[Computer-go] Project Leela Zero
sheppardco at aol.com
Fri Dec 29 12:02:22 PST 2017
Seems like extraordinarily fast progress. Great to hear that.
From: Computer-go [mailto:computer-go-bounces at computer-go.org] On Behalf Of "Ingo Althöfer"
Sent: Friday, December 29, 2017 12:30 PM
To: computer-go at computer-go.org
Subject: [Computer-go] Project Leela Zero
Hello in the round,
I am not sure how narrowly people from the list are following the progress of Gian Carlo Pascutto's project Zero Leela. Therefore, here are some impressions.
The project site is:
Shortly before Christmas some guys in the German Go mailing list claimed that LeelaZero had not yet reached a 25-kyu rank. To get an own impression I (with an inactive rating of 18-kyu) went on KGS and played a free game against LeelaZSlow. This version takes exactly 28 seconds per move, even in trivial situations. Long paragraph, short outcome: I lost clearly. You can download the sgf from here:
In the meantime the KGS versions of Leela have made considerable progress. For instance, yesterday and today two 1-dans and a 3-dan were beaten in single games.
However, Leela also has "strange" weaknesses. The most serious one concerns hopeless ladders. The only way out for Leela seems to be to play early tengen-like moves (as possible ladder breakers).
At least three versions of LeelaZero are active:
LeelaZFast, LeelaZSlow, and LeelZeroT.
As soon as a new "best" LeelaZero version has emerged in selfplay runs (of length 400 games) it substitutes the previous version for play on KGS. Currently this happens 1 or 2 times per day.
Computer-go mailing list
Computer-go at computer-go.org
More information about the Computer-go