Ratings - please read before commenting - Archived

[vaibhavv4u]

Hi guys. I apologise in advance if this question is naive. But I have gone through the forum and what I can make of the algorithm (or the version I think is in use)
I tried to use the formula mentioned and predict what the rating for a winner of a debut fight would be vs a fellow debutant or a guy with a 0 win record. It comes out to around 0.044 at best as per my calculation. Yet I see some records such as Murat Gassiev on debut against a guy with a 0-3 record getting 6.349 for a 4 round UD victory.

https://boxrec.com/en/event/628190/1626089

As someone about to make his own professional debut, this has me intrigued. What am I missing ???

[computerrank]

Hi guys. I apologise in advance if this question is naive. But I have gone through the forum and what I can make of the algorithm (or the version I think is in use)
I tried to use the formula mentioned and predict what the rating for a winner of a debut fight would be vs a fellow debutant or a guy with a 0 win record. It comes out to around 0.044 at best as per my calculation. Yet I see some records such as Murat Gassiev on debut against a guy with a 0-3 record getting 6.349 for a 4 round UD victory.

https://boxrec.com/en/event/628190/1626089

As someone about to make his own professional debut, this has me intrigued. What am I missing ???

The point is - a boxers rating at time of any of his bouts is influenced by each result before and after that bout.

You are intuitively expecting the rating of a boxer should start at 0 points before his first bout, and then with his first bout a boxer should earn his first rating, I guess.

BoxRec Whole-History ratings work a bit different.

Coming back to your example. From today's perspective Gassiev's rating at time of his first bout is 6.349 against an opponent with rating 0.002.

Gassiev's 6.349 points at time of that bout are a best estimation of his rating at that time regarding all his results after that. It means, it is is a best guess, his rating was at that level at that time. His later very impressing results gives good

If we would turn back time and we would have to estimate his rating after his first win - not knowing his later results - the best guess of his rating at that time would have been a rating for a win against an opponent with 0-4 record. I would have been more at 0.001 points.

BoxRec Whole-History ratings are balancing the ratings of each boxer at time of each of his bouts, while later and earlier bouts are enhancing that rating by a win or decreasing that rating by a loss with decreasing influence with increasing time distance.

So the rating career arc of each boxer is reflecting this result career arc.

[vaibhavv4u]

Hi guys. I apologise in advance if this question is naive. But I have gone through the forum and what I can make of the algorithm (or the version I think is in use)
I tried to use the formula mentioned and predict what the rating for a winner of a debut fight would be vs a fellow debutant or a guy with a 0 win record. It comes out to around 0.044 at best as per my calculation. Yet I see some records such as Murat Gassiev on debut against a guy with a 0-3 record getting 6.349 for a 4 round UD victory.

https://boxrec.com/en/event/628190/1626089

As someone about to make his own professional debut, this has me intrigued. What am I missing ???

The point is - a boxers rating at time of any of his bouts is influenced by each result before and after that bout.

You are intuitively expecting the rating of a boxer should start at 0 points before his first bout, and then with his first bout a boxer should earn his first rating, I guess.

BoxRec Whole-History ratings work a bit different.

Coming back to your example. From today's perspective Gassiev's rating at time of his first bout is 6.349 against an opponent with rating 0.002.

Gassiev's 6.349 points at time of that bout are a best estimation of his rating at that time regarding all his results after that. It means, it is is a best guess, his rating was at that level at that time. His later very impressing results gives good

If we would turn back time and we would have to estimate his rating after his first win - not knowing his later results - the best guess of his rating at that time would have been a rating for a win against an opponent with 0-4 record. I would have been more at 0.001 points.

BoxRec Whole-History ratings are balancing the ratings of each boxer at time of each of his bouts, while later and earlier bouts are enhancing that rating by a win or decreasing that rating by a loss with decreasing influence with increasing time distance.

So the rating career arc of each boxer is reflecting this result career arc.

Thank you. That makes so much more sense. I was wracking my head with the algorithm alone. I like playing with numbers and to min max situations when possible, at least in the first few fights. I now realize it's not a factor of the current ratings alone. I am also training with the current IBO Oceania Orient featherweight Champion, and he was perplexed as to how his ratings kept changing, since it directly affects his chances of being eligible for the next tier titles.
It's back to the drawing board for us :).

[pugilisticspecialist]

Can I see where Usyk and Huck would rank all-time at cruiserweight with the new ratings, please?

[computerrank]

Can I see where Usyk and Huck would rank all-time at cruiserweight with the new ratings, please?

- Usyk is at Heavyweight - multiply his rating with 1.73 to get his Cruiserweight rating
- Huck is not rated - he would 149 points at Cruiserweight

[pugilisticspecialist]

What happened to all-time ratings being the same as peak ratings after 15 years?

[computerrank]

What happened to all-time ratings being the same as peak ratings after 15 years?

All-time ratings go with the average rating of the top 5 defeated opponents since a while.

[pugilisticspecialist]

What happened to all-time ratings being the same as peak ratings after 15 years?

All-time ratings go with the average rating of the top 5 defeated opponents since a while.

But then why would Usyk have a higher all-time rating if you put him in cruiserweight?
Shouldn't his all-time rating be the same regardless of the division he's in, since the average rating of his top 5 defeated opponents would still be the same?

Also, Rocky Marciano's best rated win is 93.65-rated Archie Moore, but Marciano's all-time rating is 237.7, is there some kind of multiplier?

[computerrank]

What happened to all-time ratings being the same as peak ratings after 15 years?

All-time ratings go with the average rating of the top 5 defeated opponents since a while.

But then why would Usyk have a higher all-time rating if you put him in cruiserweight?
Shouldn't his all-time rating be the same regardless of the division he's in, since the average rating of his top 5 defeated opponents would still be the same?

==> The basic WHR rating is independent from the weight division and reflects the absolute strength of a boxer. But that would mean relative success in a lower weight division - with less absolute strength - would be rated lower. That doesn't matter as long as only boxers from the same weight division are compared. But in order to compare different weight divisions, BoxRec introduced a division weight multiplier in relation to the upper weight limits.

Also, Rocky Marciano's best rated win is 93.65-rated Archie Moore, but Marciano's all-time rating is 237.7, is there some kind of multiplier?

==> The exact formula is (0.67 * average_5_top_defeated_opponents + 0.33 * career_top_rating) * division_factor * 10

[pugilisticspecialist]

All-time ratings go with the average rating of the top 5 defeated opponents since a while.

But then why would Usyk have a higher all-time rating if you put him in cruiserweight?
Shouldn't his all-time rating be the same regardless of the division he's in, since the average rating of his top 5 defeated opponents would still be the same?

==> The basic WHR rating is independent from the weight division and reflects the absolute strength of a boxer. But that would mean relative success in a lower weight division - with less absolute strength - would be rated lower. That doesn't matter as long as only boxers from the same weight division are compared. But in order to compare different weight divisions, BoxRec introduced a division weight multiplier in relation to the upper weight limits.

Also, Rocky Marciano's best rated win is 93.65-rated Archie Moore, but Marciano's all-time rating is 237.7, is there some kind of multiplier?

==> The exact formula is (0.67 * average_5_top_defeated_opponents + 0.33 * career_top_rating) * division_factor * 10

Is there any chance I could see the all-time fighters by career peak, without the top 5 opponents thing and without the division factor?

[computerrank]

Is there any chance I could see the all-time fighters by career peak, without the top 5 opponents thing and without the division factor?

No there is no application for that. Here a run on the test system for Heavyweights

[pugilisticspecialist]

Is there any chance I could see the all-time fighters by career peak, without the top 5 opponents thing and without the division factor?

No there is no application for that. Here a run on the test system for Heavyweights

Is this just the career peak rating?
If so, can I see the same for cruiserweight, please?

[computerrank]

Is this just the career peak rating?
If so, can I see the same for cruiserweight, please?

[pugilisticspecialist]

Is this just the career peak rating?
If so, can I see the same for cruiserweight, please?

Can I see career peak ratings for light heavyweight and super middleweight as well, please?

[computerrank]

Can I see career peak ratings for light heavyweight and super middleweight as well, please?

Light Heavyweight:

Super Middleweight:

[Rocky-]

Okay I have to admit the ratings have vastly improved. Good job.

[pugilisticspecialist]

What ratings version are we on now, and what is the prediction ratio?

[computerrank]

What ratings version are we on now, and what is the prediction ratio?

Release 1.1.7 with winner prediction ratio 84.36 percent.

[SportsRatings]

What ratings version are we on now, and what is the prediction ratio?

Release 1.1.7 with winner prediction ratio 84.36 percent.

Have you ever calculated a retropspective "prediction" ratio with all fight information included? Obviously I don't mean with knowledge of who won a fight, but by the fighter ratings for each particular match using the backward-looking curve of the WHR?

[SportsRatings]

What ratings version are we on now, and what is the prediction ratio?

Release 1.1.7 with winner prediction ratio 84.36 percent.

Have you ever calculated a retropspective "prediction" ratio with all fight information included? Obviously I don't mean with knowledge of who won a fight, but by the fighter ratings for each particular match using the backward-looking curve of the WHR?

I guess this could also be called the real "upset" rate, since when the lower-rated fighter was the winner even based on the retrospective ratings it's an upset

[computerrank]

Release 1.1.7 with winner prediction ratio 84.36 percent.

Have you ever calculated a retropspective "prediction" ratio with all fight information included? Obviously I don't mean with knowledge of who won a fight, but by the fighter ratings for each particular match using the backward-looking curve of the WHR?

I guess this could also be called the real "upset" rate, since when the lower-rated fighter was the winner even based on the retrospective ratings it's an upset

Your upset ratio is 4.99 percent.