TalkChess.com

Laskos wrote:

Uri Blass wrote:

Laskos wrote:

IanO wrote:

carldaman wrote:What was wrong with Komodo's settings? People should at least be entitled to a proper explanation by the organizers.

The differences can be found in the initial game comment.

Before:

Code: Select all

BlackEngineOptions: Use Syzygy=true; Syzygy Probe Depth=2; Syzygy Probe Limit=6; Dynamism=110;

After:

Code: Select all

BlackEngineOptions: Contempt=7;

To summarize, removing custom Syzygy probe options and "Dynamism" (whatever that is), and adding Contempt 7.

It would be nice to hear from Mark what the effects of these changes are likely to be.

The values "Before" were close to to optimal. The ELO loss was no larger than 5-10 ELO points and hardly the reason for the lost game. Seems unfair to Houdini to replay the game.

I do not think that we know the elo loss at long time control and there is no basis to claim that the wrong setting is not the reason for the lost game.

I see nothing unfair for Houdini.

With your usual "we do not know anything" are you sure you usually see something at all? One can try to see things by a rough estimate. The drawelo here is about 250 ELO points, and if the differnce from best settings is 5 ELO points, then we have 98% that the wrong settings did not affect the result. Only if the wrong settings are affecting by an order of 250 ELO points Komodo, we can say that probably this was the cause for the loss. Is it plausible?

I do not think that the probability that the wrong setting effect the result is relevant and I do not think that we know the probability for changing the result by 5 elo reduction in s single game because it is not as simple as only changing good results to bad results and in some cases 5 elo reduction can give a positive effect and cause komodo to win instead of draw when it is a reduction only because the positive effect happens less often than the negative effect.

Mark, when you sent the new K10.1 to the TCEC Tournament Director for Stage 3, did you tell the Tournament Director the new Settings 7 and 115 before they started the first game ?

What is the default setting of K10.1 ?



The main question is: Is it the fault of Komodo Team or is it the fault of TCEC Team ?

Uri Blass wrote:

Laskos wrote:

IanO wrote:

carldaman wrote:What was wrong with Komodo's settings? People should at least be entitled to a proper explanation by the organizers.

The differences can be found in the initial game comment.

Before:

Code: Select all

BlackEngineOptions: Use Syzygy=true; Syzygy Probe Depth=2; Syzygy Probe Limit=6; Dynamism=110;

After:

Code: Select all

BlackEngineOptions: Contempt=7;

To summarize, removing custom Syzygy probe options and "Dynamism" (whatever that is), and adding Contempt 7.

It would be nice to hear from Mark what the effects of these changes are likely to be.

The values "Before" were close to to optimal. The ELO loss was no larger than 5-10 ELO points and hardly the reason for the lost game. Seems unfair to Houdini to replay the game.

I do not think that we know the elo loss at long time control and there is no basis to claim that the wrong setting is not the reason for the lost game.

I see nothing unfair for Houdini.

It is only fair if the game had been replayed no matter what.

That this game would have been replayed no matter what is far from clear. The rules are silent on this situation and what if it had been discovered only much later?

What if Komodo had drawn or won "despite its disadvantage"? I think most people would have considered it fair not to replay the game in that case.

Laskos wrote:

Branko Radovanovic wrote:

Laskos wrote:To see how disproportionate the action to replay the game is, for at most 5-10 ELO points loss in ONE game, they changed the result in the whole Stage 3 by 1 point between Komodo and Houdini, and that in 56 games each amounts to 13 ELO points in the WHOLE stage.

If you're saying that changing the result in the whole stage by 13 Elo points is a big thing, you're contravening your own argument.

The point you apparently missed was that a 5 ELO points mistake in a SINGLE game resulted in a 13 ELO points mistake for the whole stage.

And 13 ELO points mistake for the whole stage is a big thing, right? So if a small matter (5-10 ELO in one game) has significant consequences (13 ELO for the whole stage), then it's not a small matter after all, and your argument contradicts itself.

Also, you cannot argue that 5-10 ELO is a small difference that wouldn't change the outcome - because apparently it did.

Branko Radovanovic wrote:

Laskos wrote:

Branko Radovanovic wrote:

Laskos wrote:To see how disproportionate the action to replay the game is, for at most 5-10 ELO points loss in ONE game, they changed the result in the whole Stage 3 by 1 point between Komodo and Houdini, and that in 56 games each amounts to 13 ELO points in the WHOLE stage.

If you're saying that changing the result in the whole stage by 13 Elo points is a big thing, you're contravening your own argument.

The point you apparently missed was that a 5 ELO points mistake in a SINGLE game resulted in a 13 ELO points mistake for the whole stage.

And 13 ELO points mistake for the whole stage is a big thing, right? So if a small matter (5-10 ELO in one game) has significant consequences (13 ELO for the whole stage), then it's not a small matter after all, and your argument contradicts itself.

Kai's point is this:
- the disadvantage to Komodo of not replaying the game would perhaps be 0-5 Elo;
- the disadvantage to H5 of replaying the game is 13 Elo.

So complete justice cannot be done. Either way, one side is disadvantaged. But one could try to minimise the disadvantage.

Also, you cannot argue that 5-10 ELO is a small difference that wouldn't change the outcome - because apparently it did.

If you flip a coin twice, you can get two different outcomes even if you did not change the "settings" of the coin in between.

I think is indeed a very sad affair for chess computer again ...
I think if Mark don't want to let play the wrong version , He's right
May be it's because Komodo is commercial ? Jalousy ?
So I noticed since yesterday , the tcec's chat is very boring , with a lot of bad jokes and attacks against Mark , indeed this chat seems full of idiots...
We must not forget two things:
-chess is just a game
-tcec's players are just programs (full of 0 and 1 )
-chess programming is our hobby , just a hobby
Ok it is finaly 3 things lol
Ok at the end , I hope houdini is really improve by robert , for the suspense
I hope Andscacs perform very well in this stage
And at the very end , I'm very happy the monthly hgm's tournament exist, here is the real fun , no " head crash " kind guys etc ... Thanks Harm !
All the Bests for you
Dany

If a tournament has operators, representing the author, then if they set up the engine incorrectly then that is the operators problem.

When a tournament is run by another and the engine author requests certain settings then of course they have the right to insist these settings are used.

A possible solution is to ask authors for a compiled executable with all parameters preset.

syzygy wrote:

Branko Radovanovic wrote:

Laskos wrote:

Branko Radovanovic wrote:

Laskos wrote:To see how disproportionate the action to replay the game is, for at most 5-10 ELO points loss in ONE game, they changed the result in the whole Stage 3 by 1 point between Komodo and Houdini, and that in 56 games each amounts to 13 ELO points in the WHOLE stage.

If you're saying that changing the result in the whole stage by 13 Elo points is a big thing, you're contravening your own argument.

The point you apparently missed was that a 5 ELO points mistake in a SINGLE game resulted in a 13 ELO points mistake for the whole stage.

And 13 ELO points mistake for the whole stage is a big thing, right? So if a small matter (5-10 ELO in one game) has significant consequences (13 ELO for the whole stage), then it's not a small matter after all, and your argument contradicts itself.

Kai's point is this:
- the disadvantage to Komodo of not replaying the game would perhaps be 0-5 Elo;
- the disadvantage to H5 of replaying the game is 13 Elo.

So complete justice cannot be done. Either way, one side is disadvantaged. But one could try to minimise the disadvantage.

If that's Kai's point, it's still flawed: one cannot mix a priori and a posteriori effects. Komodo was harmed by 5 Elo or so a priori (in a probabilistic way - coin flipping is a good analogy), while Houdini was harmed 13 Elo a posteriori (once we flipped the coin and learned the result). Anyway, one can't mix apples and oranges: if we stick either to a priori or a posteriori view, Houdini was helped (or harmed) exactly the same as Komodo was helped (or harmed) in the other scenario (repeating vs not repeating the game), so there is no disproportionality whatsoever.

The only important difference between these two outcomes is that one is in violation of the rules, while the other isn't. It's as simple as that.

Harvey Williamson wrote:If a tournament has operators, representing the author, then if they set up the engine incorrectly then that is the operators problem.

When a tournament is run by another and the engine author requests certain settings then of course they have the right to insist these settings are used.

A possible solution is to ask authors for a compiled executable with all parameters preset.

This seems a good practical suggestion.
PK

Branko Radovanovic wrote:If that's Kai's point, it's still flawed: one cannot mix a priori and a posteriori effects. Komodo was harmed by 5 Elo or so a priori (in a probabilistic way - coin flipping is a good analogy), while Houdini was harmed 13 Elo a posteriori (once we flipped the coin and learned the result). Anyway, one can't mix apples and oranges: if we stick either to a priori or a posteriori view, Houdini was helped (or harmed) exactly the same as Komodo was helped (or harmed) in the other scenario (repeating vs not repeating the game), so there is no disproportionality whatsoever.

I agree that the two probably cannot be compared so easily. And I would say the disadvantage to Houdini is not 13 Elo but half a point.

The way I see it, each game is a coin flip. K's side of the coin may have slightly better odds with the right settings, but only slightly. But now K may have had the benefit of flipping the coin twice and taking the best of the two, which is grossly unfair. Unless the game would have been replayed no matter what.

So, to me, the real question is: what would have happened if the mistake was caught only considerably later and/or Komodo had won or drawn the game. How likely is it that Komodo would have had to replay the game if it had won it, despite its incorrect settings? I don't see that as particularly likely.

The only important difference between these two outcomes is that one is in violation of the rules, while the other isn't. It's as simple as that.

Which rule? Or rather, which rule guarantees that the game would have been replayed no matter what? There is not actually such a rule.

The rules say that if the server crashes, then the game is replayed from the start (unless a 6-men position had been reached). So that's very clear. But it does not apply to this situation. That this rule would be applied by analogy no matter what is not clear at all.