Ratings - please read before commenting - Archived

computerrank · Post by **computerrank** » 31 Jan 2007, 17:55

jujigatame wrote:Are you telling me that before he KOed Kim in Japan, Mamani was already ahead of Alcine, who had beat him dominantly less than 1 year ago?

He already was rated in Middleweight before his fight in Japan - whereas Alcine is rated in Light Middleweight.

computerrank · Post by **computerrank** » 31 Jan 2007, 17:59

JCS83MD wrote:
jujigatame wrote:Are you telling me that before he KOed Kim in Japan, Mamani was already ahead of Alcine, who had beat him dominantly less than 1 year ago?
The hometown factor likely played a large role here... as Mamani was the outsider.. thus not being penalized very much for the WIDE loss.

Which supports my thoughts that the really wide decisions shouldn't be impacted by hometown factor... as they would've lost anywhere.

Hometown factor should affect everything but real wide decision.. and it should be turned down from its current status.

Even a wide range win abroad seems to be worth more than such a win at home.

jujigatame · Post by **jujigatame** » 31 Jan 2007, 23:29

computerrank wrote:
jujigatame wrote:Are you telling me that before he KOed Kim in Japan, Mamani was already ahead of Alcine, who had beat him dominantly less than 1 year ago?
He already was rated in Middleweight before his fight in Japan - whereas Alcine is rated in Light Middleweight.

Well first of all, that begs the question of why he's ranked at MW. He should be ranked at LMW.

Second of all, even when you do the point conversion, the numbers don't come out in any way that makes sense. If Mamani gained 150 points from beating Kim then he must have been above Alcine before the fight, which makes no sense considering that Alcine basically shut him out.

computerrank · Post by **computerrank** » 01 Feb 2007, 07:16

jujigatame wrote:
computerrank wrote:
jujigatame wrote:Are you telling me that before he KOed Kim in Japan, Mamani was already ahead of Alcine, who had beat him dominantly less than 1 year ago?
He already was rated in Middleweight before his fight in Japan - whereas Alcine is rated in Light Middleweight.
Well first of all, that begs the question of why he's ranked at MW. He should be ranked at LMW.

Second of all, even when you do the point conversion, the numbers don't come out in any way that makes sense. If Mamani gained 150 points from beating Kim then he must have been above Alcine before the fight, which makes no sense considering that Alcine basically shut him out.

You cannot compare the rating points in both divisions - as they are normalized to their standard before.

jomothepure · Post by **jomothepure** » 01 Feb 2007, 07:27

computerrank wrote:
jomothepure wrote:Would someone by any chance have the patience to help an idiot like me understand how the rankings work (in a practical sense)

Basically, I want to fins out how a boxers' rating would fair if he continually fought guys making their debut.

Here's two German Cruisers, both winning on their debut against guys 0-0. One won by KO1, the other by TKO2. I thought this used to have a bearing on the rating (with TKO having a slightly higher weighting) but anyway, that's a different story.)

http://www.boxrec.com/boxer_display.php?boxer_id=377353
http://www.boxrec.com/boxer_display.php?boxer_id=356496

Anyway, they both have a points tally of 7. Basically, could I ask how their score is calculated, in practical terms? I've highlighted the rules I think relevant to calculating it, but just can't do it.

KO, TKO, RTD are rewarded with full value v=1, cd=1.

The winner gets a certain part of the opponent's points and additionaly a certain part of the rating difference to the opponent's rating. The winner gets 10 additional points, if his opponent's rating is higher than a quarter of his own rating.

Right, this is the formula
r_a_new = r_a + 0.345*v*r_b + 0.345/(1+2*cl)*(r_b - r_a)

Both boxers having a rating of 0, and v=1, cd=1, which only gives me a score of 0 for r_a_new. What's heppening
You are right, the formula gives 0 points - but

The winner gets 10 additional points, if his opponent's rating is higher than a quarter of his own rating. Both were at 0 points.

So he is at 10 points - and now the rule is applied, that he didn't defeat a quality opponent (25% of his own rating of 10) within 18 month - and he is at 7.5 points finally (he loses 25% from his 10 points).

You see 7 points in the ratings.

That's cool, thanks. So is the 25% reduction effective immediately? eg. Martin Kempf only fought in December, but already has 7 points.

And so if he were to fight another guy making his debut (ie. 0 points) he wouldn't get any more points (even if it were by KO) because 0 is less than 25% of 7?

computerrank · Post by **computerrank** » 01 Feb 2007, 07:35

jomothepure wrote:...
And so if he were to fight another guy making his debut (ie. 0 points) he wouldn't get any more points (even if it were by KO) because 0 is less than 25% of 7?

... yes - but, because 0 is less than 25% of 10 points. 7 points is already the reduced rating.

jujigatame · Post by **jujigatame** » 01 Feb 2007, 10:13

You cannot compare the rating points in both divisions - as they are normalized to their standard before.

Alright, well why don't we move Mamani back to 154 where he belongs, and then we can discuss his ranking in that division? I think that would be the appropriate action to take at this point.

JCS · Post by **JCS** » 01 Feb 2007, 10:58

jujigatame wrote:
You cannot compare the rating points in both divisions - as they are normalized to their standard before.
Alright, well why don't we move Mamani back to 154 where he belongs, and then we can discuss his ranking in that division? I think that would be the appropriate action to take at this point.

I thought he was #4 at 154...

jujigatame · Post by **jujigatame** » 01 Feb 2007, 12:03

For some reason he has been moved back to 160.

JCS · Post by **JCS** » 12 Feb 2007, 11:23

I'm still thinking the hometown factor is still a little too intense, and possibly flawed.

Take Sergio Gabriel Martinez for example. His rating was no doubt largely inflated by those 3 consecutive wins in the UK. Since then, he's faced trash. Nevertheless, he's #24 P4P and 2nd at 154. Javier Mamani has the same effect.

I'd like to see a sample set where the hometown influence is about 25% weaker. I'd also like to see a sample set where the hometown influence is not involved in wide decisions (wider than that which will lower the $fix). I guess that could be getting too technical though... either way, I think it needs to be altered.

Can you hook that up c-rank? I'm not sure what the current version is anymore... plus I hate the ultra organization of the code :)

jujigatame · Post by **jujigatame** » 12 Feb 2007, 12:00

Mamani is also still incorrectly listed at 160.

JCS · Post by **JCS** » 12 Feb 2007, 12:53

jujigatame wrote:Mamani is also still incorrectly listed at 160.

That's probably a grievance for the editors.

JCS · Post by **JCS** » 17 Feb 2007, 19:09

After some personal testing, I have deduced that the hometown factor is only part of the problem.

I think we need to look into altering the initial rating of a fighter depending on the talent pool in surrounding weights and location of their debut. This way, a fighter in Georgia can not rocket up the ratings by fighting only fighters of Georgia... since the pool of fighters is very low.

jujigatame · Post by **jujigatame** » 26 Feb 2007, 15:12

It might be time to rethink the way the divisions are "normalized" against each other, now that Darnell Wilson has debuted at #8 in the HW rankings.

JCS · Post by **JCS** » 26 Feb 2007, 15:13

jujigatame wrote:It might be time to rethink the way the divisions are "normalized" against each other, now that Darnell Wilson has debuted at #8 in the HW rankings.

.....Curious

computerrank · Post by **computerrank** » 27 Feb 2007, 05:45

When a boxer is moved between the divisions by editors, he first takes his points with him until the next recalculation of the ratings.

So when a boxer is moved from cruiserweight to heavyweight by the editors, his rating will be far too high first, until the next recalculation.

So simply have a look after the next recalculation - an hour to go ...

computerrank · Post by **computerrank** » 27 Feb 2007, 06:40

... #31

computerrank · Post by **computerrank** » 01 Mar 2007, 09:35

A new correction and update is launched - last was from October 2006:

Changes:

1. Correction for wrong usage of the two parameters "clearness of decision" $CD and "weight of bout" $FIX. There are some changes in the ratings, but no basic ones.

2. Further improved launch process for lower rated boxers.

3. Weight of not-santioned-bouts is reduced to 50%

4. Improved all-time-ratings by regarding
- the particular result
- regarding career high losses too, not only the best wins

JCS · Post by **JCS** » 04 Mar 2007, 18:56

computerrank wrote:A new correction and update is launched - last was from October 2006:

Changes:

1. Correction for wrong usage of the two parameters "clearness of decision" $CD and "weight of bout" $FIX. There are some changes in the ratings, but no basic ones.

2. Further improved launch process for lower rated boxers.

3. Weight of not-santioned-bouts is reduced to 50%

4. Improved all-time-ratings by regarding
- the particular result
- regarding career high losses too, not only the best wins

Martin,

Did you see my post and/or PM regarding the entry point of fighters from countries with lower talent pools?

computerrank · Post by **computerrank** » 05 Mar 2007, 07:00

JCS83MD wrote: Martin,

Did you see my post and/or PM regarding the entry point of fighters from countries with lower talent pools?

Yes, I did, Jason.

But it seems too complicated to me.

JCS · Post by **JCS** » 05 Mar 2007, 09:14

computerrank wrote:
JCS83MD wrote: Martin,

Did you see my post and/or PM regarding the entry point of fighters from countries with lower talent pools?
Yes, I did, Jason.

But it seems too complicated to me.

There has to be a way to limit the potential or entry point of fighters from small countries such as Georgia. Check out Shalva Jomardashvili for example... as well as the other Georgian fighters. They are rising in rank simply fighting each other rather than fighting proven talent. If you and myself got 10 friends together and kept fighting one another, the best of us would no doubt get into the Top 100 after time...

http://www.boxrec.com/boxer_display.php?boxer_id=320439

Would some kind of basic cap help?? Such as, limiting a fighter's rating to double the rating of his highest rated win?

pundit · Post by **pundit** » 08 Mar 2007, 10:24

At long last the algorithm has understood that Mundine is the world's best super-middleweight, and not somewhere between ## 13 and 20.

computerrank · Post by **computerrank** » 08 Mar 2007, 11:01

@ all time ratings - new version

The old version considered the 5 top annual (end) ratings in the division assigned.

Weaknesses - which resulted in a lot of ratings, which were not correct - more than now. You seem to forget too fast ...

- Boxers, who were rated in the division at the end of the year, but only had maybe one recent low quality bout in this division, got a high annual end rating in this division, while having earned their rating in other divisions.

- Boxers, who were rated high for long time periods, but had few quality bouts and only some spark results, got a high rating in the all time ratings

- Low quality losses in the division were not considered explicitly.

Now - the the current version:

- The actual performances in the division only are considered in a sliding average with weight 0.1, if it improves the rating. So getting the top.

- Low performances below this level reduce the ratings by the same sliding average, if suffered before the last improvement. So reduce the rating, but not in the phase of decline.

- The performance is defined as the current rating of the opponent plus/minus the decision weight = plus/minus 34.5 percent maximum. Clear losses to superior opponents cannot improve the rating.

So now it is ensured, that only real performances in the particular division are considered, and significant losses are considered too.

JCS · Post by **JCS** » 08 Mar 2007, 12:17

Cobwebcat wrote:Computerrank: That's not an answer. All you've done is described the new system and how it is better than the last one.

What we are saying is the all-time rankings are ridiculous and need changing. If they are better than before, fair enough, but they are still outrageous and people are giving examples that are obvious even to my untrained eye. Fighters who are legends and have exemplary records are below others that have never even fought for a world title and/or have average at best records.

Something is wrong and its embarrasing!

Remember when I did a test and weighted the years... that looked a lot better I thought... still didn't show much love for the old-time boxers though.

computerrank · Post by **computerrank** » 08 Mar 2007, 12:43

Cobwebcat wrote:Computerrank: That's not an answer. All you've done is described the new system and how it is better than the last one.

What we are saying is the all-time rankings are ridiculous and need changing. If they are better than before, fair enough, but they are still outrageous and people are giving examples that are obvious even to my untrained eye. Fighters who are legends and have exemplary records are below others that have never even fought for a world title and/or have average at best records.

Something is wrong and its embarrasing!

Cobwebcat

you are a man of big words ...

You like to use words like - outrageous, obvious, exemplary records, world titles ...

... and at the same time you could know, that I only will handle figures and facts:

- titles are not regarded, as you know and why
- records - pure wins and losses means near to nothing
- and emotions like outrageous and obvious are not operative

might be

- some "legends" have more bad losses in their prime, than regarded till now
- some "legends" have less big wins and only survived with low quality wins for a long time and a lot title bouts
- some "lower estimated boxers" have bigger wins, although missing title fights, than some people regarded till now