Computer Chess Club Archives


Search

Terms

Messages

Subject: Question how implement Learning in program

Author: Inmann Werner

Date: 04:47:39 03/29/99


Hello

I often heard about "Learning", and it is clear to me, that this can be very
useful. I thought about it a lot, but have big problems about how to do this
the right way.

My thoughts so far.

OK. I store the game (moves, time and evaluation) during a game. When I have a
lot of interesting games I run a "learning" program, which makes something like
a opening book (hashed?).

But what to write into it. When is a position good (play it!!) or bad(don´t
do!).
I have so much confused thoughts about it.

A simple way would be to say, if I loose, all moves get a malus, if I win, all
moves get a bonus, but I think that would not work.

To recalculate the games with deeper search would spend too much time.

You can say, if the evaluation drops down much, the move before was bad. But it
could also be the move 3 or 4 times above...

I really have no clue..

Is there any algorithm, somebody can explain to me?

Werner



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.