Page 41 of 180
Posted: 25 Sep 2007, 14:42
by JCS
emile wrote:Do this have to do with anomalies I noticed this morning? I changed the weight classes of a few South African fighters and guys were being ranked behind fighters they had just beaten.
For example, Gideon Buthelezi at minimumweight - I dropped him from Light Fly, yet he is still nine points behind Thobani Mbangani, who he beat last month. I saw at least one more like this.
I'm thinking it should resolve itself on the next run.
Posted: 27 Sep 2007, 09:59
by JCS
Joppy now at Light Heavy.. still holds a Top 20 rating...
Posted: 27 Sep 2007, 12:50
by sockdolager
Cobwebcat wrote:I don't pretend to understand all the changes but the all-time list are not as believeable as they were...Sam McVea 13th best heavyweight?!
It's a shame they were just about right before this latest change.
He has beaten some great fighters, he lost against them too. Harry Wills, Joe Jeannette and Sam Langford to name a few. His rating was probably hightened for fighting these guys so many times.
Posted: 27 Sep 2007, 13:34
by JCS
sockdolager wrote:Cobwebcat wrote:I don't pretend to understand all the changes but the all-time list are not as believeable as they were...Sam McVea 13th best heavyweight?!
It's a shame they were just about right before this latest change.
He has beaten some great fighters, he lost against them too. Harry Wills, Joe Jeannette and Sam Langford to name a few. His rating was probably hightened for fighting these guys so many times.
The All-Time rating is purely a sliding performance average. That's it.
Definitely doesn't coincide with what is perceived as all-time greatness.
Posted: 28 Sep 2007, 06:02
by computerrank
JCS wrote: ...
The All-Time rating is purely a sliding performance average. That's it.
...
... it is the career maximum of a sliding performance average ...
Posted: 28 Sep 2007, 06:03
by computerrank
Cobwebcat wrote: ...
Whatever it is it doesn't look as good plus the rating values are taking an age to return :-(
... I am waiting for John too ... priorities
Posted: 29 Sep 2007, 05:33
by computerrank
New release to be launched:
Division change mechanism now regarding inactivity and missing opponent quality penalties - William Joppy will be at #51 LHW.
Posted: 06 Oct 2007, 17:12
by m1kee50
where did the 'highest climbers' bit go to on the homepage?
Posted: 16 Oct 2007, 10:59
by computerrank
computerrank wrote:New release to be launched:
Division change mechanism now regarding inactivity and missing opponent quality penalties - William Joppy will be at #51 LHW.
... is launched now ...
Posted: 18 Oct 2007, 02:34
by Blue
Your new division change mechanism caused Yuri Foreman to be rated below a guy he beat in his last bout.
It was only 4 months ago and neither has fought since.
http://www.boxrec.com/ratings.php?count ... o&pageID=2
Posted: 18 Oct 2007, 08:01
by computerrank
This is not caused by the division change mechanism - both boxers are rated in the same division - and the winner should be rated higher than the loser - even after a close decision.
I will analyse this - thanks.
Posted: 18 Oct 2007, 17:02
by computerrank
computerrank wrote:
This is not caused by the division change mechanism - both boxers are rated in the same division - and the winner should be rated higher than the loser - even after a close decision.
I will analyse this - thanks.
... and this is caused by the division change mechanism:
- the bout is assigned to light middleweight
- the boxers are assigned to middleweigt
- Foreman is rated higher at light middleweigt
- but - as boxers are assigned to middleweight - the transition lets Thompson the edge ...
This may be interpreted as: Foreman is regarded the better boxer at light middleweight - but Thompson still holds the edge at middleweigt.
I will both assign to light middleweight ...
Posted: 18 Oct 2007, 17:32
by JCS
computerrank wrote:computerrank wrote:
This is not caused by the division change mechanism - both boxers are rated in the same division - and the winner should be rated higher than the loser - even after a close decision.
I will analyse this - thanks.
... and this is caused by the division change mechanism:
- the bout is assigned to light middleweight
- the boxers are assigned to middleweigt
- Foreman is rated higher at light middleweigt
- but - as boxers are assigned to middleweight - the transition lets Thompson the edge ...
This may be interpreted as: Foreman is regarded the better boxer at light middleweight - but Thompson still holds the edge at middleweigt.
I will both assign to light middleweight ...
The bout should probably be assigned to middleweight.
Posted: 18 Oct 2007, 17:39
by computerrank
... both are natural light middleweights from my perspective ...
Posted: 18 Oct 2007, 19:24
by Blue
Even though Foreman’s handlers have the ABCs rating him rated @ 154, neither fighter
has weighed @ that category in over 3 years. 
http://fightnews.com/rankings2.htm
Since both fighters weighed over 156 lbs, that bout is now correctly assigned to middleweight.
No matter what category U put them in, Foreman is still rated below Thompson.
http://www.boxrec.com/ratings.php?count ... o&pageID=2
Posted: 18 Oct 2007, 20:39
by JCS
Has to re-calculate overnight...
Even still, I think that should be a middleweight bout.
Posted: 22 Oct 2007, 09:21
by JCS
The method of implementing poor opposition quality is really showing itself as a black eye now.
Chris Byrd's opposition hasn't been that great as of late, but I notice today he has dropped all the way to #27 in the Heavyweight ratings. This will clearly diminish Povetkin's rise in the Heavyweight ratings if he were to win. However, if their match happened last week, it would have made a world of difference.
I realize that everything works itself out in the end, but ratings should look good at all times. Seriously, without winning or losing, how does a guy drop from #13, to #27, in one day? How is that realistic?
I strongly feel that consideration needs to be given with regards to converting these penalties into a more gradual process.
Posted: 22 Oct 2007, 12:23
by computerrank
... would make some 40 points difference for Povetkin ... perhaps
Posted: 22 Oct 2007, 12:26
by JCS
computerrank wrote:... would make some 40 points difference for Povetkin ... perhaps
Larry Donald's rating also dropped before Povetkin's win against him.
40 points can mean a dozen ranking spots or more.
Why not implement gradual penalties? They are undoubtedly more accurate and less anomalous in appearance.
Posted: 22 Oct 2007, 12:45
by m1kee50
I would have thought getting the rankings realistic was the priority, rather than worrying how quickly they look realistic.... so it changes? if its a more accurate representation, thats a good thing.
Posted: 22 Oct 2007, 12:56
by JCS
MIKEE wrote:I would have thought getting the rankings realistic was the priority, rather than worrying how quickly they look realistic.... so it changes? if its a more accurate representation, thats a good thing.
If a boxer boxes poor opposition for 17 months and 29 days, he loses nothing.. but as soon as he steps over that 18 month plateau, he can lose up to 50% of his points.
Is that accurate in any situation?
"More accurate" and "realistic" are one in the same now, because we abandoned the prediction rate and now seek the subjective acceptance. The "cliff" like penalties are ridiculous.
Posted: 22 Oct 2007, 13:02
by JCS
Let me humor you on the possibilities here. Tell me what you think is more accurate.
Current Inactive Penalty
Fighter 1 has been a dominant champion, but has been inactive for 17 months 27 days. After his last bout, where he beat the #2 contender, a quality opponent, he has a raw rating of 1500 points. If he were to come back today, he would go into the bout with the same exact rating of 1500 points.
However, if his bout was scheduled next week, rather than this week, he would go into the bout at 750 points. He would've lost half his points only for that extra week.
Proposed Gradual Penalty
Fighter 1 would enter this week's bout at say.. 755 points. If it were next week, 750 points.
In addition, I still feel the penalties are far, far too harsh. But that's a totally different story.
Proposed Gradual Solution w/ Revised Penalty
Penalties need to be phased in at a certain point.. and calculated by day, with no initial cliff-like penalty.
How about 1.25% per month starting at 9 months?
1.25% * (18-9) = 11.25
Fighter 1 loses 11.25% of 1500 = 1331.25 points
This way, a week or two can't make such a big difference.
Posted: 22 Oct 2007, 13:17
by m1kee50
I do apologise, I thought we were referring to a recalculation of all values....
as for the actual issue, I think it should be implemented from day one at a small percentage, maybe as some sort of exponential value
http://en.wikipedia.org/wiki/Exponential_function
I know we arent 'rewarding' frequent fighters, but we are punishing bad fighters, which is the other side of the coin....
Posted: 22 Oct 2007, 13:19
by JCS
MIKEE wrote:I do apologise, I thought we were referring to a recalculation of all values....
as for the actual issue, I think it should be implemented from day one at a small percentage, maybe as some sort of exponential value
http://en.wikipedia.org/wiki/Exponential_function
I know we arent 'rewarding' frequent fighters, but we are punishing bad fighters, which is the other side of the coin....
The current inactive penalty is exponential. Opposition quality penalty is based off of the best opposition encountered in 18 months (an idea I proposed -- before it was same as inactive penalty), but its not gradual at all. I think if a fighter encounters bad opposition they should enter into a gradual decay until they improve their opposition. Inactive penalties just hack off 50% at 18 months and get worse from there. This should be gradual as well, in my opinion.
Furthermore, in an MMA rating system I've been working on, I've found it extremely beneficial to implement a "loser cushion" so proven fighters don't lose as many points as an unproven fighter would in the event of a loss.
Posted: 23 Oct 2007, 02:59
by computerrank
JCS wrote:computerrank wrote:... would make some 40 points difference for Povetkin ... perhaps
Larry Donald's rating also dropped before Povetkin's win against him.
40 points can mean a dozen ranking spots or more.
Why not implement gradual penalties? They are undoubtedly more accurate and less anomalous in appearance.
... 40 points would make 1 or 2 ranks for Povetkin, if he should win.
The basic reason, why the penalty comes after 18 months completed is - a boxer's rating should be tangled only, as far as he has missed valid opposition within this period - to the last day.
Otherwise the implicit requirement would be to box this level of opposition in a shorter time period - and this is not intended.
And in addition - we would have daily sliding ratings - which are not accepted too, as we now.
Set-backs after periods for missing opposition quality or inactivity are quite usual in common ABC rankings too.