Author: Sune Fischer
Date: 06:38:10 07/24/03
Go up one level in this thread
On July 24, 2003 at 07:54:22, Steve Maughan wrote: >Russell, > > >>Could you (or anyone else) go into a little bit more detail about what you see >>as the drawbacks of UCI? > >Sure! > >The main drawbacks are those hinted at by Dieter. > >Most engines ponder by completing a search, assuming an opponents move, and if >it's actually played by their opponent they continue to seach - thus giving them >extra seach time on the position. If this is how you do pondering then there is >no problem. However, if you do something more complex e.g. after starting to >search the pondered move you find at mates sequence so switch to pondering >something else - then this is not easily implemented using UCI. > >The second common drawback stated is to do with learning. The way UCI is setup >it's not easy to tell when a game is finished - effectively the GUI decides. >This can make learning tough. However engines such as Sjeng don't seem to have >a problem. > >Another less common complaint is to do with time management. The GUI doesn't >give information about the *next* time management phase of the game. So if you >had 40 move in 2 hours followed by rest of game in 1 min (extreme controls!) >you'd want to same some of your initial two hours for the second stage - under >UCI you don't know about the second phase until it starts. > >Those are the main drawbacks but IMHO they are not that severe and the ease at >which UCI can be implemented makes it ideal for an amateur engine. Remember the >program that tops the SSDF (aka Shredder) has been using UCI for four years. I >take the approach that when Monarch (my engine) is getting more than 50% against >the latest version of Shredder I'll start looking for weaknesses in UCI to bitch >about!! > >Regards, > >Steve I agree whith those saying it is easier to implement, but the reason for that is that the engine is running in a dumber state. As you well know, at each move the position is setup'ed again. Normally this will clear hash-, killer- an historytables. This is good if you want 100% reproducable games, it is bad if you want to play bullet or reuse information from move to move, ie. some of the efficiency is lost. As mentioned learning is made harder and the engine can't resign or offer draws. So in UCI the engine is not a fully legit chessplayer, it is more like an analysis tool used by the GUI to play games. Naturally this makes it "easier" from the programmers point of view, as the GUI does some of the work for you, but don't confuse "easier" with "better". I personally see the winboard protocol as being better, although none of them are perfect of course. -S.
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.