Re: [computer-go] Pattern database

["Frank de Groot" <frank@xxxxxxxxxxxxx>]

----- Original Message ----- 
From: "Darren Cook" <darren@xxxxxxxxxxxxxxxxx>
To: "computer-go" <computer-go@xxxxxxxxxxxxxxxxx>
Sent: Wednesday, September 29, 2004 1:35 AM
Subject: Re: [computer-go] Pattern database

Hi Darren,
It seems you are attributing my answers to the Original Poster.
I will just comment on your comments.


> I'd like to echo this point. I didn't have enough, so I had layered 
> databases:
>   1. Large radius pro games only
>   2. Smaller radius, pro games only
>   3. Smaller radius, all games.
I have a similar approach to reduce learning time.
I do an initial pass over 350,000 games of 1d and up, and many more passes 
on 50,000 games on 1p and up.


> When you add liberties to your patterns you reduce the number that match. 
> So you need to make them smaller to have enough to get you through the 
> game. In my experiments the gain was mostly matched by the loss.
Yes I noticed that since I added liberties & chain sizes, I have many 
occasions that a pro played somewhere and there is no match found at all. I 
have 3 strategies to ameliorate that:

1. Have an enormous number of "harvest" games (around 350,000) and gives 
those pattern priority, meaning allow millions ofthem to be added to the 
database,

2. Do not include exact chain size and lib count, but include ranges only.

1 & 2 are in place but unsatisfactorily results, although I must say that 
the results WITH libs & stones is better than without.

3. is very easy to implement (I think less than an hour) and it involves 
APPROXIMATE matching. Not the patterns themselves but the # of stones and # 
of lib ranges will be slightly modified and a re-match is attempted. I think 
this will not delay matching significantly (a few % perhaps). 
_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/