Re: computer-go: A little Arithmetic

["Ran Xiao" <ranxiao@xxxxxxxxxxxx>]

I was estimating that 1 million games require 35MB storage for learnt
results on one computer. 1000 million games would require 35GB, a $120 hard
disk driver deal. So much as human being ever played.

Huge raw data is for a learning algorithm to find relations and rules, not
for a program to check exact match. There are many more possible playing
positions left to 9K programs.

Weimin

----- Original Message -----
From: "Andre Engels" <engels@xxxxxxxxxxxxxxxxx>
To: <computer-go@xxxxxxxxxxxxxxxxx>
Sent: Thursday, November 16, 2000 5:45 AM
Subject: Re: computer-go: A little Arithmetic


...

My rough-and-ready estimate would be around 1,000,000,000. But what does it
matter? The number of games that could _possibly_ be played is much and much
larger. If two players go and play a game, independent of who they are and
how strong they are, the chance that the game they play will be one that
has already been played will be extremely small.

> If board configurations can be generalized, there should exit a learning
> system that can generalize these configurations. The real odd balls can be
> dealt by 9K programs. If board configuration can not be generalized, no
> learning scheme will work.

All I intended to say is that the method of working that you suggested
(build an extremely large database of positions, and what to play in those,
and use a simple, weak program for the remaining possibilities) will not
be going to work.

...
--
Andre Engels, engels@xxxxxxxxxxxxxxxxx
telephone: +31-40-2474628 (work) +31-6-17774490 (mobile)
http://www.win.tue.nl/~engels/index_en.html