pohl4711 wrote: ↑Sat Oct 08, 2022 1:25 pm
michaelnill wrote: ↑Sat Oct 08, 2022 10:05 am
pohl4711 wrote: ↑Sat Oct 08, 2022 6:43 am
ShashChess24 High Tal is weaker in all categories: Less sacrifices, less short wins, more bad draws than Stockfish 15...
So, a testrun for my SPCC-ratinglist does not make sense for me...
Now this I find very interesting, did you make sure to disable NNUE on ShashChess24? My apologies for never mentioning it but with engines that make use of NNUE (such as Dragon, ShashChess) this feature need to be disabled in order for the "personality" to take over.
Other than that I have nothing to add, I am in pure disbelief to see SF 15 be more aggressive than ShashChess with HighTal personality.
Oh. No, I did not disable the nnue-net. My bad (but for my excuse: KomodoDragon switches the net off automatically, if "aggressive" is chosen). I will retry. But then ShashChess will be much weaker. So, vs. SF 15 makes no sense. I will try Stockfish final HCE as opponent.
Stockfish final HCE was too strong, so I used Rebel 15.1a as opponent. ShashChess HighTal was around +40 Elo stronger. 3500 games (singlethread 60sec+600ms, balanced Feobos openings)
Here the result of my EAS-Tool:
Code: Select all
*****************************************************************************
*** Evaluated file: resultsShash.pgn ***
*****************************************************************************
bad
Rank EAS-Score sacs shorts draws Engine/player
-------------------------------------------------------------
1 59413 10.82% 34.89% 27.47% "ShashChess24 HighTal"
2 54527 21.50% 08.72% 19.21% "Rebel 15.1a avx2"
*****************************************************************************
*****************************************************************************
*****************************************************************************
*** 2nd Ratinglist with all stats in percent-values *************************
*****************************************************************************
bad
Rank EAS-Score wins all sacs sacsQ sacs5+ sacs4 sacs3 sacs2 sacs1 all shorts short40 short45 short50 short55 short60 draws Engine/player
--------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
1 59413 1026 10.82% =[00.00% + 00.00% + 00.39% + 00.29% + 01.56% + 08.58%] 34.89% = [01.75% + 03.90% + 07.50% + 10.33% + 11.40%] 27.47% "ShashChess24 HighTal"
2 54527 642 21.50% =[00.00% + 00.16% + 00.47% + 02.02% + 04.83% + 14.02%] 08.72% = [00.47% + 00.78% + 01.09% + 02.18% + 04.21%] 19.21% "Rebel 15.1a avx2"
We see a change here (with nnue off): The number of short wins is extremly high, now. Thats good. But, sadly, the number of played sacrifices of Shash is very low, which is definitly no "Tal-style". And the number of bad (=early) draws is very high - bad, too.
In combination, these 3 parameters give an EAS-score, which is nearly equal to Rebel 15.1a. Thats OK, but definitly not overwhelming. If you look in my EAS-Ratinglist, Rebel 15.1a is on rank 16 and far behind (points!) the really aggressive playing engines (Velvet 4.1.0, Danasah 9, KomodoDragon aggressive-setting...).
Perhaps, I will try ShashChess HighTal in my SPCC-ratinglist, but I do not expect a really good EAS-ratinglist result for this setting...