idea: null-move analogy

Discussion of chess software programming and technical issues.

Moderators: hgm, Harvey Williamson, bob

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Michael Sherwin
Posts: 3016
Joined: Fri May 26, 2006 1:00 am
Location: WY, USA
Full name: Michael Sherwin

Re: idea: null-move analogy

Post by Michael Sherwin » Sun Mar 10, 2019 6:00 pm

xr_a_y wrote:
Sun Mar 10, 2019 9:06 am
Ok I found +20 elo with depth/2
depth/4 running ...
I tried depth/4 and it was not as good as just depth = 1. Do you notice a speed increase that could account for 20 elo or is something else happening? I suspect that something else is happening that takes advantage of an unmodified null move. This seems to be showing up in testing against some engines. Romi's best result ever against OliThink532 was 72%. After 28 games the score is 22w 2l 4d for 85.7%!
I hate if statements. Pawns demand if statements. Therefore I hate pawns.

User avatar
xr_a_y
Posts: 455
Joined: Sat Nov 25, 2017 1:28 pm
Location: France

Re: idea: null-move analogy

Post by xr_a_y » Sun Mar 10, 2019 6:08 pm

I don't rely on "speed" as knps a lot because the less I prune, the faster the engine seems to be. Minic can get up to 4Mnps if no pruning is perform, only 1.1Mnps with everything activated.
I your are talking time to depth, then, I didn't test... sorry.

at depth/4 I'm getting 105-81-238 right now, not really convincing, maybe +20 also ...

I'll try your depth=1 idea after that.

User avatar
xr_a_y
Posts: 455
Joined: Sat Nov 25, 2017 1:28 pm
Location: France

Re: idea: null-move analogy

Post by xr_a_y » Thu Mar 14, 2019 6:16 pm

This little test gives me another idea /question. Shall the returned score of the closed window qsearch done for razoring or the closed window little search done before nullmove be used to update evalscore? Evalscore is often used later to trigger other prunings or extension...

Maybe the evalscore can be replaced by a small iid search even at not pv nodes?

Michael Sherwin
Posts: 3016
Joined: Fri May 26, 2006 1:00 am
Location: WY, USA
Full name: Michael Sherwin

Re: idea: null-move analogy

Post by Michael Sherwin » Fri Mar 15, 2019 2:14 pm

xr_a_y wrote:
Thu Mar 14, 2019 6:16 pm
This little test gives me another idea /question. Shall the returned score of the closed window qsearch done for razoring or the closed window little search done before nullmove be used to update evalscore? Evalscore is often used later to trigger other prunings or extension...

Maybe the evalscore can be replaced by a small iid search even at not pv nodes?
That sounds reasonable. And then using that score should be a lot more accurate than using the on spot eval. I will give it a try.
I hate if statements. Pawns demand if statements. Therefore I hate pawns.

Post Reply