Page 137 of 180
Re: Whole-History Ratings - P4P current ratings
Posted: 05 Oct 2019, 07:56
by Manrae
computerrank wrote: ↑04 Oct 2019, 12:19
Code: Select all
1 1 Middleweight 348759 Saul Alvarez MX 1637 1651
** 2 4 Welterweight 447121 Terence Crawford US 1489 1134
3 2 Middleweight 356831 Gennady Golovkin KZ 1422 1254
4 3 Lightweight 659771 Vasiliy Lomachenko UA 1301 1163
5 6 Welterweight 629465 Errol Spence Jr US 1249 992.2
** 6 16 Cruiserweight 659772 Oleksandr Usyk UA 972.9 495.8
7 13 Super Featherweight 553222 Miguel Berchelt MX 917.3 520.6
** 8 26 Super Middleweight 631178 Callum Smith UK 890.2 367.2
9 7 Featherweight 374624 Leo Santa Cruz MX 861.0 586.2
* 10 5 Welterweight 6129 Manny Pacquiao PH 837.5 1025
11 8 Light Heavyweight 505927 Sergey Kovalev RU 805.4 571.2
12 14 Welterweight 472133 Shawn Porter US 780.4 516.0
** 13 109 Heavyweight 489762 Andy Ruiz Jr US 754.9 121.0
14 17 Middleweight 432984 Daniel Jacobs US 709.9 477.1
15 21 Welterweight 429442 Danny Garcia US 671.7 392.0
** 16 62 Light Heavyweight 679709 Oleksandr Gvozdyk UA 660.8 203.5
** 17 34 Middleweight 471634 Jermall Charlo US 657.3 300.3
** 18 47 Light Heavyweight 703924 Dmitry Bivol RU 625.9 250.3
19 37 Heavyweight 468841 Deontay Wilder US 622.3 291.3
** 20 43 Super Middleweight 588468 Chris Eubank Jr UK 612.9 258.8
21 20 Featherweight 479775 Gary Allen Russell Jr US 609.9 413.8
22 29 Super Featherweight 643387 Gervonta Davis US 604.3 332.8
23 44 Heavyweight 659461 Anthony Joshua UK 604.0 256.7
24 15 Featherweight 512777 Josh Warrington UK 600.5 503.4
25 25 Super Flyweight 467843 Juan Francisco Estrada MX 599.3 368.8
26 28 Welterweight 541777 Yordenis Ugas CU 599.0 336.4
27 46 Cruiserweight 514537 Mairis Briedis LV 587.4 251.0
28 27 Light Welterweight 632104 Jose Carlos Ramirez US 578.8 362.0
29 32 Light Welterweight 611370 Regis Prograis US 559.5 313.5
30 33 Light Welterweight 725709 Josh Taylor UK 555.4 301.8
31 55 Heavyweight 479205 Tyson Fury UK 549.5 218.7
32 57 Light Middleweight 659924 Jaime Munguia MX 546.7 217.7
33 24 Super Bantamweight 602423 Emanuel Navarrete MX 544.0 372.7
* 34 9 Bantamweight 628407 Naoya Inoue JP 535.6 564.8
* 35 11 Welterweight 426053 Keith Thurman US 519.9 538.0
36 56 Lightweight 563751 Richard Commey GH 509.2 218.4
** 37 129 Super Featherweight 740705 Shavkatdzhon Rakhimov RU 506.5 107.9
38 30 Super Flyweight 528191 Wisaksil Wangek TH 501.4 322.2
* 39 12 Featherweight 499601 Carl Frampton UK 488.0 530.1
40 22 Super Bantamweight 542148 Rey Vargas MX 487.0 377.6
** 41 83 Heavyweight 318081 Alexander Povetkin RU 483.2 150.7
42 35 Light Middleweight 457231 Erislandy Lara US 482.5 294.7
43 79 Light Middleweight 536011 Julian Williams US 481.9 155.4
44 36 Middleweight 468433 Demetrius Andrade US 469.7 293.9
* 45 10 Welterweight 364679 Mikey Garcia US 464.2 555.1
46 61 Super Featherweight 367450 Andrew Cancio US 462.9 203.5
47 40 Light Heavyweight 503963 Eleider Alvarez CO 461.1 281.1
* 48 18 Bantamweight 611983 Luis Nery MX 443.3 458.7
49 42 Light Heavyweight 509666 Gilberto Ramirez MX 425.4 260.7
** 50 130 Heavyweight 569964 Dillian Whyte UK 421.6 106.6
** 51 148 Welterweight 499476 David Avanesyan RU 416.0 96.00
52 76 Light Middleweight 625960 Jarrett Hurd US 412.6 158.0
53 58 Super Bantamweight 549073 Daniel Roman US 403.9 213.3
54 41 Light Flyweight 692967 Ken Shiro JP 403.2 275.9
55 39 Bantamweight 48243 Nonito Donaire PH 390.0 284.1
** 56 245 Middleweight 556657 Michael Zerafa AU 381.0 59.41
57 49 Welterweight 314558 Amir Khan UK 379.7 238.8
* 58 19 Featherweight 629933 Oscar Valdez MX 374.3 420.2
59 63 Lightweight 776269 Teofimo Lopez US 373.1 189.6
** 60 214 Welterweight 738032 Kudratillo Abdukakhorov UZ 370.4 67.87
61 74 Welterweight 684792 Sergey Lipinets RU 369.9 159.7
** 62 260 Heavyweight 498837 Jarrell Miller US 367.3 56.80
63 60 Light Middleweight 468475 Jessie Vargas US 366.8 205.9
** 64 128 Middleweight 657422 Ryota Murata JP 364.5 108.7
** 65 183 Heavyweight 528949 Luis Ortiz CU 363.6 78.92
66 77 Light Heavyweight 610132 Jesse Hart US 360.7 157.2
** 67 137 Light Middleweight 574588 Tony Harrison US 359.7 99.98
68 122 Light Heavyweight 290691 Jean Pascal CA 358.8 111.3
69 45 Lightweight 459640 Lee Selby UK 356.4 256.4
70 125 Light Middleweight 625137 Brian Carlos Castano AR 353.7 110.5
71 71 Light Welterweight 432621 Viktor Postol UA 351.4 162.2
72 133 Light Heavyweight 497268 Badou Jack SE 347.8 102.9
73 68 Lightweight 651514 Luke Campbell UK 345.0 164.8
74 51 Light Welterweight 523667 Jose Zepeda US 343.4 230.8
75 59 Bantamweight 359270 Zolani Tete ZA 343.1 212.4
* 76 23 Super Flyweight 483786 Kazuto Ioka JP 342.5 377.4
77 90 Light Heavyweight 629463 Marcus Browne US 342.3 143.9
** 78 197 Heavyweight 511850 Kubrat Pulev BG 339.4 73.55
79 81 Lightweight 741718 Devin Haney US 333.3 152.7
80 89 Middleweight 691996 Sergiy Derevyanchenko UA 332.4 144.8
81 98 Lightweight 371746 Anthony Crolla UK 331.3 134.0
* 82 38 Light Middleweight 272717 Kell Brook UK 330.4 284.3
83 54 Super Featherweight 562946 Tevin Farmer US 327.9 219.6
84 82 Super Featherweight 140100 Miguel Roman MX 322.2 151.6
85 52 Light Middleweight 433135 Jermell Charlo US 318.8 226.1
86 70 Super Middleweight 482499 Billy Joe Saunders UK 315.6 163.3
87 99 Featherweight 665698 Can Xu CN 312.9 132.3
88 147 Light Middleweight 723287 Carlos Adames DO 312.4 96.13
89 75 Welterweight 455859 Adrien Broner US 312.3 159.0
** 90 180 Super Middleweight 545557 John Ryder UK 309.1 79.87
91 53 Flyweight 666339 Kosei Tanaka JP 304.7 223.3
92 48 Super Bantamweight 492989 Guillermo Rigondeaux CU 304.0 247.4
** 93 255 Light Welterweight 352193 Pablo Cesar Cano MX 303.4 57.28
94 117 Light Welterweight 570403 Maurice Hooker US 300.9 114.1
* 95 31 Super Flyweight 215599 Donnie Nietes PH 294.1 321.5
96 86 Light Middleweight 466535 Liam Smith UK 292.3 148.1
** 97 195 Super Middleweight 546589 Rocky Fielding UK 291.1 74.11
98 73 Minimumweight 383177 Chayaphon Moonsri TH 288.5 159.8
99 119 Middleweight 635000 Jeff Horn AU 287.6 112.2
100 64 Super Flyweight 613018 Khalid Yafai UK 286.5 186.4
All 3 posts look very interesting and extremely intriguing. I want to be the 1st person to say THANK YOU. Not because I agree or disagree with the changes, but because you're CONSTANTLY DEDICATED to TRYING new ideas and giving us more to consider and think about concerning boxing's historic past and future. Don't EVER let anyone tell you not to always try to do more, you have an incredible amount of diligence and determination, PLEASE know you are appreciated!
The only question I have is that it seems that defeating a boxer with a higher score no longer automatically places you above said fighter. For example, Ruiz and Joshua, or Horn and Zerafa. Is that a correct assumption?
Re: Whole-History Ratings - P4P current ratings
Posted: 05 Oct 2019, 17:49
by computerrank
Manrae wrote: ↑05 Oct 2019, 07:56
...
The only question I have is that it seems that defeating a boxer with a higher score no longer automatically places you above said fighter. For example, Ruiz and Joshua, or Horn and Zerafa. Is that a correct assumption?
Yes, within the Whole-History Ratings former and later bouts influence the rating of a boxer.
So a boxer with an upset win may need another confirming win of same quality to finally surpass an opponent in the rating.s
Re: Whole-History Ratings - P4P current ratings
Posted: 05 Oct 2019, 19:17
by Manrae
computerrank wrote: ↑05 Oct 2019, 17:49
Manrae wrote: ↑05 Oct 2019, 07:56
...
The only question I have is that it seems that defeating a boxer with a higher score no longer automatically places you above said fighter. For example, Ruiz and Joshua, or Horn and Zerafa. Is that a correct assumption?
Yes, within the Whole-History Ratings former and later bouts influence the rating of a boxer.
So a boxer with an upset win may need another confirming win of same quality to finally surpass an opponent in the rating.s
Wow, that has me soooo excited to see more, I can't wait!
Thus far, how do you feel about the new algorithm?
Re: Ratings - please read before commenting
Posted: 05 Oct 2019, 23:38
by JCS
Other than early career and journeymen...I don't think fighters fight often enough to support such a system .
Re: Ratings - please read before commenting
Posted: 06 Oct 2019, 03:06
by computerrank
JCS wrote: ↑05 Oct 2019, 23:38
Other than early career and journeymen...I don't think fighters fight often enough to support such a system .
And why, Jason?
Re: Ratings - please read before commenting
Posted: 06 Oct 2019, 11:05
by JCS
computerrank wrote: ↑06 Oct 2019, 03:06
JCS wrote: ↑05 Oct 2019, 23:38
Other than early career and journeymen...I don't think fighters fight often enough to support such a system .
And why, Jason?
Given the matchmaking and schedules of experienced fighters.. I just feel like the incremental ratings are more appropriate. If we lived in a world where the matchmaking made sense and fighters fought often like in the past, I think the whole career system is better suited. These purely mathematical systems need frequent, consistent data.
Re: Ratings - please read before commenting
Posted: 07 Oct 2019, 19:55
by Manrae
When can we expect the new ratings to be implemented?
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 06:35
by computerrank
Manrae wrote: ↑07 Oct 2019, 19:55
When can we expect the new ratings to be implemented?
Tests regarding the prediction quality are running.
Anyway, they will not be implemented in 2019.
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 07:56
by Manrae
computerrank wrote: ↑08 Oct 2019, 06:35
Manrae wrote: ↑07 Oct 2019, 19:55
When can we expect the new ratings to be implemented?
Tests regarding the prediction quality are running.
Anyway, they will not be implemented in 2019.
I see, that's too bad. In the mean time, do you mind posting the current welterweight results with the new system like you did with the heavyweights? I'm just curious
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 09:34
by computerrank
Manrae wrote: ↑08 Oct 2019, 07:56
computerrank wrote: ↑08 Oct 2019, 06:35
Tests regarding the prediction quality are running.
Anyway, they will not be implemented in 2019.
I see, that's too bad. In the mean time, do you mind posting the current welterweight results with the new system like you did with the heavyweights? I'm just curious
I will provide a link to the Whole-History Ratings.
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 09:40
by computerrank
computerrank wrote: ↑08 Oct 2019, 06:35
Tests regarding the prediction quality are running.
Anyway, they will not be implemented in 2019.
The raw prediction ratio for 5728 qualified bouts in 2015 based on the year's end ratings of 2014 (none of the additional 16 other factors included) is:
- 72.2 % for the current ratings
- 76.8 % for the Whole-History ratings
So I am quite surprised. This solution seems to be strict, simple, fast and effective.
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 11:27
by computerrank
computerrank wrote: ↑08 Oct 2019, 09:34
Manrae wrote: ↑08 Oct 2019, 07:56
I see, that's too bad. In the mean time, do you mind posting the current welterweight results with the new system like you did with the heavyweights? I'm just curious
I will provide a link to the Whole-History Ratings.
http://151.boxrec.com/~martin/ratings.php
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 11:46
by BroughtonRulesRefuge
- RE Current welter ratings: Boxrec currently has both Porter and Danny Garcia over Thurman who beat both of them with only one loss to an all time great compared to their multiple losses. That's intuitively wrong, so further:
#2 Spence
#3 Manny
#4 Porter
#5 Danny Garcia
#6 Yordenis Ugas with 4 freaking losses
#7 Thurman
And by the current point totals it looks like their fights had no bearing on their prefight rankings, but I don't recall the prefight rankings.
The highlighted above is a collective grouping of the same era in the same point in time with 2 identical official boxing results, Split Decisions for Spence vs Porter and Manny vs Thurman, yet completely different point outcomes by your formula:
Manny loses significant points while Thurman gains significant points
https://boxrec.com/en/event/788305/2353535
Spence gains nominal points while Porter gains nominal points.
https://boxrec.com/en/event/792079/2374294
Porter with two losses, one against Thurman started vs Spence with 771.3 points whereas Thurman who was undefeated and actually beat Porter could only start vs Manny with a scrawny 483.4 points.
Probably I'll hear some blah-blah about Thurman being inactive for 2 years, but he did have a comeback rust remover in January 2019 before fighting Manny, so all I can add is
"Something is rotten in the state of Denmark."
Intuition is funny thing that we are born with at various levels. It makes drivers having a green light go sailing through an intersection without a care while grooving to their headphones, yet meanwhile just outside their distracted vision disaster comes barreling down on them, a drunk driver. The intuitive driver knows he best not press his luck and either jams the brakes or buries the throttle to escape, but it's amazing the number of people who suffer tragic consequences in thinking all is well because they have the legal right away.
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 12:38
by JCS
computerrank wrote: ↑08 Oct 2019, 09:40
computerrank wrote: ↑08 Oct 2019, 06:35
Tests regarding the prediction quality are running.
Anyway, they will not be implemented in 2019.
The raw prediction ratio for 5728 qualified bouts in 2015 based on the year's end ratings of 2014 (none of the additional 16 other factors included) is:
- 72.2 % for the current ratings
- 76.8 % for the Whole-History ratings
So I am quite surprised. This solution seems to be strict, simple, fast and effective.
But as we know.. going strictly with prediction makes for some unhappy viewers and 'odd' results.
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 13:56
by computerrank
JCS wrote: ↑08 Oct 2019, 12:38
computerrank wrote: ↑08 Oct 2019, 09:40
The raw prediction ratio for 5728 qualified bouts in 2015 based on the year's end ratings of 2014 (none of the additional 16 other factors included) is:
- 72.2 % for the current ratings
- 76.8 % for the Whole-History ratings
So I am quite surprised. This solution seems to be strict, simple, fast and effective.
But as we know.. going strictly with prediction makes for some unhappy viewers and 'odd' results.
I didn't expect Whole-History ratings would be better predictive. I just wanted to see, whether they are at same level or worse.
The Whole-History ratings consider the results from a wider perspective and are more stable regarding odd outlier results. And they are more stable comparing boxers with more or less bouts per year. And there is no need for additional points. The update ratings program uses the old rating values as a starter and the program run time is down to about 1/2 of the current time needed.
A nice side result is, the Whole-History ratings also provide an estimator for the rating uncertainty.
I just generalized the Whole-History model to use detailed Boxrec results (including draws and decisions on points, bouts weighted regarding rounds boxed and points conversion between weight divisions). No points reduction for inactivity and missing opponent quality needed any longer.
Top career ratings seem to be a good measure for the all time ratings now.
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 15:05
by JCS
computerrank wrote: ↑08 Oct 2019, 13:56
JCS wrote: ↑08 Oct 2019, 12:38
But as we know.. going strictly with prediction makes for some unhappy viewers and 'odd' results.
I didn't expect Whole-History ratings would be better predictive. I just wanted to see, whether they are at same level or worse.
The Whole-History ratings consider the results from a wider perspective and are more stable regarding odd outlier results. And they are more stable comparing boxers with more or less bouts per year. And there is no need for additional points. The update ratings program uses the old rating values as a starter and the program run time is down to about 1/2 of the current time needed.
A nice side result is, the Whole-History ratings also provide an estimator for the rating uncertainty.
I just generalized the Whole-History model to use detailed Boxrec results (including draws and decisions on points, bouts weighted regarding rounds boxed and points conversion between weight divisions). No points reduction for inactivity and missing opponent quality needed any longer.
Top career ratings seem to be a good measure for the all time ratings now.
It seems like the all-time ratings would benefit from such a system... that much I agree with.
Re: Ratings - please read before commenting
Posted: 08 Oct 2019, 16:23
by computerrank
BroughtonRulesRefuge wrote: ↑08 Oct 2019, 11:46
- RE Current welter ratings: Boxrec currently has both Porter and Danny Garcia over Thurman who beat both of them with only one loss to an all time great compared to their multiple losses. That's intuitively wrong, so further:
#2 Spence
#3 Manny
#4 Porter
#5 Danny Garcia
#6 Yordenis Ugas with 4 freaking losses
#7 Thurman
And by the current point totals it looks like their fights had no bearing on their prefight rankings, but I don't recall the prefight rankings.
The highlighted above is a collective grouping of the same era in the same point in time with 2 identical official boxing results, Split Decisions for Spence vs Porter and Manny vs Thurman, yet completely different point outcomes by your formula:
Manny loses significant points while Thurman gains significant points
https://boxrec.com/en/event/788305/2353535
Spence gains nominal points while Porter gains nominal points.
https://boxrec.com/en/event/792079/2374294
Porter with two losses, one against Thurman started vs Spence with 771.3 points whereas Thurman who was undefeated and actually beat Porter could only start vs Manny with a scrawny 483.4 points.
Probably I'll hear some blah-blah about Thurman being inactive for 2 years, but he did have a comeback rust remover in January 2019 before fighting Manny, so all I can add is
"Something is rotten in the state of Denmark."
Intuition is funny thing that we are born with at various levels. It makes drivers having a green light go sailing through an intersection without a care while grooving to their headphones, yet meanwhile just outside their distracted vision disaster comes barreling down on them, a drunk driver. The intuitive driver knows he best not press his luck and either jams the brakes or buries the throttle to escape, but it's amazing the number of people who suffer tragic consequences in thinking all is well because they have the legal right away.
You already mentioned the 2 points:
- Thurmann lost half of his points due to his inactivity
- The loss of Thurman was closer so he won points and Pacquiao lost some less due to his additional points
- The loss of Porter was wider to he won less points and Spence won some too due to his additional points
Re: Ratings - please read before commenting
Posted: 09 Oct 2019, 00:45
by Manrae
Thanks!
Re: Ratings - please read before commenting
Posted: 09 Oct 2019, 06:59
by Manrae
Kell Brook as #1 at 154 will be controversial
Re: Ratings - please read before commenting
Posted: 09 Oct 2019, 11:25
by JCS
Manrae wrote: ↑09 Oct 2019, 06:59
Kell Brook as #1 at 154 will be controversial
Whoa, yeah that's bad... and it is probably because Zerafa beat Horn afterward.
I told you guys that this isn't the way to go...
Re: Ratings - please read before commenting
Posted: 09 Oct 2019, 13:37
by JCS
Cobwebcat wrote: ↑09 Oct 2019, 13:32
Manrae wrote: ↑09 Oct 2019, 06:59
Kell Brook as #1 at 154 will be controversial
Considering the two fighters that beat him it’s not that crazy.
It is, because he didn't do anything directly to earn it.. he earned it based on the achievements of those he fought, after he fought them. If Zerafa didn't beat Horn just recently, Brook is not #1.
Re: Ratings - please read before commenting
Posted: 09 Oct 2019, 14:08
by computerrank
Cobwebcat wrote: ↑09 Oct 2019, 13:42
JCS wrote: ↑09 Oct 2019, 13:37
It is, because he didn't do anything directly to earn it.. he earned it based on the achievements of those he fought, after he fought them. If Zerafa didn't beat Horn just recently, Brook is not #1.
That’s as maybe. All I’m saying is I wouldn’t be surprised if Brook was actually the best in the division however he got there.
I will test this.
Re: Ratings - please read before commenting
Posted: 09 Oct 2019, 14:16
by computerrank
Cobwebcat wrote: ↑09 Oct 2019, 13:37
...
The system is virtually made for an all-time peak. The fact that it has better predictability for current ratings, by quite a margin, is remarkable.
Has any system you have tried in the past improved on 76.8%?
Not for the raw ratings based on rating points only.
Boxrec currently uses a prediction system with 16 additional factors. This yields around 86% percent.
Re: Ratings - please read before commenting
Posted: 09 Oct 2019, 15:34
by computerrank
Cobwebcat wrote: ↑09 Oct 2019, 15:21
computerrank wrote: ↑09 Oct 2019, 14:16
Not for the raw ratings based on rating points only.
Boxrec currently uses a prediction system with 16 additional factors. This yields around 86% percent.
So would you expect Whole History to beat the current system once the additional factors are added?
Also is the current system the one with the best predictability ever or did you have to sacrifice some predictability to keep people happy such as the winner always jumps above the loser?
Current system is best raw predictability. Setting a winner above the loser doesn't influence the predictability in a measurable manner.
Yes, I expect the Whole-History ratings to top the current ratings, when the additional factors are included.
Re: Ratings - please read before commenting
Posted: 09 Oct 2019, 17:42
by computerrank
computerrank wrote: ↑09 Oct 2019, 14:08
Cobwebcat wrote: ↑09 Oct 2019, 13:42
That’s as maybe. All I’m saying is I wouldn’t be surprised if Brook was actually the best in the division however he got there.
I will test this.
Yes, Brook would be #2 only, if Zerafa wouldn't have defeated Horn.