Newbie question about PVS zws verification

Discussion of chess software programming and technical issues.

Moderators: bob, hgm, Harvey Williamson

Forum rules
This textbox is used to restore diagrams posted with the [d] tag before the upgrade.
Post Reply
tpetzke
Posts: 686
Joined: Thu Mar 03, 2011 3:57 pm
Location: Germany
Contact:

Newbie question about PVS zws verification

Post by tpetzke » Sun Jun 19, 2011 4:05 pm

Hi,

I have a question about the researches in a PVS framework.

Situation:
I run a search up to 14 ply on the initial position (no book). I'm considering the moves and the returned scores at the root node.

The lower ply searches give e2e4 and sometimes d2d4 as best move. Ply 13 has e2e4 at the end.

So the Ply 14 search is started with e2e4 first. It gets a score of +9.
The other moves are searched with a 0 window. When it comes to d2d4 the 0 window search fails high with a score of +10, so a research is done and the research then only gives a score of +9 (so it does not increase alpha).

Is this a bug or just bad luck ? I read somewhere in post from Bob that false fail highs indicate a bug (but is a 0 window fail high a real fail high ?)

The research burned a lot of nodes (more than the 1.st move of that ply) and just lead to nothing.

My explanation for the research requiring more nodes is that the TT favors e2e4 because that was the pv at ply 13 and d2d4 did not cause a research on ply 13 so the TT contains probably less useful entries for d2d4. But that's just an idea.

Thomas...

User avatar
hgm
Posts: 24740
Joined: Fri Mar 10, 2006 9:06 am
Location: Amsterdam
Full name: H G Muller
Contact:

Re: Newbie question about PVS zws verification

Post by hgm » Sun Jun 19, 2011 6:50 pm

Such search instability is normal, when you use null-move pruning or hash tables. White might have used some null moves to push the score to +10, but when you open the window for the re-search, +10 is no longer enough to make these null moves fail high. So it has to search real moves in stead,and perhaps these were not so hot, because they look deeper, and bring good moves of the opponent within the horizon. With hash tables a re-search can use hash entries that did not exist yet in the null-window search, and look over the horizon through grafting (if you don't do exact-depth hits only).

tpetzke
Posts: 686
Joined: Thu Mar 03, 2011 3:57 pm
Location: Germany
Contact:

Re: Newbie question about PVS zws verification

Post by tpetzke » Mon Jun 20, 2011 6:42 pm

Thanks for your explanation, so if this is something that is about to happen from time to time I will try some ideas to lessen the effect when it happens.

I just did not want to cure only a symptom of a something that should rather be fixed :-)

Thomas...

Post Reply