Author: Christophe Theron
Date: 09:23:58 10/21/99
Go up one level in this thread
On October 21, 1999 at 11:20:10, blass uri wrote:
>On October 21, 1999 at 06:29:12, Dave Gomboc wrote:
>
>>On October 20, 1999 at 09:41:21, Christophe Theron wrote:
>>
>>>BTW: Tiger has NO agressive learning. It does not try to repeat won games. It
>>>just try to avoid repeating losses. This game appears twice only by bad luck...
>>>
>>>
>>> Christophe
>>
>>In auto-232 play, isn't this a mistake?
>>
>>Dave
>
>It depends on the target.
>If the target is to improve the engine in the future then it is not a mistake.
>
>You cannot learn much from seeing the same game agains and again and you can
>learn more when you get more information.
>
>I also think that part of the customers do not want to see programs win the same
>game again and again in comp-comp games so if the target is to sell the program
>it is not clear if it is a mistake.
>
>Uri
I assume that people interested in computer chess not only look at the results,
but also look at how they were achieved.
I assume people understand that a program that achieves the best place by
replaying the same won games does not deserve the first place.
When a customer buys an engine, he is interested in ENGINE STRENGTH. You can say
that every part of the engine has to be taken into account, including learning,
but I pretend users are not much interested in learning. I mean "learning" as it
is done currently in chess programs. The kind of parrot learning.
Users:
* either use their programs to play against them, and in this case learning is
not important because the player will not replay a lost game (but the program
has to learn how to avoid losing again the same game).
* either use their programs for analysis, and also in this case learning is
useless.
Christophe
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.