Author: Bas Hamstra
Date: 02:34:06 01/04/01
Go up one level in this thread
Some extra info: the program is less than a year old, and so there is still much to improve. In particular it has not much endgame knowledge, apart from the obvious things, when for example compared to Crafty. And it has a very lousy "book". It is my second program. Version 4.4 differs from the dutch open version mainly in that the hash size is settable and that it extends less (not sure the latter is an improvement). Right now I am working on ways to automatically tune evaluation parameters. The idea is to invest time once, and get rid of the time consuming watching of games and adjusting parameters via trial and error. So far the results with Temporal Difference learning and genetic tuning are far from impressive, but still ongoing. If you want to know more, let me know. Ciao! Bas. On January 03, 2001 at 18:19:29, Peter Berger wrote: >While currently checking on new Winboard engines I also checked this one at last >. > >There hasn't been much public upset about it so far . > >I noticed this one great tournament result but never heard much more about it , >so I thought the WinBoard masters checked on it and weren't too impressed after >all. > >I played two games at slowish time control ( game 1 hour , 10 secs increment per >move ) against Chesstiger 13.0 with a little hardware advantage for the outdog >and watched lines and action live to get a first impression , a very usual >test/event for me for new winboard engines where usually they get an >entertaining spanking . > >First game was a pretty win for Tao with black in a technical endgame with Black >where it completely _outsearched_ (!) the Tiger . > >Second game it started with White with what looked like a complete beginner's >opening 1. g3 e5 2. Bh3 only to get a completely blocked position where it >showed great skills in sophisticated knight manouvres . While Tiger evaluated >more than 1.5 for itself for some time after Tao looked cramped after grabbing >what looked like a pawn better not to be taken it coolly freed itself until it >reached a very promising, advantegous position only it was too late for me to >play it out ( kind of drawish in the end probably ) . > >Yet I was deeply impressed by both games and this was by far the best newcomer >performance I have ever seen by a WinBoard engine . > >This might have been kind of lucky for Tao , yet it really is an experiment I do >_very_ often and I have never seen a performance like this one from a newish >Winboard engine . > >I understand that maybe with its habit of pondering all the time it might not >have got the attention it deserves as being not suitable for single computer >matches . > >So : before I dream and sleep bad and with the excitement of a great first >impression ;-) : any tests and results tried with Tao at longer or Blitz time >controls by others ? > >I am definitely _very_ interested to find more about it . > >pete
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.