Life In 19x19

RobertJasiek wrote:What do you mean with "weighted by significance"? Which significance?

later rounds have higher significance than earlier rounds?

What I mean is,

if we take a game to be a test, each test gives us some information about the player, there is a pairing based on that information, which leads us to the second test. That test now contains some of the information of the previous round of both players (limited by the pairing algorithm).

As you continue, later rounds contain more and more information. Hence, you should be able to develop a value for the significance of each round. In my previous example, the player's score at the end of the 5 rounds is 4-1.

now suppose he's tied with another 4-1. A very simple weighting algorithm (and probably wrong, since the significance of certain bouts may be reduced by other factors) would be the following:

0.5 0.6 0.7 0.8 0.9

which by a weighted average would give the player a tiebreaker score of 0.8
in the unlikely circumstance that two players share the same weighted score, you then do the sum of opponents scores, weighted by the same values that the wins were.

A more likely outcome would use the shannon information, you ask, based on the comparative ratings (your previous knowledge of both players), how much can this pairing actually tell us?

Then one wins and now you compute new performance ratings. You then attempt to maximize the information gleaned out of the next round, by judicious pairing.

so for this particular player, based on the field, suppose the information gained in each round looked something like this:
(remember, this player went 1,1,0,1,1)

0.4 bits | 0.8 bits | 0.5 bits | 0.9 bits | 1 bit

so in the first round he faced someone he overmatched, so we get very little information out of it, in the second the pairing was close, in the third there was a gap in player strength, he rose "too high" as it were. In the fourth he had a very well placed match, and in the 5th he had a perfect match.

so the cumulative information would be:

0.4 | 1.2 | 1.7 | 2.6 | 3.6

and his weighted average would be 0.62


A tournament is a test, the goal is to structure the test to get the most information out of it, as well as to interpret results based on how much information we gleaned from them.

breakfast wrote:I am sure, that making supergroup smaller solves all our problems with SOS/SODOS champions.
I will vote against any new systems (cannot vote, actually, but I prepared a tesuji). I will be Belorussian representative this time
Someone has to defend the interests of top Europeans. I will be the first one!

Well, I don't support a reduction in supergroup size.

Reducing the supergroup size may be in the interests of the top (known) European players - but it would reduce the chance of an outsider becoming champion I believe.

For me, as a participant and spectator and with the interests of as successful an EGC as possible, I would like the restrictions on Asian players in the supergroup to be relaxed.

I feel that somehow we should also reward top players who attend a lot of European events.

I also liked the suggestion that lightning matches be used to resolve drawn places.

shapenaji, the simplest tiebreaker according to your idea that later rounds have monotonously increasing significance is IROS: r points in round r. However, it is wrong to assume such. E.g., a player having already beaten all the top competitors then necessarily is paired against weaker players again.

RobertJasiek wrote:
Currently the EGF Rules Commission consists of Matti Siivola (chairman, FI) and me.

That is not very much, that is not even one representative per member organisation.

RobertJasiek wrote:
If you want to suggest that there is a conflict of interests between the strong players among the Tournament Supervisors and objective supergroup forming: Yes, there is. We have been aware of it and usually have let third persons double check when some of us supervisors was at the lower boundary of a supergoup. If you consider that insufficient, then please propose capable and rules-firm weak players who want to do the job of the Tournament Supervisors! It would save me of 8 - 20 hours of work every year.

I did some thinking and allready alone in the UK I can find 6 candidates (not counting myself) who are capable for different parts of a tournament supervisor job.
(but for some problems might need a strong player)

I strongly suspect that the only reason that that commission is so small is that no one else wants to do that job.

RobertJasiek wrote:shapenaji, the simplest tiebreaker according to your idea that later rounds have monotonously increasing significance is IROS: r points in round r. However, it is wrong to assume such. E.g., a player having already beaten all the top competitors then necessarily is paired against weaker players again.

Sorry Robert, I know it's been a while, but I was thinking about this.

If we define a concept the "information gained from a match", (For which we use the Shannon's information based on the probability of winning given the current tournament ratings in that round), to weight the monotonically increasing values, then if a player runs out of players his own strength and has to start playing weaker players, the weighting would be dropped by using the corresponding drop in the information gained from the match.

Of course this would mean that in order for players to have an opportunity to get the best tiebreaking scores they can, the pairing software should act in such a way as to maximize the Shannon information gained from each round.