Author: Uri Blass
Date: 07:08:04 02/14/04
Go up one level in this thread
On February 14, 2004 at 09:42:42, Uri Blass wrote: >On February 14, 2004 at 07:51:01, Peter Fendrich wrote: > >>On February 13, 2004 at 10:35:06, Uri Blass wrote: >> >>>On February 13, 2004 at 10:13:26, Peter Fendrich wrote: >>> >>>>On February 13, 2004 at 00:28:13, Paul Doire wrote: >>>> >>>>>Hi All, >>>>> >>>>>I am interested in knowing the strengths of all who post here. >>>>>Whether it is USCF or FIDE.To import chess knowledge into chess programs >>>>>seems to require the programmer to be strong or at minimum, their resources to >>>>>be strong. Who dares to tell...and dares to tell of those who will not tell. >>>>>Some human analysis we see would carry more weight knowing the strength of the >>>>>analyst. Do you dare to tell? >>>>> >>>>>Regards, >>>>>Paul >>>> >>>>I'm quite convinced that the correlation between being a strong chess player and >>>>a strong chess programmer is not very high. It's far more important to be a good >>>>programmer than a good chess player in order to produce a strong chess program. >>>>Of course the programmer must have rather good knowledge about different chess >>>>elements but that is not at all the same as being strong in OTB play. I even >>>>beleive that a very strong OTB player might have some troubles to lower his >>>>level of play to the level of an evaluation function in a chess program... >>>>/Peter >>> >>>I do not agree with the last claim. >>>strong players do not need to lower their level of play. >>> >>>Even 1500 players know about fortress positions when most chess engines do not >>>have the knowledge. >>> >>>This is not a problem for the programmers so for the same reason the fact that >>>some programmer is a strong player should not be a problem. >>> >>> >>>If you try to teach a chess program everything that you know you have problems >>>and the question if you are a strong player or not strong player is not >>>important and the problem is that you simply do not know to give definitions to >>>your knowledge. >>> >>>Uri >> >>I really shouldn't speak for others than myself but I had some troubles with >>that and I think that a very strong OTB player might have the same or even more >>of it. >>Not a big obstacle compared to all others but it's one of them... >>/Peter > >Movei does not know a lot of things that I know but at these days I decided to >try to teach it things that I do not know about KPK endgame(I know that it is >not the best way to improve the program as fast as possible). > >I try to write a function that will get the position and give me the result win >draw without looking at a table. > >I use Dieter's table for debugging my program. > >At this point I have almost 500 lines of code when most of them are for white to >move and still have 3000 cases when I return do not know as an answer. > >I only look at positions when the pawn is white pawn in column a,b,c,d and the >black king is not in check so I still have a lot of undecided cases. > >There is probably better ways to improve the program but I want at least to >solve the problem of one simple endgame not by tables. > >I try to get progress in 2 directions: > >1)inventing more rules for undecided cases(I look at the first undecided >position and try to define a rule based on it but unfortunately there are cases >when the rule is only good for less than 1% of the remaining position). >2)trying to look at my code to see if I can generalize rules to do the code >shorter(I am practically sure that it is possible). > >Uri Note also that one of the nice effects of 2 is sometimes solving part of the unsolved positions because I generalize also cases that I did not consider. Now I have only 2916 undecided positions with white to move 1634 out of them are wins for white. With black to move I have 36604 undecided positions but I decided to work first on white to move and I almost did nothing with black to move. Uri 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.