Author: Andrew Platt
Date: 16:38:31 09/17/04
Go up one level in this thread
On September 17, 2004 at 18:04:48, martin fierz wrote: >2) when i receive a ponderhit command but have used a lot of time, i abort the >search. so my PV is now reduced to crap (except for the first ply) and my engine >ponders on a garbage move. is there any more elegant/better way to solve this >than memorizing the PV of the last ply? for instance, i might have had a >fail-high for a new root move, and i don't construct PVs in the middle of the >search, only after every completed ply. but i should construct PVs all the time >(i.e. on every ply, and on every new best move at root level) if i want to >ponder, correct? Like Dann said this is no different from regular abort. What you do is just unwind the stack and not update the ply 0 PV which you have from the previous iteration. I also create a one move PV if I fail high figuring that that's probably the move to make. So far that hasn't caused me problems but my engine isn't very strong! The other thing you need to consider with ponder is that you might not have a move to ponder on if you had a truncated PV. I just do a shallow search in that case. Andy.
This page took 0 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.