Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Movei-Jonny Crazybishop so far report.

Author: Dann Corbit

Date: 21:19:12 07/30/04

Go up one level in this thread


On July 31, 2004 at 00:14:26, Uri Blass wrote:

>On July 30, 2004 at 23:29:19, Dann Corbit wrote:
>
>>On July 30, 2004 at 15:51:24, Uri Blass wrote:
>>
>>>Movei-Jonny CrazyBishop 36.5-13.5  32.5-17.5 at 1  minutes per game
>>>Movei-Jonny CrazyBishop 34.5-15.5  35-15     at 2  minutes per game
>>>Movei-Jonny             33.5-16.5  34-16     at 3  minutes per game
>>>Movei-Jonny             27-23      29-21     at 4  minutes per game
>>>Movei-Jonny Crazybishop 23.5-26.5  27-23     at 10 minutes per game
>>>Movei-Jonny CrazyBishop 23.5-26.5  31.5-18.5 at 20 minutes per game
>>>Movei-Jonny             19-21                at 40 minutes per game
>>>
>>>more details about result of the last match from movei point of view
>>>1-3 positions 1-2
>>>9.5-2.5 positions 3-8
>>>0.5-3.5 positions 9-10
>>>6-2 positions 11-14
>>>2-10 positions 15-20
>>>
>>>total after 40 games
>>>19-21 for Jonny
>>>
>>>If you take only 12 games it can be 10-2 for Jonny or 9.5-2.5 for Movei
>>>dependent of the positions.
>>
>>Proper hash table utilization will do movei wonders.
>>
>>As you move to the longer time controls, movei approaches a 50% (70 Elo) time
>>penalty.  Not to mention the better move ordering you could achieve.  I think it
>>also shows that your evaluation function is quite good.
>
>I agree that movei probably can earn much from hash but
>part of my ideas may contradict efficient use of hash
>Movei is using evaluation that is based on the path and not only on the leaf
>position.
>
>I tested that it was productive at the time that I implemented it when movei
>even did not use hash.
>
>I have also many ideas for improvement without efficient use of hash and I feel
>that only the sky is the limit for improvements.
>
>It will probably take time to implement things and one of the reason is that
>movei is badly written(probably all the programs at movei's level or higher have
>a better code).

I think it is a hard path to start with TSCP for an outline.

It is very simple, but too many shortcuts are taken that will be expensive
later.

Especially the extensive use of globals (which is fine for single threading but
will become a headache if you want threads).



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.