Push a patch

[Lyudmil Tsvetkov]

Joerg, is it possible to widen the tree somehow, when a 0.0 score is reached in search. Only in that case?

I have done this on Andscacs long ago, and it was good. Of course the testing field was a gauntlet of different opponents, not selftest.

Meaning that on fishtest, where you match 2 equal engines, such a patch will have zero elo impact, or even might be a regression, but, if it is 0 elo patch, it is good to implement, as this will improve SF score against weaker opponents without involving any form of contempt.

Anything better than that?

[cdani]

Joerg, is it possible to widen the tree somehow, when a 0.0 score is reached in search. Only in that case?

I have done this on Andscacs long ago, and it was good. Of course the testing field was a gauntlet of different opponents, not selftest.

Meaning that on fishtest, where you match 2 equal engines, such a patch will have zero elo impact, or even might be a regression, but, if it is 0 elo patch, it is good to implement, as this will improve SF score against weaker opponents without involving any form of contempt.

Anything better than that?

Well, with a bit of luck, also will be good on selftest.

[Lyudmil Tsvetkov]

Joerg, is it possible to widen the tree somehow, when a 0.0 score is reached in search. Only in that case?

I have done this on Andscacs long ago, and it was good. Of course the testing field was a gauntlet of different opponents, not selftest.

Meaning that on fishtest, where you match 2 equal engines, such a patch will have zero elo impact, or even might be a regression, but, if it is 0 elo patch, it is good to implement, as this will improve SF score against weaker opponents without involving any form of contempt.

Anything better than that?

Well, with a bit of luck, also will be good on selftest.

I came to the conclusion, that in computer chess, you need luck for everything.

[cdani]

I came to the conclusion, that in computer chess, you need luck for everything.

When one defines a model statistically, it is that it has given up, at least for the moment, on defining it by reasoning. So mostly you can not predict the results of the changes, as you know.

[lucasart]

Lyudmil, do you wish a single patch to tune the pieces values+psqt tables?
Or 1 patch for the pieces values + 1 patch for the psqt, etc?
I ask this because Joerg is currently trying your values for the piece values but nothing about your psqt values for king middle game nor rook middle game-end game.

Please don't waste resources with Lyudmil's random tweaks. The current PST is incredibly well tuned, far beyond the reach of hand-tuning. What are the odds that, in a 334-dimensional space of parameters, a random move leads to a 1.5-2 elo gain (below that we can't distinguish from noise) ?

SPRT(0,4) is costly, and fishtest is low on resources, so I would appreciate if people do some homework (eg. SPSA tuning locally) before wasting resources with random tuning tests.

This is not aimed at you, but a general comment for all who throw random tuning patches at the framework.

[Isaac]

Please don't waste resources with Lyudmil's random tweaks. The current PST is incredibly well tuned, far beyond the reach of hand-tuning. What are the odds that, in a 334-dimensional space of parameters, a random move leads to a 1.5-2 elo gain (below that we can't distinguish from noise) ?

SPRT(0,4) is costly, and fishtest is low on resources, so I would appreciate if people do some homework (eg. SPSA tuning locally) before wasting resources with random tuning tests.

This is not aimed at you, but a general comment for all who throw random tuning patches at the framework.

No problem Lucas, I just wanted Lyudmil to clarify this point. I think there was a patch that passed the 2 SPRT with like 93 parameters tuned, that's basically the reason why I asked whether Lyudmil had in mind a huge general tweak or just several small and specific tweaks.
If I had the hardware I would maybe have tried the Lyudmil's big tuning and if I had a good score after over 1 k game I'd have tried it in fishtest. But my hardware is prehistoric.

[lantonov]

Joerg, is it possible to widen the tree somehow, when a 0.0 score is reached in search. Only in that case?

This will involve no contempt, but will avoid a fair portion of early draws, might even gain some elo?

Sorry for being so obtuse, you know I am a non-specialist.

At the moment Stefan is testing a patch that restricts pruning if at pv node and no preliminary best move is found. This patch is doing well at STC.

[Lyudmil Tsvetkov]

Joerg, is it possible to widen the tree somehow, when a 0.0 score is reached in search. Only in that case?

This will involve no contempt, but will avoid a fair portion of early draws, might even gain some elo?

Sorry for being so obtuse, you know I am a non-specialist.

At the moment Stefan is testing a patch that restricts pruning if at pv node and no preliminary best move is found. This patch is doing well at STC.

What is the connection?

[Lyudmil Tsvetkov]

As many tries have already been done on changing the scoring of penalties for pawns on squares the colour of the bishop in SF, and all of them more than convincingly failed, here what maybe makes the most sense without going to extremes and changing just the necessary minimum, as otherwise SF simply refuses to obey, if you introduce more sweeping changes, even if reasonable.

- leaving the way bishop pawns are scored now
- adding just one single specific case - double the penalty for a blocked central d4 or e4 pawn on square the colour of the bishop

I think this makes very much sense and is easy to do.

the d4 and e4 blocked central pawns are by far the most important for the evaluation of bishop pawns.

I have watched too many games where SF fails on that, as it makes no distinction between central and non-central pawns, blocked or not, but, as you know, SF does not like sweeping changes...

Still, I think this is a very reasonable rule.

So, in this case, in the specific case of d4 or e4 pawns blocked by enemy pawns the penalty will become 16;28 for the mg/eg instead of 8;14, if I remember correctly the values.

[Lyudmil Tsvetkov]

This rule is extremely important - an engine that does not consider properly at least blocked d4 and e4 pawns on squares the colour of the bishop often plays ugly chess, staying with a bad bishop, as SF frequently does in a variety of lines, although it tries to compensate for that by every imaginable tactical way...

But again, I have abandoned all hope for SF...