Author: Albert Silver
Date: 06:54:44 03/06/04
Go up one level in this thread
>>Yes, but on this matter I do not agree with you. If a program has no learning
>>feature and will play a lost game more than once it has to be tested as it is
>>and the double game included in the list.
>>Why?
>>Because the owner will be facing the same situation in a match against it and
>>since this effects the streght of the program it must be included to get more
>>realistic data.
>
>There is a second possibility
>Not to include programs that behave like that in the list.
>
>Suppose that program A has no learning and can lose the same game again and
>again.
>
>Program B played against A 40 games when program C played against A only 15
>games
>Is it fair that B will get better rating than C?
>
>Uri
In a nutshell: absolutely. The first programs had no book learning of any kind.
I remember the energetic discussions in RGCC that this developed when it was
introduced. The result as I recall more or less, was that the only proper way to
deal with this was for everyone to incorporate it. It made sense for more than
one reaason, and the SSDF was not the only one. Users themselves would not want
to be faced with the decision of having to change THEIR plays to avoid repeating
a win because the program is too dumb to avoid it.
The first time I saw book learning wasn't even in a computer program, but on a
Fidelity Mach III with a new book learning feature set in a wooden box. Some
Excalibur (I'm probably getting this last name wrong) or other as I recall. It
had arrived in the Librairie Games, and I spent an afternoon testing this
feature for Pierre Nolot. Little by little I got it to avoid an entire bad line
in the Italian as it tried the possible moves and kept losing. I was very
impressed.
Albert
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.