Author: Uri Blass
Date: 10:54:21 12/20/03
Go up one level in this thread
On December 20, 2003 at 12:57:02, Thomas Mayer wrote: >Hi Uri, > >> I think that the main reason is simply the fact that engines with hash tables >> are usually stronger not only because of hash tables and have more knowledge >> about endgames. > >well, I have seen it several times, that as long the programmers did not >implement hashtables I have some chances (as I said, when I survive in the >middle game) but as soon as they implement hashing in whatever way my chances >were zero. > >> movei cannot play without hashtables but if I release a version without hash >> you will probably not have chances against it also in the endgame. > >There you are most likely right. But of course I have also no chance at all to >survive in the middle game - I am simply to weak... :) > >> I believe that hash does not help a lot for movei espacially when today I do >> not use it for pruning and I will first learn crafty's code about hash. > >maybe you start with some code that is easier to understand like e.g. MSCP or >TSCP (I think the newest TSCPs have hashtables implemented ?!) I understood most of tscp before I started with movei. I know that new tscp has hash keys to detect repetition but not hash tables. I did not look at mscp but I do not believe that I can learn much from it. >It is sometimes better to start with the basics and then try to implement your >own ideas... I learned about hash from bruce moreland's page but I found by testing that it is better for me only to store position with score>beta in the hash tables. > >> I believe based on some test positions with one entry for hash tables that >> today being 2 times slower is a bigger demage for movei than not having hash >> (I do not believe that one entry changes much). > >There you are for sure right ! Well, when I mean having implemented hash then I >had in mind that one would use at least 16k or more entries... > >>I expect it to be changed when I implement hash correctly. >Well, I had some discussions with Dieter Bürßner about that topic - at least you >could spent some thought on the idea that your implementation is more correct >then what we do, especially in combination with extensions and forward >pruning... > >Greets, Thomas I know that my implementation of hash today is very simple and I simply did what worked in my tests. Extension is also one of the thing that I need to change in the structure of movei and today the remaining depth practically does not give me the partial extensions(I have a different varaible for it and I think that having all in one varaible is better). There are things that I do not do the same as Crafty but I still think that I should learn from crafty (I plan to rewrite my alphabeta and I think before rewriting it to understand the structure of Crafty in order to have a better design and my alphabeta is connected with the hash tables). The structure of Crafty already helped me when I got almost 2% speed improvement by adding a file iterate.c that does the same job as crafty when previously it was in the same file as alphabeta. 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.