Accelerated vs Abominated Swiss Pairings :)

[Garvinator]

In testing, the following system, which he called the Baku System, performed extremely well for events of 9+ rounds:

* Give top-half players one notional bonus point for rounds 1-3
* Reduce the notional bonus to half a point for 4-5
* Remove the notional bonus point for remaining rounds.

He also found that one notional bonus point for rounds 1-2 with half a bonus point for 3 worked quite well.

I will admit I have not read the report as I have just a couple of questions at this stage. Was there any discussion or study of the effect of:

Retaining the notional one or two bonus points for the top half players until all top half players have dropped at least half a point, or until only one player is still on a perfect score?

Does the effect of the Baku System change depending on the rating range of the field? And yes I am thinking of the monster swiss weekenders we have here in Australia, where players are rated from 2500 to unrated and totalling sometimes 80 or so players.

[Kevin Bonham]

Retaining the notional one or two bonus points for the top half players until all top half players have dropped at least half a point, or until only one player is still on a perfect score?

As far as I know, no contingent system of this kind was modelled.

Does the effect of the Baku System change depending on the rating range of the field? And yes I am thinking of the monster swiss weekenders we have here in Australia, where players are rated from 2500 to unrated and totalling sometimes 80 or so players.

I asked this question too. The test tournaments include similar events so it does work for monster swisses like ours. It might not be suitable for 7 rounds though.

[Chigoresov]

So acceleration for 7 Round tournaments is dubious and for 5 Round tournaments is just silly, right?

[jammo]

So acceleration for 7 Round tournaments is dubious and for 5 Round tournaments is just silly, right?

Looks like a bad case of asking for the result that you want. So far as I can see the conclusion of the report relates to events of 9 rounds or more but does not draw a conclusion about events with fewer rounds.

[Otto]

Hi!
I had the presentation in Baku, and will answer questions about the Baku acceleration and other questions.

In a previous post it is described a system “Retaining the notional one or two bonus points for the top half players until all top half players have dropped at least half a point, or until only one player is still on a perfect score”. Does it exist a more detailed description of the method, or is it actually two methods, one with one points added, and another with two points added. I will run a test on the method.

The Baku acceleration was tested against a large number of different tournament types, and performed better for all tournaments we tested. The largest test set was 432 players with rating from 2700 to unrated.

Since the aim for this work was to have an accelerated system that could be used in title tournaments, the system was not tested on 7 rounds tournaments.
From my experience I will guess that
Give top-half players one notional bonus point for rounds 1-2
Reduce the notional bonus to half a point for 3-4
Remove the notional bonus point for remaining rounds.
will work OK, but it still to be proved.

For 5 rounds I doubt any method will work good. A scheme with 1+1+1/2 may work, but only two rounds without acceleration are too short to smooth the effect of the acceleration points. Anyway it may be interesting to test it out.

[Kevin Bonham]

So acceleration for 7 Round tournaments is dubious and for 5 Round tournaments is just silly, right?

Any form of acceleration for a 5 round event with a very large number of entrants should reduce the risk of having multiple winners who have not played each other. But for the "abominated" system the chance is too high that the winner will have got there with the assistance of a mismatch win in round 3, which not only gives them an easy point but also gives them an advantage in freshness for round 4. It is better even to have multiple winners who do not play each other than to allow this risk.

The use of the reduced half-point benefit for round 3 for a 5-rounder is worth trying but even then I think there could be issues. There is still some chance that a bottom-halfer gets to 3/3 playing only other bottom-halfers along the way. For instance in the 2010 Ballarat Begonia (an extreme case) nine of the 22 players on 2/2 were from the bottom half. Only four of the ten players on 1.5 were from the top half, and one of these would have had to play a floater from among the 13 top-halfers on 2. So the notional 2-point group for pairing round 3 would have three top half players and the nine bottom-half players, forcing at least three bottom-half vs bottom-half pairings in the 2/2 group.

There also seems to be a big problem with half-point add-ons using Swiss Perfect. It seems to only accept whole point add-ons, and to process half-point add-ons as zero. It would be useful to know how many other pairing programs only accepted full point add-ons.

[Kevin Bonham]

Hi!
I had the presentation in Baku, and will answer questions about the Baku acceleration and other questions.

In a previous post it is described a system “Retaining the notional one or two bonus points for the top half players until all top half players have dropped at least half a point, or until only one player is still on a perfect score”. Does it exist a more detailed description of the method, or is it actually two methods, one with one points added, and another with two points added. I will run a test on the method.

I think Garvinator actually meant to write "“Retaining the notional one or two bonus points for the top half players until all bottom half players have dropped at least half a point, or until only one player is still on a perfect score”. This captures some of the principle of Stewart Reuben's system from his Arbiter's Handbook first edition while being easy to implement with a computer. (By the way Stewart actually told me that he started his version because he at first misunderstood the concept of the Haley (bonus point) system!)

If I am right then I have used what Garvinator is suggesting in some local tournaments, but the fields are so small that it is rare to have to keep the acceleration on for round 3.

I suggest testing it with one bonus point rather than two retained for the top-half players, and with the bonus point taken off once any of the following conditions are reached:

* There are zero bottom-half players on 100%
* There is only one player on 100%
* There are three rounds to go

[Otto]

I did the analyses with both 1 and 2 accelerate points.
The first test is performance. With 108 linear distributed players we may expect a player to have a better score in the tournament if (and only if) he has a higher rating. Acceleration with 1 point performs good, but the method with two points has a big jump on the border between accelerated players and unaccelerated players.
performance.jpg
The second test is a huge tournament with 432 players with range from 2700 to unrated. The rating distribution will reflect a real tournament, with most of the players in the range 1800-2200 (mixed Gaussian distribution). We will look at the percentage of games played with rating difference > level. The x-axis shows the level, and y-axis the percentage. The acceleration with 1 point performs OK, and the acceleration with 2 point behaves bad.
meaningless.jpg
A round per round analyses shows that acceleration with 1 point is all over good. It seems that acceleration will last for three or four rounds, and then the average rating difference increase in round 5. Acceleration with 2 points will last for four or five rounds, and round 6 will have a mean rating difference on 400 rating points. Not good at all.
oppoment.jpg
The last analysis is a probability for achieving a GM norm. It’s based on distribution om GMs, IMs and FM’s in a tournament, where we analyze rank x, and calculate the probability for he/she will get a GM norm.
norm.jpg
Conclusions:
Australian acceleration with 1 acceleration point performs OK. The disadvantage is the big mean difference in rating in round 5.
Australian acceleration with 2 acceleration points should not be used. It is not fear, and it performs bad.

[Kevin Bonham]

Thanks very much for posting this, Otto.

For comparison people should see the chart on page 4 of Otto's report at http://pairings.fide.com/images/stor...pairing-v3.pdf

The red line "Mean Haley" is the system used at Begonia and other such events, where the acceleration is taken off for round 3 no matter what is happening.

Under "Mean Haley" the average rating difference is well over 400 points in round 3. In "Australian 1" there is a peak after round 5, making it less good than the Baku system in terms of avoiding mismatches, but it is nowhere near as bad as if acceleration is always taken off after round 3.

[pappubahry]

I've done a similar sort of analysis to Otto but looking at small, Western-Australia-size fields (20-40 players) for 6-round weekenders. A long discussion and some interactive tables (so many numbers...) is here.

My main conclusions are that, if all players are correctly seeded, then one point for the first three rounds followed by half a point for round four (almost Baku) is the best of the non-permanent acceleration methods that I tested -- it reduces the mean rating difference across all games by the most, and generally introduced the least amount of inequality of the mismatches faced by the stronger players.

But all forms of acceleration do introduce that inequality. And furthermore, the longer the acceleration lasts, the worse it is for handling a strong player incorrectly seeded into the bottom half of the field -- that player gets shielded from the top players until late in the tournament, leading to an abnormally high percentage of tournament wins or top-3 finishes.

Even though the mismatch inequality provides an advantage to the strong players who get those extra mismatches (and such victories might therefore be unsatisfying, at least to those who missed out!), overall each of the top players won a similar percentage of the time regardless of the acceleration method. So I would say there is still arguably a place for acceleration in "low-stakes" events where you want somewhat more evenly-matched games.

[Kevin Bonham]

^^^

Excellent work.

These two conclusions are rather bleak for acceleration as a fair system, especially taken in tandem:

Using any sort of (non-permanent) acceleration leads to a more unequal distribution of mismatches, so that some top players get an easier path to a high score than others. The least unequal tested was (again) 1-1-1-0.5-0-0.

If a strong player is mistakenly seeded into the bottom half of the field, the 1-1-1-0.5-0-0 method was the worst-performing method, in the sense that the under-rated strong player becomes much more likely to win or finish in the top three than a correctly-seeded player of the same strength. This is because the wrongly-seeded player gets easier games, being shielded by the acceleration from facing the top players until very late in the tournament.

The Reuben-book method (Australian 1 in Otto's tests) can run into a similar problem to the second if the under-rated strong player is actually extremely strong (the player keeps winning against weaker opponents than they should be getting, and the acceleration doesn't end until they stop), hence the supreme importance of accurate ratings.

If all players are correctly seeded, then the distribution of winners is usually* approximately the same under any of the acceleration systems tested. i.e., a top seed who wins 40% of simulated Swisses without acceleration wins about 40% of simulated Swisses with acceleration, whatever the method.

This is interesting. I had previously suspected otherwise based on a trivially small number of simulations so it is good to know that that at least is not a problem.

[Garvinator]

In this thread, has there been much, or any analysis, on the Fide Baku acceleration system for 80 to 100 players with 6 rounds where the top players are likely to be rated around 2400 and there will be a long tail of unrated or very low rated players?

This does not seem to be covered, but is rather relevant to Qld weekenders.