It is quite well possible to derive the resulting php rating numbers from the short description. The performance at bout time just is the opponent's BoxRec rating at bout time modified by the bout result.conan_the_cribber wrote: ↑17 Feb 2023, 05:32Firstly, it's a rough draft, I understand that. But as someone who in real life designs the biggest computing systems in Europe, I think you need to work on your specification skills. It is not possible from the description above to know how points a) through e) combine to generate the number you came up with.Here is a first approach for such a p4p rating:
a) finds the boxer's bout performances in the last 5 years
b) sum of all performances in the last 2 years
c) finds the best performance in the last 2 years
d) sum of all performances in the 3 years before with a cap of the best performance in the last 2 years (in order devaluate descending boxers)
e) winner above loser rule within the last 18 months for a division difference of at most 1 division
So going back to the drawing board for a second, tell me what's going wrong with the normal divisional ratings? If you're satisfied with them, then I'm surprised that they don't suffice for a p4p rating. However you did mention, that there is some divisional scaling (where you made an adjustment and sent the results to me recently). Does that mean the higher weight divisions require scaling down? How did you determine that? My guess is it's a side of effect of your desire for a predictability validated ratings system (our old dispute). Bigger fighters defeat smaller fighters disproportionally, I'm assuming that this is something that the underlying mathematical model does not account for. I ask you this, because in points a) b) c) and d) above you use this rating (I assume in an unfiltered way).
Secondly, I would like to think about the problem domain. What are we trying to automate here? Here are some rough domain thoughts.
1) Marquis fighters, those who's name comes up in p4p discussions, generally fight about twice a year these days.
2) Marquis fighters are near the top of their divisional ratings.
3) A division may have multiple Marquis fighters (Aj, Fury, Wilder OR Canelo Golovkin).
4) These Marquis fighters may not have met yet.
5) When two Marquis fighters in a division meet, and there is a result for one of these fighters, then immediately after the fight, the winner should ALWAYS be rated above the loser in the p4p ratings.
6) What should be the criteria for invalidating this direct result in point 5)? Well I guess it's either
6a) a truly stellar achievement by the direct result loser in a higher weight class. If Tarver beat Jones, then Tarver should be p4p higher. If Jones fought anyone in his division and won in his next fight, then Tarver should still ALWAYS be above Jones still. However, if in his next fight, Jones jumped two divisions and fought Ruiz and won, then I could make an argument that Jones is ahead of Tarver on the p4p basis because Tarver has yet to achieve such a feat. My gut feeling is Marquis fights in a higher weight division than normal, should be weighted more in p4p calculations.
6b) A very long time. Given the fight frequency, I think if both fighters are avoiding each other, yet both are still competing in similar divisions (max one weight class apart), then the direct result should stand. I think it's fair to say, that whatever Golovkin achieved in his own weight division in the four years since he fought Canelo in 2018, no-one would rank him above Canelo because Canelo did not lose in that time. So I would suggest that the 18 months offered above is too short for p4p.
6c) A very bad loss for the winner. Boxing is a cruel sport. So many boxers reach a point where they just get old. Should the winner have a loss against a weak opponent, then it's fair to say, that they're not a p4p prospect any more. So I'm thinking, that in addition to points a) through e) above, the nature of the losses also need to be accounted for.
7) In general beating someone on the p4p list, does not mean you inherit their spot. It is not a ladder competition. When Andy Ruiz beat AJ, he should not have been #1 on anyone's heavyweight list, as the fight only proved that a) Ruiz was better than AJ and b) that AJ was weaker than people previously thought and probably behind Tury and Wilder (none of the three having fought each other at the time).
8) The nature of the result should probably influence the p4p status. KO wins weigh more than PTS win, and a lopsided PTS win is worth more than a close PTS win or SD.
9) No-one should be considered for the p4p ratings who has not fought another Marquis fighter. Compiling a 40-0 record like Brian Nielson did in the 90s, should be worth nothing. I'm assuming the divisional ratings take care of this.
10) A fighter who has proven themselves superior in multiple Marquis fights should probably be rated above people with a single Marquis win.
Any further domain comments from enthusiasts? Once the domain is sort of specified we can deduce factors that are implementable.
So the rating i not from scratch at all. But the compiliation might be different for p4p purposes.