ShashChess

[amchess]

Hi, Khalid.

For the experience file, there can't be two entries for the same position's hashKey, because, if there is another (or the same) move, but with a greater depth, the entry is updated.
For the contempt, stockfish uses contempt here:

Code: Select all

Score score = pos.psq_score() + me->imbalance() + pos.this_thread()->contempt;

in evaluate.cpp (classical eval), but classical evaluation is used very rarely.
There was also a recent discussion on discord channel and it was confirmed this:
https://discord.com/channels/4359437104 ... 6409731082
Someone also proposed to completely remove it.
ShashChess, anyway, has always contempt to 0 (and also no dynamic contempt).
Infact, it must correctly define a position nature based on score and with contempt this is misleading.
After by choosing personalities, you can bootstraap an analysis/game play based on your needs.

Andrea

[khalidomar]

You are right about experience fragmentation. I forgot that in Shashchess and Brainlearn it is saving the whole experience file every time after each game. For that reason, it is doing the merging on the fly before it actually does the saving.

About Contempt, it is at least a debatable point, and of course, depends on the engine. I don't think the classical evaluation is used very rarely, but yes, it is used less often than before.

My idea is: If contempt was good for classical evaluation (before NNUE), then it will be also good for classical evaluation (after NNUE), even if classical evaluation is used less often.

I ran a quick test from the start position and found that classical evaluation is called 16% of the time. This percentage will increase in unbalanced positions and towards the endgame.

[criko]

Mac version - anybody please

here you go...
https://acepoint.de/

direct link ...
https://acepoint.de/home/wp-content/upl ... ess-16.zip

[amchess]

You are right about experience fragmentation. I forgot that in Shashchess and Brainlearn it is saving the whole experience file every time after each game. For that reason, it is doing the merging on the fly before it actually does the saving.

About Contempt, it is at least a debatable point, and of course, depends on the engine. I don't think the classical evaluation is used very rarely, but yes, it is used less often than before.

My idea is: If contempt was good for classical evaluation (before NNUE), then it will be also good for classical evaluation (after NNUE), even if classical evaluation is used less often.

I ran a quick test from the start position and found that classical evaluation is called 16% of the time. This percentage will increase in unbalanced positions and towards the endgame.

Hi, Khalid.
You can read the discussion on discord channel.
Anyway, this isn't a problem for ShashChess, as I explained you.

[amchess]

Mac version - anybody please

here you go...
https://acepoint.de/

direct link ...
https://acepoint.de/home/wp-content/upl ... ess-16.zip

Thanks a lot.
Andrea

[carldaman]

Contempt could be useful to the Tal personality, especially if the user chooses to turn NNUE off, since it can benefit from 'optimism' and draw-avoidance by its very nature. This would also apply to analysis done by the Tal personality.

[criko]

You are right about experience fragmentation. I forgot that in Shashchess and Brainlearn it is saving the whole experience file every time after each game. For that reason, it is doing the merging on the fly before it actually does the saving.

About Contempt, it is at least a debatable point, and of course, depends on the engine. I don't think the classical evaluation is used very rarely, but yes, it is used less often than before.

My idea is: If contempt was good for classical evaluation (before NNUE), then it will be also good for classical evaluation (after NNUE), even if classical evaluation is used less often.

I ran a quick test from the start position and found that classical evaluation is called 16% of the time. This percentage will increase in unbalanced positions and towards the endgame.

Is is possible to merge 2 or more exp files?
If YES ... how can I do this?

regards
Chris

[acepoint_de]

Mac version - anybody please

here you go...
https://acepoint.de/

direct link ...
https://acepoint.de/home/wp-content/upl ... ess-16.zip

Thanks a lot.
Andrea

It's the version for the new arm64, not Intel...just in case .

Ciao

acepoint

[Ozymandias]

Run a test with version 16 against github's version of FF2:

Code: Select all

   # PLAYER        :  RATING   POINTS  PLAYED   (%)  CFS(%)     W      D     L  D(%)
   1 FF2           :  2000.0  12643.5   25000    51     ---  3646  17995  3359    72
   2 Default       :  1998.0   2485.5    5000    50     ---   674   3623   703    72
   3 Tal           :  1997.2   2480.0    5000    50     ---   722   3516   762    70
   4 Tal+Capa      :  1996.6   2476.0    5000    50     ---   662   3628   710    73
   5 Capablanca    :  1996.1   2472.5    5000    49     ---   663   3619   718    72
   6 Petrosian     :  1991.9   2442.5    5000    49     ---   638   3609   753    72

White advantage = -1.56
Draw rate (equal opponents) = 50.00 %

Looks like Petrosian needs a hand.

[Cornfed]

Run a test with version 16 against github's version of FF2:

Code: Select all

   # PLAYER        :  RATING   POINTS  PLAYED   (%)  CFS(%)     W      D     L  D(%)
   1 FF2           :  2000.0  12643.5   25000    51     ---  3646  17995  3359    72
   2 Default       :  1998.0   2485.5    5000    50     ---   674   3623   703    72
   3 Tal           :  1997.2   2480.0    5000    50     ---   722   3516   762    70
   4 Tal+Capa      :  1996.6   2476.0    5000    50     ---   662   3628   710    73
   5 Capablanca    :  1996.1   2472.5    5000    49     ---   663   3619   718    72
   6 Petrosian     :  1991.9   2442.5    5000    49     ---   638   3609   753    72

White advantage = -1.56
Draw rate (equal opponents) = 50.00 %

Looks like Petrosian needs a hand.

Isn't this kind of a pointless test?
I mean, not all positions are 'Petrosian'...or 'Tal'...or...Capa or Tal+Capa positions. That is why for such a test ONLY the Default (which uses each personality as needed) makes any sense.

Even the real Petrosian could sac like a Tal when the position called for it.