BoxRec

You guys should be happy. When this penalty was first added, it was a flat rate 50% cut. I talked Martin into the sliding scale :). I recall thinking even that was too harsh... 35% or so with the sliding scale seemed to be the best option to me.

conan_the_cribber wrote:

JCS wrote:I understand what you mean conan.

I think how it works is.. everyday, an 18 month lookback is performed. If a fighter didn't fight someone within 50% of their current rating, they are penalized up to 50% based on the pre-fight rating of the best opponent.

In other words, if I have 650 points today, and the best opponent in the past 18 months was 300 points (46.1%), I am penalized 3.9%. IMO, this is pretty harsh.. I'd probably only penalize if the best opponent didn't reach 25 or 33%.

So the Klits will be continually penalised unless they fight each other? As the nearest fighter currently is Adamek with 45% of the lower Klitschko. NO algorithm can fit every case, but getting a points penalty for fighter the 3rd best fighter in your division is really odd. Same with Haye, getting penalised for fighting the best.

ComputerRank, are you listening? I think the time frame next to be extended and 50% is really high for Marquee fighters. hard to find opponents that fit that criteria.

conan

Neither W. Klitschko nor Haye boxed the best available opponents within 18 months.

A bout W. Klitschko vs Haye within these 18 months was available and would have seen Haye at 741 points and so above the 50% margin for W. Klitscho and himself.

A winner in a bout against an equivalant opponent can win 33% of the his current rating.

A winner in a bout against an opponent rated at 50% can win 11% of his current rating.

A winner in a bout against an opponent rated at 25% can withn no further points.

A boxer performing against an opponent better than 50% of his own rating within 18 months keeps his rating. The approved ratio so ist 1:2 at least.

A boxer performing against a best opponent rated 25% of his own rating within 18 months loses 25% of his rating down to 75%. The approved ratio so still is 1:3.

18 months is a long time ... and there are too many boxers missing eqivalent opposition over years, once having got a belt or a top ranking.

Maybe the Klitschkos will lose some points from their current top level.

But the ratings can only award an approved relation to the specific current level.

computerrank wrote:

conan_the_cribber wrote:

JCS wrote:I understand what you mean conan.

I think how it works is.. everyday, an 18 month lookback is performed. If a fighter didn't fight someone within 50% of their current rating, they are penalized up to 50% based on the pre-fight rating of the best opponent.

In other words, if I have 650 points today, and the best opponent in the past 18 months was 300 points (46.1%), I am penalized 3.9%. IMO, this is pretty harsh.. I'd probably only penalize if the best opponent didn't reach 25 or 33%.

So the Klits will be continually penalised unless they fight each other? As the nearest fighter currently is Adamek with 45% of the lower Klitschko. NO algorithm can fit every case, but getting a points penalty for fighter the 3rd best fighter in your division is really odd. Same with Haye, getting penalised for fighting the best.

ComputerRank, are you listening? I think the time frame next to be extended and 50% is really high for Marquee fighters. hard to find opponents that fit that criteria.

conan

Neither W. Klitschko nor Haye boxed the best available opponents within 18 months.

A bout W. Klitschko vs Haye within these 18 months was available and would have seen Haye at 741 points and so above the 50% margin for W. Klitscho and himself.

A winner in a bout against an equivalant opponent can win 33% of the his current rating.

A winner in a bout against an opponent rated at 50% can win 11% of his current rating.

A winner in a bout against an opponent rated at 25% can withn no further points.

A boxer performing against an opponent better than 50% of his own rating within 18 months keeps his rating. The approved ratio so ist 1:2 at least.

A boxer performing against a best opponent rated 25% of his own rating within 18 months loses 25% of his rating down to 75%. The approved ratio so still is 1:3.

18 months is a long time ... and there are too many boxers missing eqivalent opposition over years, once having got a belt or a top ranking.

Maybe the Klitschkos will lose some points from their current top level.

But the ratings can only award an approved relation to the specific current level.

Martin, the ratings can do whatever they like. And as long as it makes a sense and is defendable, then no-one has a right to complain. But right now, anyone the Klitchkos fight means they lose points, which is kind of absurd. Haye steps up and loses points. That's mad. Vitali fights Adamek and will lose points although Adamek is the best available fighter out there. That's mad.

I'm not sure how the rating work in detail. You really need to update it somewhere thoroughly. I'm not sure why you put the "winner of a bout against an opponent rated at 50%....", it is not clear how that influences the algorithm.

JCS mentions a second parse, which affects ratings because of inactivity against better or similar opponents. In principle I think this is OK, as you can argue, you have to keep your skill set up in order to maintain your rating. A little bit of decay is OK. But when someone dominates the division that is hardly correct. You maintain your dominance and keep losing points.

In my opinion, the period that needs to be taken into account is wrong for today's domain, where the best (marquee) fighters fight once or twice a year max. These fighters are the most visible in the ratings. Ratings lose credability, if people like Haye lose points for finally fighting Wlad.

conan

this may be a stupidly complex question... but how would the ratings look if actual scorecards were taken into account for cut/swelling stoppages, rather than being recorded as a stoppage?

@matt___s wrote:this may be a stupidly complex question... but how would the ratings look if actual scorecards were taken into account for cut/swelling stoppages, rather than being recorded as a stoppage?

There is no dedicated indicator for the criteria cut/swelling stoppages.

We only have the official decision TD, which indicates a stoppage for technical reasons. In this case the Boxrec Ratings evaluate the scorecards and assign a decision on points based on the number of rounds boxed.

computerrank wrote:

conan_the_cribber wrote:

JCS wrote:I understand what you mean conan.

I think how it works is.. everyday, an 18 month lookback is performed. If a fighter didn't fight someone within 50% of their current rating, they are penalized up to 50% based on the pre-fight rating of the best opponent.

In other words, if I have 650 points today, and the best opponent in the past 18 months was 300 points (46.1%), I am penalized 3.9%. IMO, this is pretty harsh.. I'd probably only penalize if the best opponent didn't reach 25 or 33%.

So the Klits will be continually penalised unless they fight each other? As the nearest fighter currently is Adamek with 45% of the lower Klitschko. NO algorithm can fit every case, but getting a points penalty for fighter the 3rd best fighter in your division is really odd. Same with Haye, getting penalised for fighting the best.

ComputerRank, are you listening? I think the time frame next to be extended and 50% is really high for Marquee fighters. hard to find opponents that fit that criteria.

conan

Neither W. Klitschko nor Haye boxed the best available opponents within 18 months.

A bout W. Klitschko vs Haye within these 18 months was available and would have seen Haye at 741 points and so above the 50% margin for W. Klitscho and himself.

A winner in a bout against an equivalant opponent can win 33% of the his current rating.

A winner in a bout against an opponent rated at 50% can win 11% of his current rating.

A winner in a bout against an opponent rated at 25% can withn no further points.

A boxer performing against an opponent better than 50% of his own rating within 18 months keeps his rating. The approved ratio so ist 1:2 at least.

A boxer performing against a best opponent rated 25% of his own rating within 18 months loses 25% of his rating down to 75%. The approved ratio so still is 1:3.

18 months is a long time ... and there are too many boxers missing eqivalent opposition over years, once having got a belt or a top ranking.

Maybe the Klitschkos will lose some points from their current top level.

But the ratings can only award an approved relation to the specific current level.

Martin,

Given this criteria, why did Vitali lose points between Johnson and Sosnowski fight? Vitali fought JC Gomez on 2009-03-21, when Gomez was above 50% of his ranking. He fought Sosnowski on 2010-05-29. That is 14 months later. Admittedly, Johnson was a lowly ranked contendor 20% of Vitali's ranking. But in the end Vitali won a shut out, every round on two cards. That confirms his dominance, rather than reduces it.

And it would be great, if there was a place, where the whole algorithm in reasonable detail is explained. Then we wouldn't have to keep asking specific questions.

conan

Boxer a KO boxer b, a has 1000 points, b has 500 points. launch state 4, v=1, cd=1,

earn= 0.33 * 1 * (500*1 + (500-1000)/(1+2*1)) = 111
r_a_new = 1000 + 111 = 1111
r_b_new = 500 - 111 = 389

That is taking from the link on page one of this thread. According to my calculations.

earn
= 0.33 * 1 * (576 + (576-1243)/3)
= 0.33 * (576 + (-667)/3)
= 0.33 * (576 - 222) // rounded
= 0.33 * (252)
= 84

However Vitali only profitted by 59 pts.

What did I get wrong or has the algorithm moved on?

conan_the_cribber wrote: ...

Martin,

Given this criteria, why did Vitali lose points between Johnson and Sosnowski fight? Vitali fought JC Gomez on 2009-03-21, when Gomez was above 50% of his ranking. He fought Sosnowski on 2010-05-29. That is 14 months later. Admittedly, Johnson was a lowly ranked contendor 20% of Vitali's ranking. But in the end Vitali won a shut out, every round on two cards. That confirms his dominance, rather than reduces it.

And it would be great, if there was a place, where the whole algorithm in reasonable detail is explained. Then we wouldn't have to keep asking specific questions.

conan

Gomez had a pre_bout rating of 572 points on 2009-03-21.

V. Klitschko had 1270 points after the bout against Sosnowksi, which were covered by Peter's pre-bout rating of 824 on 2008-10-11. But the coverage only lasted until 2010-04-11 (18 months).

After this date, the best result within 18 months was Gomez' pre-bout rating, which no longer covered V. Klitschkos's rating.

So V. Klitschkos rating was reduced.

I will add the reduction formula to the description.

conan_the_cribber wrote:

Boxer a KO boxer b, a has 1000 points, b has 500 points. launch state 4, v=1, cd=1,

earn= 0.33 * 1 * (500*1 + (500-1000)/(1+2*1)) = 111
r_a_new = 1000 + 111 = 1111
r_b_new = 500 - 111 = 389

That is taking from the link on page one of this thread. According to my calculations.

earn
= 0.33 * 1 * (576 + (576-1243)/3)
= 0.33 * (576 + (-667)/3)
= 0.33 * (576 - 222) // rounded
= 0.33 * (252)
= 84

However Vitali only profitted by 59 pts.

What did I get wrong or has the algorithm moved on?

V. Klitschko would even have got more points, as

=0.33 * (576 - 222)
=0.33 * 354
= 118

if it would have been covered by a confirming opponent rating within the last 18 months.

But ...

V. Klitschko had a rating of 1243 of the points before the bout. So the rating was immediatly reduced by some amount and the resulting improvement was only half the value expected.

The same thing already happend after the bout against Solis.

computerrank wrote:

conan_the_cribber wrote: ...

Martin,

Given this criteria, why did Vitali lose points between Johnson and Sosnowski fight? Vitali fought JC Gomez on 2009-03-21, when Gomez was above 50% of his ranking. He fought Sosnowski on 2010-05-29. That is 14 months later. Admittedly, Johnson was a lowly ranked contendor 20% of Vitali's ranking. But in the end Vitali won a shut out, every round on two cards. That confirms his dominance, rather than reduces it.

And it would be great, if there was a place, where the whole algorithm in reasonable detail is explained. Then we wouldn't have to keep asking specific questions.

conan

Gomez had a pre_bout rating of 572 points on 2009-03-21.

V. Klitschko had 1270 points after the bout against Sosnowksi, which were covered by Peter's pre-bout rating of 824 on 2008-10-11. But the coverage only lasted until 2010-04-11 (18 months).

After this date, the best result within 18 months was Gomez' pre-bout rating, which no longer covered V. Klitschkos's rating.

So V. Klitschkos rating was reduced.

I will add the reduction formula to the description.

Martin, I've said in the past, the ratings algorithm should reflect common sense i.e. the way people think in a domain. For example, it makes no sense in the Fifa rankings to make Luxemburg number one, if forever what reason they have a lucky win against current number one Holland.

This part of the algorithm makes no sense to someone involved in boxing. Vitali comes back and has a crap rating ang fights Peter, who was probably 2nd or 3rd at the time. Then he fights Gomez who was, as far as ratings points go, in the same level. Now because of these stellar victories, he has too many points afterwards and gets penalised for CORRECTLY taking on the best challengers and winning.

Inactivity against top opponents should be a factor in the algorithm, without a doubt. But it should be defined as "How long is it since you fought somebody at your level". Vitali fought Gomez, who was at his level, so he should have 18 months grace from then.

conan

computerrank wrote:

conan_the_cribber wrote:

Boxer a KO boxer b, a has 1000 points, b has 500 points. launch state 4, v=1, cd=1,

earn= 0.33 * 1 * (500*1 + (500-1000)/(1+2*1)) = 111
r_a_new = 1000 + 111 = 1111
r_b_new = 500 - 111 = 389

That is taking from the link on page one of this thread. According to my calculations.

earn
= 0.33 * 1 * (576 + (576-1243)/3)
= 0.33 * (576 + (-667)/3)
= 0.33 * (576 - 222) // rounded
= 0.33 * (252)
= 84

However Vitali only profitted by 59 pts.

What did I get wrong or has the algorithm moved on?

V. Klitschko would even have got more points, as

=0.33 * (576 - 222)
=0.33 * 354
= 118

if it would have been covered by a confirming opponent rating within the last 18 months.

But ...

V. Klitschko had a rating of 1243 of the points before the bout. So the rating was immediatly reduced by some amount and the resulting improvement was only half the value expected.

The same thing already happend after the bout against Solis.

Martin, again I think a tweak is needed. Common sense screams out "Wlad should not lose points for fighting Haye" and "Vitali shoudl not lose points for fighting Adamek". Both opponents were number three at the time of the fight. In other weight divisions, you can fight someone in a higher weight class, that also has points. In heavyweight, there is nowhere to go. This should be reflected in the reduction algorithm.

Thanks

conan

conan_the_cribber wrote:
Martin, I've said in the past, the ratings algorithm should reflect common sense i.e. the way people think in a domain. For example, it makes no sense in the Fifa rankings to make Luxemburg number one, if forever what reason they have a lucky win against current number one Holland.

This part of the algorithm makes no sense to someone involved in boxing. Vitali comes back and has a crap rating ang fights Peter, who was probably 2nd or 3rd at the time. Then he fights Gomez who was, as far as ratings points go, in the same level. Now because of these stellar victories, he has too many points afterwards and gets penalised for CORRECTLY taking on the best challengers and winning.

Inactivity against top opponents should be a factor in the algorithm, without a doubt. But it should be defined as "How long is it since you fought somebody at your level". Vitali fought Gomez, who was at his level, so he should have 18 months grace from then.

conan

It is common sense in professional boxing, that a winner should be rated above the loser. So in the Boxrec ratings the winner will be rated at 50 + x % of the loser's pre_bout rating at least.

The Boxrec ratings represent the level a boxer has actually fought the recent time (18 months). If there is no matching opponent to really challenge the boxer, there is no evidence, that he kept his level - and so he will lose rating points.

Gomez was at his level at this time - regarding the Boxrec ratings - Gomez wasn't at a level to grant him a later higher rating.

conan_the_cribber wrote:
Martin, again I think a tweak is needed. Common sense screams out "Wlad should not lose points for fighting Haye" and "Vitali shoudl not lose points for fighting Adamek". Both opponents were number three at the time of the fight. In other weight divisions, you can fight someone in a higher weight class, that also has points. In heavyweight, there is nowhere to go. This should be reflected in the reduction algorithm.

Thanks

conan

There is no evidence that the heavyweights suffer from this.

Maybe in other divisions boxers lose more points against opponents from higher divisions than they win against those from lower divisions - or maybe it is the other way round.

@matt___s wrote:Funny question I know -

RING magazine has its titles dished out when #1 and #2 fight, as we all know, my question is, would it be possible to tell me when divisional #1 and #2 on the Boxrec database have fought?

It's #1 vs #2 or #3. That's why Wlad won it when he beat Ruslan Chagaev. I could be wrong, but as I said, I'm pretty sure.

keithmoonhangover wrote:

@matt___s wrote:Funny question I know -

RING magazine has its titles dished out when #1 and #2 fight, as we all know, my question is, would it be possible to tell me when divisional #1 and #2 on the Boxrec database have fought?

It's #1 vs #2 or #3. That's why Wlad won it when he beat Ruslan Chagaev. I could be wrong, but as I said, I'm pretty sure.

This question was answered here:

http://boxrec.com/forum/viewtopic.php?f ... start=2154

computerrank wrote:

conan_the_cribber wrote:
Martin, I've said in the past, the ratings algorithm should reflect common sense i.e. the way people think in a domain. For example, it makes no sense in the Fifa rankings to make Luxemburg number one, if forever what reason they have a lucky win against current number one Holland.

This part of the algorithm makes no sense to someone involved in boxing. Vitali comes back and has a crap rating ang fights Peter, who was probably 2nd or 3rd at the time. Then he fights Gomez who was, as far as ratings points go, in the same level. Now because of these stellar victories, he has too many points afterwards and gets penalised for CORRECTLY taking on the best challengers and winning.

Inactivity against top opponents should be a factor in the algorithm, without a doubt. But it should be defined as "How long is it since you fought somebody at your level". Vitali fought Gomez, who was at his level, so he should have 18 months grace from then.

conan

It is common sense in professional boxing, that a winner should be rated above the loser. So in the Boxrec ratings the winner will be rated at 50 + x % of the loser's pre_bout rating at least.

The Boxrec ratings represent the level a boxer has actually fought the recent time (18 months). If there is no matching opponent to really challenge the boxer, there is no evidence, that he kept his level - and so he will lose rating points.

Gomez was at his level at this time - regarding the Boxrec ratings - Gomez wasn't at a level to grant him a later higher rating.

Martin, this is the way I view this.

A boxer wants to claim that he is at a certain level e.g. “I am an elite marquee boxer, because I beat x, y and z in a row”. After a certain time however, you can say, “yes, you did beat x, y and z in a row to reach your elite status, but you’ve only been fighting bums b1, b2 and b3 since then, so we don’t know if you’re elite any more”. The victories against x, y and z are causal, they are responsible for the elite rankings. Therefore they should be used for the expiry of belief in the boxer’s claim. You could tell the boxer “You haven’t fought anyone good since ‘z’, so there is now a question mark against your elite marquee claim and I have to reduce your rating”.

However the current implementation is “you beat ‘z’, so your win against ‘x’ is now retrospectively meaningless”. This is actually a self-criticism of your rating algorithm. You don’t trust the results generated by the win against ‘z’. It might be, that ‘x’ was around the same level as b1 and b2, however, at the time of fighting x, the fighter himself was at that level. Your implementation also contradicts the common understanding of the progression of a talented boxer’s career i.e. fight bums gather experience, take on a few warhorses, crossroads fight against another up and comer, title fight, elite fight. The fight against the up-and-comer was extremely significant at the time, yet because of the elite fight, is now worthless. That doesn’t seem to make sense.

Conan

p.s. where did the 50 come from? That’s not mentioned anywhere.

computerrank wrote:

conan_the_cribber wrote:
Martin, again I think a tweak is needed. Common sense screams out "Wlad should not lose points for fighting Haye" and "Vitali shoudl not lose points for fighting Adamek". Both opponents were number three at the time of the fight. In other weight divisions, you can fight someone in a higher weight class, that also has points. In heavyweight, there is nowhere to go. This should be reflected in the reduction algorithm.

Thanks

conan

There is no evidence that the heavyweights suffer from this.

Maybe in other divisions boxers lose more points against opponents from higher divisions than they win against those from lower divisions - or maybe it is the other way round.

Martin, mathematical theory says, "this must be the case". How many possible straights in poker can I make with a King. Is is less or more than I can make with a 7? The ends of a range are always special cases. Minimum weight also has the same problem. There is no chance to make a mega-fight with a boxer going up in weight. In both cases, the pool of similar fighters that I have to potentially compete against is smaller.

conan

Actually Martin, I think the real solution to the reduction is relativity and there are two measures of that. One measure is the rating. The other measure is ranking at the time of the fight.

How do you determine the "best available contenders". The current implementation is a comparison of the rating. This ignores the fact, that a boxer might fight the best available contender out there. I think anyone will tell you, that if the number 2 boxer fights the number 3 boxer in a weight division (Vitali vs Adamek) then he should not be penalised for doing that. Same with Wlad. If the number 1 fights the number 3, then this is not a class mismatch. This is doing his duty as a champion to fight the best contenders out there.


conan

conan_the_cribber wrote:
Martin, this is the way I view this.

...

However the current implementation is “you beat ‘z’, so your win against ‘x’ is now retrospectively meaningless”. This is actually a self-criticism of your rating algorithm. You don’t trust the results generated by the win against ‘z’. It might be, that ‘x’ was around the same level as b1 and b2, however, at the time of fighting x, the fighter himself was at that level. Your implementation also contradicts the common understanding of the progression of a talented boxer’s career i.e. fight bums gather experience, take on a few warhorses, crossroads fight against another up and comer, title fight, elite fight. The fight against the up-and-comer was extremely significant at the time, yet because of the elite fight, is now worthless. That doesn’t seem to make sense.

Conan

p.s. where did the 50 come from? That’s not mentioned anywhere.

Conan,

would you please stuff your reasoning with some figures - I won't get it elsewise.

The 50% is part of the Boxrec Ratings despription - rule 23:

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

Best regards
Martin

conan_the_cribber wrote:
Martin, mathematical theory says, "this must be the case". How many possible straights in poker can I make with a King. Is is less or more than I can make with a 7? The ends of a range are always special cases. Minimum weight also has the same problem. There is no chance to make a mega-fight with a boxer going up in weight. In both cases, the pool of similar fighters that I have to potentially compete against is smaller.

conan

Conan,

I won't correct this as the ratings are intended to reflect the achievements (by effective opponents' ratings) in the last 18 months. There is no prove that the rating IS higher - where it sure MIGHT be higher.

Best regards
Martin

computerrank wrote:

conan_the_cribber wrote:
Martin, mathematical theory says, "this must be the case". How many possible straights in poker can I make with a King. Is is less or more than I can make with a 7? The ends of a range are always special cases. Minimum weight also has the same problem. There is no chance to make a mega-fight with a boxer going up in weight. In both cases, the pool of similar fighters that I have to potentially compete against is smaller.

conan

Conan,

I won't correct this as the ratings are intended to reflect the achievements (by effective opponents' ratings) in the last 18 months. There is no prove that the rating IS higher - where it sure MIGHT be higher.

Best regards
Martin

Hi Martin,

there is no need to correct anything, if the rankings i.e. 3rd in division are also taken into account as I wrote above. Currently the rating i.e. 1200pts is the only thing used.

cheers

conan

computerrank wrote:

conan_the_cribber wrote:
Martin, this is the way I view this.

...

However the current implementation is “you beat ‘z’, so your win against ‘x’ is now retrospectively meaningless”. This is actually a self-criticism of your rating algorithm. You don’t trust the results generated by the win against ‘z’. It might be, that ‘x’ was around the same level as b1 and b2, however, at the time of fighting x, the fighter himself was at that level. Your implementation also contradicts the common understanding of the progression of a talented boxer’s career i.e. fight bums gather experience, take on a few warhorses, crossroads fight against another up and comer, title fight, elite fight. The fight against the up-and-comer was extremely significant at the time, yet because of the elite fight, is now worthless. That doesn’t seem to make sense.

Conan

p.s. where did the 50 come from? That’s not mentioned anywhere.

Conan,

would you please stuff your reasoning with some figures - I won't get it elsewise.

The 50% is part of the Boxrec Ratings despription - rule 23:

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

Best regards
Martin

Hi Martin,

the only way to understand the description is if you have the code in front of you. 8)

One day, I might get around to trying to put it into layman's terms, which you can tweak.

conan

computerrank wrote:

conan_the_cribber wrote:
Martin, this is the way I view this.

...

However the current implementation is “you beat ‘z’, so your win against ‘x’ is now retrospectively meaningless”. This is actually a self-criticism of your rating algorithm. You don’t trust the results generated by the win against ‘z’. It might be, that ‘x’ was around the same level as b1 and b2, however, at the time of fighting x, the fighter himself was at that level. Your implementation also contradicts the common understanding of the progression of a talented boxer’s career i.e. fight bums gather experience, take on a few warhorses, crossroads fight against another up and comer, title fight, elite fight. The fight against the up-and-comer was extremely significant at the time, yet because of the elite fight, is now worthless. That doesn’t seem to make sense.

Conan

p.s. where did the 50 come from? That’s not mentioned anywhere.

Conan,

would you please stuff your reasoning with some figures - I won't get it elsewise.

The 50% is part of the Boxrec Ratings despription - rule 23:

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

Best regards
Martin

OK, I'll try. I'll use Vitali.

2008-10-11 with a rating of 100, he fights Peter (800+) and wins and gets a massive to 1098. This is clearly a good matchup.
2009-03-21 with a rating of 1098 he then fights Gomez (572 i.e. >50% of Vitali rating), and moves in the stellar p4p range of 1231. This is clearly a good match up at the time.

At this point you can say "Vitali, respect, you are an elite fighter right now, because of these wins".

He then fights top 10 ranked fighter Aerola taking him up to 1280 (moderate fight 50%>x>25% of Vitali rating) and Johnson (meaningless fight x<25% of Vitali) to leave him at 1280.

2010-05-29 he fights Sosnowski (meaningless x<25% of Vitali). He loses points beforehand because of the level of competition clause.

Now the important part. You have retrospectively changed the criteria

On the 2009-03-21, Vitali had been fighting fighters better than or equal to him. However, on the 2010-05-29 you now declare Gomez as unworthy because of Vitali's further successes. In other words, you penalise a boxer for getting the rating points that your algorithm assigns due to his wins.

Domain view

I think the boxing public's general opinion on the 2010-05-29 would be "Peter and Gomez were very could wins. The win against Aerrola was ok, he was top 10. But the defences against Johnson and Sosnowski were a bit week. We are looking to you to have a meaningful fight next, as it has been 14 months since you beat Gomez."

I do not think the boxing public's general opinion would be "The win against Peter was great as you were coming back from a layoff. But you have since proved by beating Aerrola and Johnson, that despite getting a ton of rating points for it, the win against Gomez was insignificant". This makes completely no sense at all, because Gomez was rated higher than Aerrola and Johnson when they fought.

Do you understand the problem now?

In my eyes, a fighter should have 18 months (or some other period of time) to fight somebody worthy again. There should be a date field tracking this. In Vitali's case here, on the day he fought Sosnowski, the date field would be the date that he fought Gomez. The period is still within the 18 months grace period, so that is all cool. When the Briggs fight came around though on 2010-10-16, then his 18 months is up and his pre-fight rating should be adjusted. It is fair to assume, as part of the domain problem, that a boxer's skill declines if he is not competing at his own level.

I think this is the common sense fix that is needed. 18 months grace is given from an earned date, and not by retrospectively comparing points.

Thanks.

conan

p.s. However I just realise, that Vitali's pre-bout rating did not decline for the fight against Briggs, why is that so?

@conan,

thanks for your explanations.

There is a point with your example.

I will make a test implementation and have a closer look at the results.

Best regards
Martin

Your question regarding the Briggs bout showing no reduction of the pre-bout rating of Vitali:

In conjunction with a point reduction or a reduced point win caused by the missing opponent quality rule, the tracked best opponent rating is set to 50% of the new rating.

So this tracked best opponent rating was set to

- 824 after the Peter bout
- still to 824 after the Gomez bout
- still to 824 after the Arreola bout
- still to 824 after the Johnson bout
- 1213 / 2 = 606 after the Peter bout got out of scope on 2010-03-11 and the rating was reduced to 1213, based on 572 points, the tracked best opponent rating from the Gomez bout
- 1217 / 2 = 608 after the Sonsnowski bout, caused by reduced points win not matched by the 606 points tracked best opponent rating before the bout
- still to 608 after the Briggs bout
- 1243 / 2 = 621 after the Solis bout, caused by reduced points win not matched by the 608 points tracked best opponent rating before the bout
- 1302 / 2 = 651 after the Adamek bout, caused by reduced points win not matched by the 621 points tracked best opponent rating before the bout

So there was no points reduction before the Briggs bout.

computerrank wrote:@conan,

thanks for your explanations.

There is a point with your example.

I will make a test implementation and have a closer look at the results.

Best regards
Martin

Your question regarding the Briggs bout showing no reduction of the pre-bout rating of Vitali:

In conjunction with a point reduction or a reduced point win caused by the missing opponent quality rule, the tracked best opponent rating is set to 50% of the new rating.

So this tracked best opponent rating was set to

- 824 after the Peter bout
- still to 824 after the Gomez bout
- still to 824 after the Arreola bout
- still to 824 after the Johnson bout
- 1213 / 2 = 606 after the Peter bout got out of scope on 2010-03-11 and the rating was reduced to 1213, based on 572 points, the tracked best opponent rating from the Gomez bout
- 1217 / 2 = 608 after the Sonsnowski bout, caused by reduced points win not matched by the 606 points tracked best opponent rating before the bout
- still to 608 after the Briggs bout
- 1243 / 2 = 621 after the Solis bout, caused by reduced points win not matched by the 608 points tracked best opponent rating before the bout
- 1302 / 2 = 651 after the Adamek bout, caused by reduced points win not matched by the 621 points tracked best opponent rating before the bout

So there was no points reduction before the Briggs bout.

Hi Martin,

Sorry, try as I might, I still need the code to understand your implementation. From the information above,
- you have some number N which you use to compare opponents. (The rules for this are not completely stated).
- at some point, this number N was set to 824 after the Peter fight
- this number N has an expiry date D. You mentioned 2010-03-11, which is 17 months after the Peter fight (probably a typo)
- upon reaching D, if some (unnanmed) criteria is not met, then N is set to 50% of the current ranking. (Note: Theoretically, for an active fighter, the value of N could increase). In this example, it means Vitali's ranking gets reduced to 606.
- After the Sosnowski bout this value of N is revised to 608.

Now we come to Briggs fight, and the part that is missing is, "what is the use of this new value 608"? According to your previous explanation (see below), Vitali continues to fight no-one of significance within the last eighteen months and should be penalised again for fighting Briggs. This would make sense, as the gap between fighting a credible opponent is getting larger.

A boxer performing against an opponent better than 50% of his own rating within 18 months keeps his rating. The approved ratio so ist 1:2 at least.

A boxer performing against a best opponent rated 25% of his own rating within 18 months loses 25% of his rating down to 75%. The approved ratio so still is 1:3.

In your explanation above, you seem to indicate that because of the new value of N, that some criteria is met that the fight against Briggs is not that bad and not worthy of a further demotion. However, I can't understand that, from the description that you have given.

cheers

conan