Geschrieben von:/Posted by: Daniel Shawul at 03 May 2004 12:08:19:
Als Antwort auf:/In reply to: Re: DanChess 1.04d-beta available! geschrieben von:/posted by: Olivier Deville at 03 May 2004 10:54:36:
Hello
DanChess 1.04d beta version is available for download at my homepage.
This version is by far different from the previous version 1.04c.
1) 50% slower.
This is mainly due to attack tables. They are generated at every node including internal nodes.
2) Every node is evaluated.
3) no incremental move generation
4) book format is changed
5) experimental reductions like reducing losing captures??
I did this mainly to enhance the evaluation function and i hope i did. It usually scores equal against the fast 1.04c. I did a 100 games round robin b\n 1.04c and 1.04d at 40/3 control and score is 58.5-41.5 in favour of 1.04d. But it also looses more games against gothmog 0.4.7 than 1.04c.Being fast seem to be better against other engines. Gerhard Schawager made a lot of games with this version and his results are in favour of 1.04c. So strength is still in question. But i am still optimistic about this version. If only it is a bit faster in endgames it will defeat 1.04c convicingly. That's why i want to make a beta release first.
One other problem, this version once it loses a game on time it will,for some unknown reason, stop playing the next games. So if you got results like 30-1 against DanChess please check out the pgn.[unless ofcourse it is shredder
]
Testers please test.
thanks in advance
daniel
Hi Dan
I just downloaded the beta version. In the first game I try, as Black DanChess plays 1.e4 c5 2.Nf3 d6 3.d4 cxd4 4.Nxd4 Nf6 5.Nc3 (all right) and then 5..Bd7
Here are the statistics for 5th black move in the book :
The rather unusual 5...Bd7 is playable (2 draws and 1 win for Black, and Shirov couldn't win vs Bischoff in this variation), and maybe this is what you want DanChess to do. But 3 games are very few, compared to the 582 Najdorf variation games.
Olivier
>Move       Games  %     Av      Perf
>5...a6     582    44    2633    2585
>5...Nc6    198    37    2595    2504
>5...e6     58     42    2613    2554
>5...g6     41     29    2572    2433
>5...Bd7     3     67    2552    2734
>5...Qb6     1      0    2654    1798
>5...Nbd7    1      0    2542    1798
>
Hi Oliver
This one is a beta version I am still working on the book. But i think the book problem is solved.
Here is the book statistics after Nc3
move played + - = score learn sortv
a6 475 22.11% 33.05% 44.84% 44.42% 1.00 61.31
Nc6 189 18.52% 44.97% 36.51% 36.51% 1.00 33.21
e6 49 26.53% 36.73% 36.73% 44.90% 1.00 41.92
g6 41 12.20% 56.10% 31.71% 26.83% 1.00 4.68
Bd7 3 33.33% 0.00% 66.67% 66.67% 1.00 6.00
N8d7 1 0.00% 100.00% 0.00% 0.00% 1.00 7.00
Qb6 1 0.00% 100.00% 0.00% 0.00% 1.00 9.00
sortv is 6 for Bd7.
Also i think i can go back to the previous pgn collection with the new scoring method. For example the e4 a6?? problem is solved now
initial position
move played + - = score learn sortv
e4 5148 34.32% 20.12% 45.55% 57.09% 1.00 73.51
d4 4290 31.91% 18.46% 49.63% 56.71% 1.00 75.06
Nf3 918 29.30% 19.83% 50.87% 54.68% 1.00 49.75
c4 709 31.88% 21.44% 46.69% 55.15% 1.00 50.12
g3 76 38.16% 14.47% 47.37% 61.84% 1.00 7.00
b3 13 30.77% 23.08% 46.15% 53.85% 1.00 4.00
f4 5 40.00% 40.00% 20.00% 40.00% 1.00 9.00
Nc3 2 50.00% 0.00% 50.00% 50.00% 1.00 8.00
c3 1 0.00% 100.00% 0.00% 0.00% 1.00 6.00
after e4
move played + - = score learn sortv
c5 2425 20.41% 35.13% 44.45% 42.64% 1.00 59.17
e5 1342 19.00% 30.77% 50.22% 44.11% 1.00 50.11
e6 666 20.27% 37.24% 42.49% 41.44% 1.00 40.15
c6 420 22.86% 32.86% 44.29% 45.00% 1.00 38.00
d6 124 22.58% 34.68% 42.74% 43.55% 1.00 12.35
g6 73 21.92% 39.73% 38.36% 41.10% 1.00 0.00
d5 50 10.00% 52.00% 38.00% 28.00% 1.00 2.00
Nf6 39 7.69% 35.90% 56.41% 35.90% 1.00 8.00
b6 5 0.00% 60.00% 40.00% 20.00% 1.00 3.00
a6 2 100.00% 0.00% 0.00% 100.00% 1.00 4.00
Nc6 2 50.00% 50.00% 0.00% 50.00% 1.00 1.00
sort value for a6 is 4 so it will never be played despite the 100% score.
what do you think?
best wishes
daniel
