Author: Bruce Moreland
Date: 09:27:28 06/10/01
Go up one level in this thread
On June 10, 2001 at 06:47:31, Mogens Larsen wrote: >On June 10, 2001 at 00:02:19, Bruce Moreland wrote: > >>1) It is using 0x88 move generation. The function is something under 200 lines >>long, which seems big, but there are a lot of comments and there's some repeated >>code. >> >>2) It is using vanilla windowed alpha-beta with no PVS or anything like that, >>and I plan to keep it this way, since that is clear and performs acceptably. >> >>3) It has a check extension. I think that is enough for now. I may try to do >>some pruning in the quiescent search, which uses MVV/LVA. I don't play to write >>a static exchange evaluator. >> >>4) This is already a more or less fully functional winboard engine, an effort >>that consumed a whole bunch of time, not that Mann's standard is hard to write >>to. I just had to spend a lot of time messing with threads. >> >>5) I will make it available for source and executable download after it is more >>or less finished, which should be pretty soon. I'm trying to write to a >>reasonable standard. As you can see by the fact that it has played games, it >>already can do that, but I would like to finish commenting it, clean up the data >>structures, add some error trapping, etc., as well as some obvious features. >> >>Things to add: >> >>1) Opening book. I am getting tired of 1. Nc3. >>2) An eval function that does something other than piece/square. >>3) Pawn structure eval. It currently has absolutely none. >>4) King safety eval. Likewise. >> >>I may leave one or more of the above out. >> >>Current features: >> >>1) It's doing about 250K nps on a single processor of my quad 450 Xeon, and it >>gets 6 or 7 plies in the middlegame of 5 0 blitz games, which is decent. >> >>2) It has a transposition hashing system and can solve Fine #70 in something >>under two seconds. >> >>3) It can do repetition detection, it claims 3x repetition draws, and it detects >>and claims 50-move draws (this latter hasn't happened yet, so I'm speaking >>hypothetically). >> >>4) It's scored alright against a bunch of other non-professional chess programs. >> It won one game against Chess System Tal on similar hardware (this seems hard >>to believe and possibly the account "rb" has some problems), and it whacked >>Vincent once in several tries. It hasn't played any titled players yet but I'll >>try to fix that. >> >>Right now I'm working on a test suite eater so it can suck up EPD's so I can see >>how it does on suites. >> >>It's hard to find time to work on it, but this is fun so I'll work on it a lot, >>and I hope I can have it done soon. > >The other educational attempt called Beowulf relies on tablebases for most of >its basic endgame knowledge. Will that also be your approach? My basic approach will be to call the eval function, which thanks to a change made two days ago will probably recognize that the game is an endgame, and reward the program if its king is in the center. That's certainly good enough to win KQ vs K, and is probably good enough to win KR vs K. What more do you need than that? bruce > >Mogens.
This page took 0.01 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.