Author: Odd Gunnar Malin
Date: 18:05:58 04/15/04
Go up one level in this thread
On April 15, 2004 at 20:07:02, John Merlino wrote: >On April 15, 2004 at 19:04:01, Odd Gunnar Malin wrote: > >>On April 15, 2004 at 15:31:22, Stefan Meyer-Kahlen wrote: >> >>> >>>Below you find an extension of the UCI protocol which was discussed with many >>>engine authors at the IPCCC earlier this year in Paderborn and finally worked >>>out with Rudolf Huber of SOS, Dieter Bürßner of YACE and myself. This new >>>version is fully compatible with the old protocol (just one tiny exception, see >>>below), so you can load and old engine in a new user interface and a new engine >>>in an old user interface. This is the reason why we did not call it UCI2 or >>>something. >>> >>>The main changes are: >>> >>>* a possibility to set the engine to any playing strength measured in Elo, >> >>I like this option very much, hope it encourage engine authors to create engines >>that support it. > >It's a good idea, but I don't expect too many authors to put a lot of time into >trying to make it remotely accurate. Think how long it would take Dieter (for >example) to be able to program his engine to play something reasonably >approximating 1000 ELO....then 1200 ELO...then 1400...or whatever different >settings this new protocol is going to allow. A difficult task. > It isn't necessary to have exact numbers, but an estimate could be made by letting the engine play on chess-servers. If the author want his engine to be used by many players then he will hit most people at 1200-1700. If the average player are 1400-1500 in rating, creating an engine that is stronger than 1700 (engine score 75%) would cut off 50% of possible users. >I suspect that most authors are simply going to artificially limit the search >depth and/or NPS and say that they now "support" this feature. > There is a method that is only slightly harder for reducing the tactical strength. In the root of the search (where a change of prefered move happend), each change of move could be timestamped and go into a table for later lookup when a move should be made. My current formula for selecting move from this table are: strengthdivider=2^((100-strength)/5) where strength=0-100% then I take the move from the table where movetime=usedtime/strengthdivider. Knoweledge should be easy too to give with a number, but here of course you need to study games to get a feel what to add at each level (a bit randomness is needed too). The third parametre for strength reduce are blunder move/rate. Here is where the hard work begins. >The King is the only engine I know of that does this well, and I would guess >that Johan spent a great deal of time making sure that his engine could be >"dumbed down" in a believable way. This is likely not what the typical (unpaid) >engine programmer is going to want to spend his time on. > >jm A look at the average rating (above), no wonder why CM are the most populare chessprogram. Odd Gunnar
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.