Thanks for supporting movei.Michael Sherwin wrote:Alaric
Bright
Movei
I am not very optimistic.
I could send later version of movei that is probably slightly better than 403 10 10 10 but I did not care about it.
my notes:
1)If you believe the ccrl list then 423 is slightly better than 403 10 10 10 but this is not enough improvement that I can consider as a big improvement(here is the ccrl data)
Movei 0.08.423 2691 +24 −24 52.3% −19.4 32.0% 575
90.6%
Movei 0.08.403(10 10 10) 2650 +25 −25 46.6% +21.7 26.0% 572
Main reason for improvement of 423 relative to 403 10 10 10 is better evaluation.
2)I started to work about using hash for pruning but unfortunately my tries did not work and I will need to look at the tree to find bugs.
I plan to do it step after step and the first thing is to find why my new replacement rules do not work even when I do not use pruning.
Movei by default use hash only for order of moves(even not to avoid null move pruning) and it only replace the entry in case of entry with bigger depth.
It always has a generation number of 16 bits that is updated in every search so practically movei does not use hash information from previous searches(I plan to keep it in the near future because I do not want to have bugs that I cannot reproduce)
Uri