Author: Uri Blass
Date: 11:55:55 08/15/04
Go up one level in this thread
On August 15, 2004 at 14:13:35, Stuart Cracraft wrote: >On August 15, 2004 at 12:39:56, Uri Blass wrote: > >>On August 15, 2004 at 12:31:13, Stuart Cracraft wrote: >> >>>On August 15, 2004 at 11:58:28, Uri Blass wrote: >>> >>>>On August 15, 2004 at 11:31:11, Stuart Cracraft wrote: >>>> >>>>>On August 15, 2004 at 10:42:46, Dan Honeycutt wrote: >>>>> >>>>>>On August 15, 2004 at 10:25:43, Stuart Cracraft wrote: >>>>>> >>>>>>>So what makes a program more aggressive? >>>>>>> >>>>>>>Better king safety? >>>>>>> >>>>>>>Points for control or occupation of square near the enemy king? >>>>>>> >>>>>>>I've tried obvious things and never been satisfied with the >>>>>>>aggression-level. >>>>>> >>>>>>Asymmetrical king safety. Program needs to not mind moving a piece which >>>>>>shelters it's own king to a position that menaces the enemy king. >>>>>> >>>>>>Dan H. >>>>> >>>>>My current king safety is very limited but doesn't involve any >>>>>hardwiring of friendly pieces to my king. They are free to roam. >>>>>At the same time, I've noticed no aggressive tendancy. It plays >>>>>passively and reaction-only to what the opponent with few exceptions. >>>>>On the other hand, it really fights for the center, develops quickly >>>>>and castles on a timely basis. But once the middlegame hits, nothing >>>>>much happens except wood-shuffling. >>>>> >>>>>I do have a tropism factor to get queens, rooks, and knights to >>>>>minimize the distance to the enemy king. Perhaps something is wrong >>>>>with them. I don't use attack tables in evaluation since my program >>>>>has none. I'll have to revamp the whole program some day to add >>>>>them incrementally but haven't found a good paradigm yet Even bitboard. >>>>>I liked the thing that Atkin/Slate did with incremental updates >>>>>in their makemove/unmakemove. >>>>> >>>>>So basically "middlegame" malaise is my program's problem. I need >>>>>to tighten the tropism to just the few squares around the enemy >>>>>king and heighten the bonus. There is already a substantial bonus >>>>>for loosening pawns protecting the king but I need to get some >>>>>heavy firepower over there, sans attack tables, using tropism >>>>>to get something real happening. >>>>> >>>>>I wish there were some test suites that gauged early midle-game >>>>>aggression. Not checkmate/mate type things but simply threats against >>>>>the castled king. >>>>> >>>>>Stuart >>>> >>>>I think that based on your WAC results search and not evaluation is the main >>>>problem of your program. >>>> >>>>You should get easily more than 290/300 in a few seconds if your search work >>>>correctly. >>>> >>>>Uri >>> >>>Uri, >>> >>>I wish. It's a Pentium 3 1ghz. Not very fast by today's standards. >>>Also, I can only run with 500,000 hash table entries. >>> >>>If my search is broken, it will be years before I find the bugs >>>through trial-and-error. >>> >>>Also, my results just went up after improving the evaluation only. >>> >>>Stuart >> >>better evaluation can also help but if you compare your results with the results >>of programs like gerbil that has almost no evaluation you can find that even >>Gerbil is clearly better than your program on the same hardware with the same >>size of hash tables. >> >>Uri > >I tend to design my programs without too much direct reference to >other source code en masse. There are exceptions such as the SEE code >recently donated to me and an individual idea or two from the board >or from looking at a program. > >However, I try not to review other's entire programs with a view to >duplicating their effort. It is unrewarding. I realize I am not the >most creative of person's but that is taking it just a bit too far. > >That said, when I have looked at other program's in the past, I find >them practically unreadable so it wouldn't do that much good anyhow. >I have an aversion to other programmer's code in large measure. > >I know nothing about Gerbil except through heresay and certainly nothing >about its results on WAC at 1 second per move on a 1ghz P3. So nothing >remains proven. > >Stuart <y point is that Gerbil is not the best program but as far as I remember it did in 1 second per move better result than the result that your program did in 30 seconds per move so I guess that there is a lot to improve in your program only in the search. Uri
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.