. But there is no GUI to do this automatically (?) so I counted solved positions manually. Note that if 2 move had same score I counted it as solved. Questions to answer: is MV=3 still better? And how about other engines?Are engines really better in analysis with MV=2 !?
Moderators: hgm, Rebel, chrisw
-
Jouni
- Posts: 3283
- Joined: Wed Mar 08, 2006 8:15 pm
Are engines really better in analysis with MV=2 !?
I was curious and tested Crystal in Hard-Talkchess. One minute and 4 cores. In normal mode Crystal solved 94 positions. With 2 best lines it solved 110 . But there is no GUI to do this automatically (?) so I counted solved positions manually. Note that if 2 move had same score I counted it as solved. Questions to answer: is MV=3 still better? And how about other engines?
. But there is no GUI to do this automatically (?) so I counted solved positions manually. Note that if 2 move had same score I counted it as solved. Questions to answer: is MV=3 still better? And how about other engines?Jouni
-
yurikvelo
- Posts: 710
- Joined: Sat Dec 06, 2014 1:53 pm
Re: Are engines really better in analysis with MV=2 !?
Task solving and elo play strength contradict each other.
Tasks - are specially selected positions (1 in a billion) where traditional human or alpha-beta approach "stuck", that's why they are considered interesting.
If you decrease alpha-beta pruning at this specific positions - you can get solution faster. But this hurts overall play strength and overall analysis strength. There still are 999 999 999 other positions where tweaking alpha-beta only hurts strength.
Old Houdini (<=4) had option "Tactical" which decrease alpha-beta pruning, engine get lower depth for given amount of time, but if position is known to be "task solving" - it can help. This option widen search windows at any Depth.
multiPV>1 widen search only at root move.
Special version of Stockfish to solve endgames is called Matefinder. It is similar to old Houdini it tactical mode.
Tasks - are specially selected positions (1 in a billion) where traditional human or alpha-beta approach "stuck", that's why they are considered interesting.
If you decrease alpha-beta pruning at this specific positions - you can get solution faster. But this hurts overall play strength and overall analysis strength. There still are 999 999 999 other positions where tweaking alpha-beta only hurts strength.
Old Houdini (<=4) had option "Tactical" which decrease alpha-beta pruning, engine get lower depth for given amount of time, but if position is known to be "task solving" - it can help. This option widen search windows at any Depth.
multiPV>1 widen search only at root move.
Special version of Stockfish to solve endgames is called Matefinder. It is similar to old Houdini it tactical mode.