SugaR AI ICCF

[M.Z]

https://github.com/Zerbinati/SugaR-AI-ICCF
SugaR AI ICCF 1.00

Release:
https://github.com/Zerbinati/SugaR-AI-ICCF/releases/tag/ICCF100

[peterm]

https://github.com/Zerbinati/SugaR-AI-ICCF
SugaR AI ICCF 1.00

Release:
https://github.com/Zerbinati/SugaR-AI-ICCF/releases/tag/ICCF100

Great news, many thanks!

Biggest wish of mine so far would be a Use NNUE Pure UCI option like in Cfish.

Anyhow glad to see SugaR back regards

[Engine]

https://prabusari.blogspot.com/p/raubfisch-x44annsl-ssse3-vs-sugar-ai.html?m=1

[timetraveller]

https://prabusari.blogspot.com/p/raubfisch-x44annsl-ssse3-vs-sugar-ai.html?m=1

SugaR ICCF to let play in a 3 minute game without time bonus is of course extremely meaningful wacko

[M.Z]

https://github.com/Zerbinati/SugaR-AI-ICCF
SugaR AI ICCF 1.10
Learning logic enhancement and full Linux support
https://github.com/Zerbinati/SugaR-AI-ICCF/commit/69a9171fe681cf48e0d5dddd68e2a807f353cfac
changed some logic to avoid experience book if doing infinite analysis
or searching for mate.
Khalid Omar authored
master AI110
@Zerbinati
Zerbinati committed


Release:
https://github.com/Zerbinati/SugaR-AI-ICCF/releases/tag/ICCF110

[peterm]

changed some logic to avoid experience book if doing infinite analysis

Good point!
My next one big wish besides Use NNUE Pure (or does it ICCF version anyhow per defaut?) would be:
Possibility to give different names to different experience- files like Eman has. Now one always has to shift different SugaR.exp files being trained by different games, openings, positions, to differntly named folders telling by name of directory what the file is for.
In Eman you just call the .exp- files differently and significantly as for their content and can leave all of them together with binary of engine and don't always have to delete one in root directory if you want the engine to ponder without any experience or create a new one.
That was a big advantage in handling of more then one experience file. Enjoy that feature in Eman much regards

[M.Z]

https://github.com/Zerbinati/SugaR-AI-ICCF
SugaR AI ICCF 1.20

Adjust definition of unsafeSquares

and adjust related bonus values. The bonus is now not given whenever
there is an enemy piece in front of the pawn.

Add 'Minimum Thinking Time' UCI option.
Release:
https://github.com/Zerbinati/SugaR-AI-ICCF/releases/tag/ICCF120

[M.Z]

As soon as I have a moment I will compile the new versions
Khalid fixed some important things in the learning code
However the updated sources are already available in the repositary.

[M.Z]

https://github.com/Zerbinati/SugaR-AI-ICCF
SugaR AI ICCF 1.30

Fix experience book problem if draw is detected
https://github.com/Zerbinati/SugaR-AI-ICCF/commit/5da9b1ff089f7d7a9f3b0b0bdcd24ff1c6551c77
Avoid no move situation when an experience move causes 3-fold or 50-move
draw.
master (#3)
@khalid-a-omar
khalid-a-omar committed
Fix exception when best move has depth zero
https://github.com/Zerbinati/SugaR-AI-ICCF/commit/ca0f06f8a18144614bb23cf26c0dd6c86d2cea6e
master (#1)
@khalid-a-omar
khalid-a-omar committed
Release:
https://github.com/Zerbinati/SugaR-AI-ICCF/releases/tag/ICCF130

[peterm]

https://github.com/Zerbinati/SugaR-AI-ICCF
...
Release:[/color][/b]
https://github.com/Zerbinati/SugaR-AI-ICCF/releases/tag/ICCF130

Thanks for the update, Marco!
Didn't notice until recently, contempt can be set to 1000, one question to that:
As far as I knwo, contempt at SF NNUE doesn't matter, if NNUE- eval is used.
How about that as for SugaR AI ICCF?
Does contempt influence eval only outside of hybrid- tresholds?
I don't see change of numeric eval in output in "balanced" positions within eval- range of SF- tresholds when changing contempt even to high values.
Is only output- eval unchanged or does contempt matter only in positions of hard coded ("classic") eval?

[peterm]

https://github.com/Zerbinati/SugaR-AI-ICCF
SugaR AI ICCF 1.30

Sorry for insisting, Marco, but I think the question yet could be of some interest to others too.
To ask once again, but this time in the shortest possible way:
Does contempt have any relevance in SugaR AI ICCF, if NNUE is used or only if it's not used (if NNUE is unchecked)?

[khalidomar]

Hi Peter,

In SugaR ICCF, the answer is No. Contempt isn't of relevance when NNUE evaluation is used.

However, it is different in SugaR AI (non ICCF), because the engine may utilize NNUE if the classical score is less than a certain threshold. Since classical evaluation contains contempt, it is possible for large contempt values to cause the engine to favor classical over NNUE. It is also possible for large negative values of contempt to make the engine utilize NNUE more often than usual.

[peterm]

Hi Peter,

In SugaR ICCF, the answer is No. Contempt isn't of relevance when NNUE evaluation is used.

However, it is different in SugaR AI (non ICCF), because the engine may utilize NNUE if the classical score is less than a certain threshold. Since classical evaluation contains contempt, it is possible for large contempt values to cause the engine to favor classical over NNUE. It is also possible for large negative values of contempt to make the engine utilize NNUE more often than usual.

Thanks for the prompt and detailed answer, Khalid!
Thought so, but wasn't sure. Pity in Sugar non ICCF 100 is the max. contempt, which isn't really much for SF dev, so it won't be for SugaR neither.
Good to know too, that SugaR ICCF doesn't use Hybrid- Eval and to know, that usage of NNUE in Hybrid- mode can be influenced by contempt, which I wouldn't have expected neither, at least not for SF dev.
So that would mean, contempt changes eval before tresholds get measured.
In this case, contempt enabled to be set higher than 100 would be fine for SugaR non ICCF and for Eman too, wouldn't it?

[khalidomar]

I believe the ICCF version of SugaR is meant for analysis and therefore it allows extreme values of contempt such as +/-1000 to give more control to the user.

But the normal version of SugaR (as well as Eman) is meant for normal gameplay and therefore only safe values of contempt are possible.

I do agree that larger contempt values may prove to be beneficial in certain positions, however, on average, the default contempt value is best (for normal gameplay)

[peterm]

I believe the ICCF version of SugaR is meant for analysis and therefore it allows extreme values of contempt such as +/-1000 to give more control to the user.

But the normal version of SugaR (as well as Eman) is meant for normal gameplay and therefore only safe values of contempt are possible.

I do agree that larger contempt values may prove to be beneficial in certain positions, however, on average, the default contempt value is best (for normal gameplay)

I know SugaR ICCF being meant for analysis and I wouldn't let an engine play with more then 100 normally. But from certain positions it's a kind of alternative to armageddon modus (komodo) too and especially for analysis it does help to find tricky solutions now and then sooner too.
SugaR ICCF won't benefit from higher values together with NNUE- usage ("Pure" I presume according to your explanation) SugaR non ICCF and Eman in hybrid eval and analysis would sometimes maybe.
Thanks again, dear Khalid regards