computer-go: A little Arithmetic

["Ran Xiao" <ranxiao@xxxxxxxxxxxx>]

Let's say we have 362 computers (361 computers evaluate positions and one
does everything else), and 1000,000 good games (not 100,000). Assume every
good game has about 250 moves, and no duplicated board configuration and
move (even though we know many of them will be duplicated configurations.)

On average, each of the 361 computers will have 1000,000 x 250/361 = 692,520
cases stored and prepared to learn rules. At most, 500 bytes will be needed
to store one case, thus 346,260 KB or 350 MB storage will be needed.

If rules can be learnt at 10% raw data storage size (who knows, if the game
can not be generalized, all raw data by itself will forms a rule, and 350 MB
will be needed.), 35 MB storage is good for each good position. That size is
small from my standard (think a $120 35GB HD.)

Now, if a board configuration can not be evaluated by stored rules, most of
the time, the board configuration may be a low level player's work, may be
7K. So, the residual evaluation can be passed to the last computer we have,
with HT, WL, or MFG installed, to do today's 7K computer playing.

Weimin

----- Original Message -----
From: "David Fotland" <fotland@xxxxxxxxxxxxxxxxx>
To: <computer-go@xxxxxxxxxxxxxxxxx>
Sent: Sunday, November 12, 2000 7:53 PM
Subject: Re: computer-go: Data Mining and Rulebase for GO



Please compare the number of positions in the number of good game
records you can get (perhaps 100,000), to 3^360 and describe
what fraction of the game space can be learned by your system....

Please guys, do a little arithmetic before you propose these
schemes.

David Fotland