KO'd Linares in his previous bout, which was less then 7 months prior..

Is this something worth consideration, Martin? How would a change like this affect the ratings?

This feels like deja vu.. perhaps I asked this previously..

There is a new ratings release in the pipeline, which is about 1 percent point better than the old one (81 percent correct prediction vs 80 percent correct predictions for the top boxers based on the rating points only. With an additional set of parameters the are about 86 percent correct predictions for the top boxers.

By the way, with the new ratings Lomachenko does not get a penalty in this case at all.

[computerrank]

The new release is in place now.

# - no further ranking points for current rating
#
# - points reduction for inactivity or missing opponent quality has
# -- a bottom value of 10 percent of a boxers career top rating adapted to division now
# -- first calculate inactivity or missing opponent quality then change division

Improvement for top 10 decile bouts result prediction (based on ratinings only)
83..04 percent instead of 81.53 percent

The combined prediction formula including further 16 parameters will improve from currently 84.5 percent further in the direction of 86 percent.

[JCS]

Martin... Are you saying the worst an inactivity or opposition penalty can affect a boxer is to take them down to 10% of their max rating?