Page 1 of 1

Experiments and curiosities

Posted: Mon May 14, 2018 8:39 pm
by Rebel
Short articles about various aspects of computer chess.

http://rebel13.nl/rebel13/experiments.html

Re: Experiments and curiosities

Posted: Wed May 16, 2018 9:07 pm
by jstanback
Hi Ed!

I get similar results when running fixed depth matches between Wasp3 and Stockfish 8. Wasp wins 96% against SF at depth=4 :)

Code: Select all

depth wins draws losses  
 1     306   271    423  44.15%
 2     707    68    225  74.10%
 3     915    54     31  94.20%
 4     956     8     36  96.00%
 5     922    57     21  95.05%
 6     885    83     32  92.65%
 7     884    95     21  93.15%
 8     764   143     93  83.55%
 9     484   309    207  63.85%
10     340   439    221  55.95%
I didn't test at higher depths, but of course SF will soon start to dominate Wasp. Fixed depth tests against other strong engines are somewhat similar also. I guess SF plays so weak at low depths due to it's huge amount of pruning and reduction. But I haven't found pruning and reduction conditions for Wasp that give enough reduction in tree size to offset the mistakes ....

John

Re: Experiments and curiosities

Posted: Fri May 18, 2018 2:26 pm
by mhull
jstanback wrote:
Wed May 16, 2018 9:07 pm
Hi Ed!

I get similar results when running fixed depth matches between Wasp3 and Stockfish 8. Wasp wins 96% against SF at depth=4 :)

Code: Select all

depth wins draws losses  
 1     306   271    423  44.15%
 2     707    68    225  74.10%
 3     915    54     31  94.20%
 4     956     8     36  96.00%
 5     922    57     21  95.05%
 6     885    83     32  92.65%
 7     884    95     21  93.15%
 8     764   143     93  83.55%
 9     484   309    207  63.85%
10     340   439    221  55.95%
I didn't test at higher depths, but of course SF will soon start to dominate Wasp. Fixed depth tests against other strong engines are somewhat similar also. I guess SF plays so weak at low depths due to it's huge amount of pruning and reduction. But I haven't found pruning and reduction conditions for Wasp that give enough reduction in tree size to offset the mistakes ....

John
I ran some tests with SF8 (cfish) on Mac IIsi (MC68030 @20Mhz), which at 40/120 time control would reach between 6 and 12 ply depths, depending on the position. I ran it against Crafty, running on another machine at fixed depth 9 (moving instantly of course). There seemed to be some parity in strength there. Some estimated SF8 on that hardware at around 2600 Elo based on average NPS (300-400) over those games.