Ratings - please read before commenting - Archived

[jujigatame]

Even though my last post was very critical, I think what CR has done here is getting misunderstood. It seems like he's basically tried to chart a curve of a boxer's "true" ability level throughout their career, accounting for all the data available throughout the entire database. It's an interesting thing to do, and seems like it would be exceptionally valuable for gambling/handicapping purposes. It's just not the way most boxing fans think of boxing rankings, and not the way they want their rankings to work. These rankings are attempting to be predictive and forward-looking, whereas most boxing rankings are backward-looking, reacting only to a boxer's specific accomplishments as they happen.

I think it's an interesting alternative system, I just wish it didn't replace the old one. FightMatrix.com for example maintains multiple different ranking algorithms that users can switch between. That's a really nice feature.

[computerrank]

Even though my last post was very critical, I think what CR has done here is getting misunderstood. It seems like he's basically tried to chart a curve of a boxer's "true" ability level throughout their career, accounting for all the data available throughout the entire database. It's an interesting thing to do, and seems like it would be exceptionally valuable for gambling/handicapping purposes. It's just not the way most boxing fans think of boxing rankings, and not the way they want their rankings to work. These rankings are attempting to be predictive and forward-looking, whereas most boxing rankings are backward-looking, reacting only to a boxer's specific accomplishments as they happen.

I think it's an interesting alternative system, I just wish it didn't replace the old one. FightMatrix.com for example maintains multiple different ranking algorithms that users can switch between. That's a really nice feature.

Boxrec will be back with parallel rating options too. I already said that. But it needs time.

There are some priority non-rating features do be implemented first.

I am working on a solution, to show the ratings impact of a bout directly in the current version.

[marcianofan]

Boxrec will be back with parallel rating options too. I already said that. But it needs time.

There are some priority non-rating features do be implemented first.

I am working on a solution, to show the ratings impact of a bout directly in the current version.

I know in the past you said there's no specific timeframe for this, and I know you'd probably be hesitant to announce one even if you do have an idea because of the pressure of knowing everyone will throw a fit 2 days before it's due on account of it's not early...but is there even a worst-case/best-case sort of ballpark estimate you might be able to float for how much time the parallel ratings might take to implement? We're an impatient lot, I know, but having a concrete time period that we know 100% not to expect an update by might help that impatience somewhat.

I'll say this- I do have more respect for what you're trying to do with the new rankings than I did at rollout, but like many others I still prefer the old ones for my own purposes, and will be thrilled when they are at least available for viewing again.

[pugilisticspecialist]

Even though my last post was very critical, I think what CR has done here is getting misunderstood. It seems like he's basically tried to chart a curve of a boxer's "true" ability level throughout their career, accounting for all the data available throughout the entire database. It's an interesting thing to do, and seems like it would be exceptionally valuable for gambling/handicapping purposes. It's just not the way most boxing fans think of boxing rankings, and not the way they want their rankings to work. These rankings are attempting to be predictive and forward-looking, whereas most boxing rankings are backward-looking, reacting only to a boxer's specific accomplishments as they happen.

I think it's an interesting alternative system, I just wish it didn't replace the old one. FightMatrix.com for example maintains multiple different ranking algorithms that users can switch between. That's a really nice feature.

Boxrec will be back with parallel rating options too. I already said that. But it needs time.

There are some priority non-rating features do be implemented first.

I am working on a solution, to show the ratings impact of a bout directly in the current version.

The new system is the best it's ever been. Especially the all-time rankings.
I wouldn't bother with parallel rating options, it just overcomplicates things. Just use the system that best predicts results.
If you have multiple rating options, you're just going to get tons of bettors coming to the forum asking which system is best for predicting results.
You don't need a solution to show the ratings impact of a bout directly either. We already pretty much know the impact of a bout by the result and how many points the opponent has.
The old system was stupid because it basically implied that Andy Ruiz was the ~15th best heavyweight in the world at the opening bell of his first fight with Joshua, but 7 rounds later he'd magically developed into the best heayweight in the world. The new system is closer to reality.

[pugilisticspecialist]

FightMatrix.com also needs to figure out the best system for predicting bouts and get rid of the other systems.
They currently have FOUR options and basically make end-users like me guess which one is the best. That's THEIR job to figure out, not mine.

[JCS]

FightMatrix.com also needs to figure out the best system for predicting bouts and get rid of the other systems
They currently have FOUR options and basically make end-users like me guess which one is the best. That's THEIR job to figure out, not mine.

Read the FAQ for more info.. Prediction rate isn't a primary goal for the main system. The alternate options are just to give folks something to look at and pick whichever one they like as there are slight differences.. All the systems have comparable prediction rate. MMA is too variable for a "magic bullet" in terms of overall prediction rate.

Going to be rolling out some additional functionality soon that show fighters in the standard system who vary wildly from the pure mathematical counterparts.

[Rocky-]

What is the logic behind a fighter winning a 4 star bout and in the following match against a higher ranked guy the system downgrades the event to 3 stars?!? In other words 4 VS 5 = 3

Would please post the bout_ids and boxer ids?

A couple of examples…

Boxer ID# 176707 won a 4 star bout Fight:1969536 against boxer ID# 477723 then fought boxer ID# 457231 in Fight:2004275

Boxer ID# 488033 won a 4 star bout Fight:1981524 against boxer ID# 014452 then fought boxer ID# 356831 in Fight:2037019

There should be a rule implemented that ensures the winner of a 4 star fight doesn't drop-off to 3 stars in his very next bout against higher ranked opposition

[jujigatame]

FightMatrix.com also needs to figure out the best system for predicting bouts and get rid of the other systems
They currently have FOUR options and basically make end-users like me guess which one is the best. That's THEIR job to figure out, not mine.

I don't think most users of either BoxRec or FightMatrix particularly value prediction rate as a criteria for their rankings. Maybe some gamblers do but that's probably a relatively small percentage of the user base.

[DazBoxingFan]

FightMatrix.com also needs to figure out the best system for predicting bouts and get rid of the other systems
They currently have FOUR options and basically make end-users like me guess which one is the best. That's THEIR job to figure out, not mine.

I don't think most users of either BoxRec or FightMatrix particularly value prediction rate as a criteria for their rankings. Maybe some gamblers do but that's probably a relatively small percentage of the user base.

Exactly. This model may be good for a gambling website but not a boxing one. As previously mentioned I too appreciate CR always looking to improve but it’s never good how he bypasses any criticism of any kind and doesn’t ask for users opinion before changing things - just as he feels

[computerrank]

Would please post the bout_ids and boxer ids?

A couple of examples…

Boxer ID# 176707 won a 4 star bout Fight:1969536 against boxer ID# 477723 then fought boxer ID# 457231 in Fight:2004275

Boxer ID# 488033 won a 4 star bout Fight:1981524 against boxer ID# 014452 then fought boxer ID# 356831 in Fight:2037019

There should be a rule implemented that ensures the winner of a 4 star fight doesn't drop-off to 3 stars in his very next bout against higher ranked opposition

Ok, I now understand.
Both boxers were in decline at end of their career in this time period. So, when these boxers match their next bout, the ratings are down retrospectively, as the ratings know the following results.

[Rocky-]

Ok, I now understand.
Both boxers were in decline at end of their career in this time period. So, when these boxers match their next bout, the ratings are down retrospectively, as the ratings know the following results.

I don't dislike that the system takes future achievements into consideration but surely an undefeated fighter who just earned his career best scalp is not "in decline" by the time of his next bout? James Kirkland completely ruined Glen Tapia yet their fight is still rated 4 stars, so there are exceptions to the current rule.

[JCS]

I actually like the new system BUT only for getting a picture of the past and analyzing career trends. It makes it much easier to see the quality of a boxer at a given point in time.

However, must agree with the majority on trying to analyze the effects of fight movement and for the state of the current rankings.

If only we could somehow combine the two... Use the old system for the current rankings, bake the new system into some kind of visualization piece for the profile pages.... as for the fight-by-fight changes, not sure there's an easy answer for that. MAYBE scrap the rating changes and show ranking changes based on the old system... I like that idea.

Current Rankings = Old System
Career Trend Visualization/Analysis = New System
Per-Fight Profile Changes = Old System, but Show Rank not Rating

[jujigatame]

Per-Fight Profile Changes = Old System, but Show Rank not Rating

I think you really need to have both for context, especially in MMA since it's such a young sport. Some divisions barely existed 20 years ago.

[pugilisticspecialist]

FightMatrix.com also needs to figure out the best system for predicting bouts and get rid of the other systems
They currently have FOUR options and basically make end-users like me guess which one is the best. That's THEIR job to figure out, not mine.

Read the FAQ for more info.. Prediction rate isn't a primary goal for the main system. The alternate options are just to give folks something to look at and pick whichever one they like as there are slight differences.. All the systems have comparable prediction rate. MMA is too variable for a "magic bullet" in terms of overall prediction rate.

Going to be rolling out some additional functionality soon that show fighters in the standard system who vary wildly from the pure mathematical counterparts.

If you're pushed for time, you could just simulate every UFC title fight in history using each system, then whichever system is most accurate should be the one and only.
You could resimulate every couple of years to make sure the "current" system is still the best.
Even if it's only 0.1% more accurate than the next best, there must be one system that reigns supreme.

[computerrank]

Ok, I now understand.
Both boxers were in decline at end of their career in this time period. So, when these boxers match their next bout, the ratings are down retrospectively, as the ratings know the following results.

I don't dislike that the system takes future achievements into consideration but surely an undefeated fighter who just earned his career best scalp is not "in decline" by the time of his next bout? James Kirkland completely ruined Glen Tapia yet their fight is still rated 4 stars, so there are exceptions to the current rule.

Regarding Tom Wade, after Soliman he had the straight loss to Golovkin and the he had 2 low-level wins. For the ratings This is a decline.
Tapia was at a bit higher level with 30 points compared with Wade's 20 points. So the 4 star-level was missed earlier by Wade.

[computerrank]

Here is a try to show the impact of a bout to the ratings - an additional field in brackets:

[Rocky-]

So the reason Victor Ortiz vs Floyd Mayweather Jr is not a 5 star fight is because Josesito Lopez broke Victor's jaw 5 months later and Ortiz wasn't the same fighter ever again? Why can't the system take "fights that ruin a guy" into account?

[computerrank]

So the reason Victor Ortiz vs Floyd Mayweather Jr is not a 5 star fight is because Josesito Lopez broke Victor's jaw 5 months later and Ortiz wasn't the same fighter ever again? Why can't the system take "fights that ruin a guy" into account?

The ratings have Ortiz at peak at time of his bout against Nate Campell, a year before.
How and why should the system have an extra look at "fights that ruin a guy" . The effects of such a ruin are already accounted for, as the later results are ruined, or?

[JCS]

Read the FAQ for more info.. Prediction rate isn't a primary goal for the main system. The alternate options are just to give folks something to look at and pick whichever one they like as there are slight differences.. All the systems have comparable prediction rate. MMA is too variable for a "magic bullet" in terms of overall prediction rate.

Going to be rolling out some additional functionality soon that show fighters in the standard system who vary wildly from the pure mathematical counterparts.

If you're pushed for time, you could just simulate every UFC title fight in history using each system, then whichever system is most accurate should be the one and only.
You could resimulate every couple of years to make sure the "current" system is still the best.
Even if it's only 0.1% more accurate than the next best, there must be one system that reigns supreme.

Probably, but UFC title fights are 0.1% of the total bout database? Not large enough of a statistical representation.

[marcianofan]

The new system is the best it's ever been. Especially the all-time rankings.
I wouldn't bother with parallel rating options, it just overcomplicates things. Just use the system that best predicts results.
If you have multiple rating options, you're just going to get tons of bettors coming to the forum asking which system is best for predicting results.
You don't need a solution to show the ratings impact of a bout directly either. We already pretty much know the impact of a bout by the result and how many points the opponent has.
The old system was stupid because it basically implied that Andy Ruiz was the ~15th best heavyweight in the world at the opening bell of his first fight with Joshua, but 7 rounds later he'd magically developed into the best heayweight in the world. The new system is closer to reality.

Most rankings measure accomplishment in some particular timeframe, not inherent skill. Otherwise we could just rank based on subjective eyeball tests. And I'm fairly disgusted by the fact that you don't think we should have options that most people would like to see and would benefit from, just because you like this one better and don't want to have to be confronted with the option.

So long as Andy Ruiz had the best set of wins, he was the top fighter. It doesn't mean he was the best necessarily, although the two concepts go hand in hand more often than not. I'd rather have a 100% clear idea of who the top fighter is, compared to having an 85% clear idea of who the best fighter is, especially when I can probably get a pretty similar view just from watching the performances and skills of the candidates. Maybe not with the precision a gambler would like to see, but I guess that might depend on the eye of the beholder.

You are entitled to your opinion, but most of us have flocked to boxrec wanting to get a clear view of a fighter's standing based on what he's actually done in the ring at that point- not have fight results predicted for us. I don't object to the latter being added for the benefit of those who do want to know that aspect, but the fact that you're lobbying for the rest of us to permanently lose the functionality we care about just so you don't have to...I don't know...see an extra unselected radio button is beyond selfish, to the point of sadism.

[JCS]

Saw a strange one.. Bobby Czyz took a huge dip in May 1985.

Seems odd..

1985-07-26 169¾ 78.09 Murray Sutherland
1985-05-09 172¼ 25.76 Mike Fisher
1985-01-18 171 63.17 Tim Broady

[felinoboxing]

Greetings to all.
I was making P4P lists per year based on boxrec ratings. I had made 11 lists when the formula to calculate the points was modified.

Here are two P4P lists from late 1968. One old system and one new system.
I remind you that in those days Ali was suspended, for that reason he is not on the list.
Compare, and help me choose which one to stay with.

Old system

Curtis Cokes 5170
Carlos Ortiz 3648
Carlos Teo Cruz 3261
Bob Foster 3101
Sandro Mazzinghi 3027
Nino Benvenuti 3017
Nicolino Locche 2996
Jose Napoles 2423
Luis Rodriguez 2396
Ismael Laguna 2199

New system

Joe Frazier 283.3
Jimmy Ellis 97.04
Jose Napoles 96.53
Bob Foster 95.56
Oscar Bonavena 89.87
Nicolino Locche 74.77
Floyd Patterson 69.53
Emile Griffith 45.10
Curtis Cokes 43.55
Luis Rodriguez 41.50

[computerrank]

Saw a strange one.. Bobby Czyz took a huge dip in May 1985.

Seems odd..

1985-07-26 169¾ 78.09 Murray Sutherland
1985-05-09 172¼ 25.76 Mike Fisher
1985-01-18 171 63.17 Tim Broady

The Fisher bout was not assigned to a division, so Heavyweight is default.

[computerrank]

Greetings to all.
I was making P4P lists per year based on boxrec ratings. I had made 11 lists when the formula to calculate the points was modified.

Here are two P4P lists from late 1968. One old system and one new system.
I remind you that in those days Ali was suspended, for that reason he is not on the list.
Compare, and help me choose which one to stay with.

Old system

Curtis Cokes 5170
Carlos Ortiz 3648
Carlos Teo Cruz 3261
Bob Foster 3101
Sandro Mazzinghi 3027
Nino Benvenuti 3017
Nicolino Locche 2996
Jose Napoles 2423
Luis Rodriguez 2396
Ismael Laguna 2199

New system

Joe Frazier 283.3
Jimmy Ellis 97.04
Jose Napoles 96.53
Bob Foster 95.56
Oscar Bonavena 89.87
Nicolino Locche 74.77
Floyd Patterson 69.53
Emile Griffith 45.10
Curtis Cokes 43.55
Luis Rodriguez 41.50

Your list NeW System is not correct. I just looked at those and found ????
Curtis Cokes 189.5
Carlos Ortiz 160.8
Sandro Mazzinghi 171.4

[Jens S]

Saw a strange one.. Bobby Czyz took a huge dip in May 1985.

Seems odd..

1985-07-26 169¾ 78.09 Murray Sutherland
1985-05-09 172¼ 25.76 Mike Fisher
1985-01-18 171 63.17 Tim Broady

The Fisher bout was not assigned to a division, so Heavyweight is default.

Why not make a logarithm that looks at the previous fight and the next fight and decide the default value from that?