New tournament form

[hgm]

I have an idea for a new form of conducting a tournament. It is especially useful if there are only a limited number of participants, that are spread out over a large Elo range.

The idea is to subdivide the participant into classes, based on their strength. But to have a large enough variation of opponents, every participant plays everyone else, also the participants that belong in other classes. But to make the games a meaningful test of strength, rather than the obligatory collection of a 1 or 0, when participants from different classes play each other, the strongest one gets a time-odds handicap. The magnitude of the handicap is determined by the number of classes that seperates them.

For example, if a participant from the 3rd class plays against one from the 6th class, he gets a time-odds factor of 2^(6-3) = 2^3 = 8. Thus the 6th classer plays, say, 40/40', and the 3rd classer 40/5'. (The weakest player always plays the nominal time, to avoid extremely short games if two heavily handicapped engines meet each other.) As a factor 2 in time is supposed to change the engine strength by abot 70 Elo, it should be possible to handicap all engines into a range of ~70 Elo. That means that the expected scores in every game should be between 40% and 60%, i.e. none of the results wil be a 'dead cert', and all tests are meaningful.

To assign the classes, a normal promotion / degradation system can be used: those who score above 60% from all their games promote to the next higher class, those that score less than 40% to the next lower class. The over-all champion is of course the engine that ends highest of all first-class engines, even if some engines from lower classes score more points. As these engines benifited from time odds, that is not much of an achievement, after all. Their use was to provide a large-enough gauntlet for the first class engines to make a meaningful and reliable discrimination by strength.

[swami]

I don't know of any GUI that allows automated tournaments with time handicap for particular group of engines. Definitely not in Arena and Fritz, we have to manually do it.

[hgm]

PSWBTM + WinBoard_F would.

I will have to make a small change to WinBoard_F to make it exactly work as described, though. In the time-odds tourneys I did so far, if two handicapped engines met, they both played at the reduced time. But it seems more logical to renormalize such a case.

Another way to do it, would be to normalize the total time, rather than the longest of the two. Perhaps I should add an argument /timeOddsMode=N, where you can choose how to handle the case where botth /firstTimeOdds and /secondTimeOdds are set different from 1.

[Dr.Wael Deeb]

All this idea about the time-odds handicap games is not my cup of tea,sort them into classes,put them in their own classes and collect the damn results,why to complicate things

[Matthias Gemuh]

I don't know of any GUI that allows automated tournaments with time handicap for particular group of engines. Definitely not in Arena and Fritz, we have to manually do it.

I may implement this in ChessGUI in some future version.

Matthias.

[Matthias Gemuh]

All this idea about the time-odds handicap games is not my cup of tea,sort them into classes,put them in their own classes and collect the damn results,why to complicate things

Does it mess up the absolute elo ratings ?.

Matthias.

[hgm]

All this idea about the time-odds handicap games is not my cup of tea,sort them into classes,put them in their own classes and collect the damn results,why to complicate things

Like I said, this is mainly useful if you have not enough engines of similar strength to make a meaningful class of 16-20 engines. Perhaps because you have only 20 engines in total, and some are rated 2800, while others are rated 1800.

With over 500 engines for Chess you probably wouldn't have that problem. But it might already be not so simple if you wanted to make a league for SMP engines. And for Gothic Chess I currently only have 8 engines, spread out over a range of 600-800 Elo.

[hgm]

Does it mess up the absolute elo ratings ?.

The results of engines within the same class will be directly translatable into relative Elo ratings, as they played the same opponents under the same conditions. Between classes it would be more difficult, as it is not sure that every engine will lose exactly 70 Elo for every factor 2 of handicapping. So purists might argue that every differently handicapped version of an engine should have its own independent rating.

But any league system will have a similar problem, to gauge the ratings obtained in different classes w.r.t. each other. Using a few promotees / demotees for this, is also not very accurate.

In fact, I would have more confidence in the absolute rating difference between two engines if I knew that one of them needs a time handicap factor of 1000 to perform equal to the other (against the same set of opponents), then when it had to be derived in an indirect chain of intermediate-strength engines.

[Dr.Wael Deeb]

All this idea about the time-odds handicap games is not my cup of tea,sort them into classes,put them in their own classes and collect the damn results,why to complicate things

Like I said, this is mainly useful if you have not enough engines of similar strength to make a meaningful class of 16-20 engines. Perhaps because you have only 20 engines in total, and some are rated 2800, while others are rated 1800.

With over 500 engines for Chess you probably wouldn't have that problem. But it might already be not so simple if you wanted to make a league for SMP engines. And for Gothic Chess I currently only have 8 engines, spread out over a range of 600-800 Elo.

Even though,I don't trust this method,how do you know the exact amount of time that you must handicap the stronger engines with

As for the 20 engines rated 2800 Elo,look here at my currently running tournament:

_micro-Max 4.8 i
Alf 1.08
BiBiChess 0.9 i
ChessAlex 1.4 r3
ChessAlex 2.0 r4
Clueless 1.4 i
FireFly 2.2.0 MP
FireFly 2.5.0 MP
Gullydeckel 2.16 pl1 i
Holmes 0.8.20m
Monarch 1.7 i
Smash 1.0.3 MP i
Warrior 1.0.3

Two games every engine against all the other engines,156 games in total at long time control....

[Dr.Wael Deeb]

All this idea about the time-odds handicap games is not my cup of tea,sort them into classes,put them in their own classes and collect the damn results,why to complicate things

Does it mess up the absolute elo ratings ?.

Matthias.

Definitely yes