[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

Re: computer-go: Learning from existing games

To: computer-go@xxxxxxxxxxxxxxxxx
Subject: Re: computer-go: Learning from existing games
From: Markus Enzenberger <compgo@xxxxxxxxxxxxxxxxxxxxx>
Date: Wed, 15 Jan 2003 20:15:12 -0700
In-reply-to: <200301142204.00693.oleg_inconnu@xxxxxxxxxxxxxxxxx>
References: <Pine.LNX.4.44.0301141957260.31137-100000@xxxxxxxxxxxxxxxxx> <200301141943.09995.compgo@xxxxxxxxxxxxxxxxx> <200301142204.00693.oleg_inconnu@xxxxxxxxxxxxxxxxx>
Reply-to: computer-go@xxxxxxxxxxxxxxxxx
Sender: owner-computer-go@xxxxxxxxxxxxxxxxx
User-agent: KMail/1.4.3

On Tuesday 14 January 2003 20:04, Oleg wrote:

> How does the percentage of wins against GnuGo depend on the board
> size and the amount of time GnuGo is allowed to think?

the number is from games on 9x9 boards.
Percentages close to 50% usually scale well to the 19x19 board,
but the network needs a retraining on increasing board sizes with the 
9x9 weights as a starting point for this.

- Markus

References:
- Re: computer-go: Learning from existing games
  - From: schraudo
- Re: computer-go: Learning from existing games
  - From: Markus Enzenberger
- Re: computer-go: Learning from existing games
  - From: Oleg

Prev by Date: computer-go: Upcoming competitions?
Next by Date: Re: computer-go: Learning from existing games
Previous by thread: Re: computer-go: Learning from existing games
Next by thread: Re: computer-go: Learning from existing games
Index(es):
- Date
- Thread