Hi,
My experience is the same: My CNN was a very poor judge of life and
death. Part of the problem is that I couldn't get Pachi to behave
exactly the way I wanted (play to maximize score; play to the bitter
end, assuming everything left after two passes is considered alive). But
perhaps there
I understand that using games from humans to learn about life and death
introduces all sorts of biases. That's why I tried to use games from an
engine instead.
In their standard configuration, MCTS engines will sometimes let lots of
groups die after they know the game is hopeless, or if they have
Well my point is more in the direction to get a good live-and-death
solver, maybe completely independent of the game as a whole. Just train
a DNN on some life-and-death collections with solutions, where all of
the usual tricks are applied. I think this has the potential to get
similar if not
Hi!
On Fri, Jan 15, 2016 at 04:54:18PM -0500, Álvaro Begué wrote:
> In their standard configuration, MCTS engines will sometimes let lots of
> groups die after they know the game is hopeless, or if they have a large
> advantage and they still see a guaranteed victory after the group dies.
>
On Fri, Jan 15, 2016 at 5:14 PM, Petr Baudis wrote:
> Two ideas re Pachi:
>
> (i) In uct/uct.c maximize_score section, try changing the
> max_losing_komi from 30 to, say, 300.
>
> (ii) Try to work out that if Pachi decides it's losing, the gameplay
> is switched over to GNUGo
On 1/15/16 2:07 PM, Michael Sué wrote:
Well my point is more in the direction to get a good live-and-death
solver, maybe completely independent of the game as a whole. Just train
a DNN on some life-and-death collections with solutions, where all of
the usual tricks are applied. I think this has
> Regarding 9x9, I believe Alvaro Begue has explored this idea in a way
> which perhaps would work better in a go engine. He used pachi to generate a
> database of games by playing against itself and then trained a model in a
> similar fashion to what I did. I'm not sure about the results of his
>
Hey everyone,
I recently trained a CNN to do board evaluation in Go. You can see the
work on github:
https://github.com/jmgilmer/GoCNN
The network was trained on 15000 professional games which didn't end in
resignation, I had the network try to predict the final ownership based on
current
Although it cannot replace MC simulations altogether, it *could* be used
for more accurate prior values I suppose. Do you plan to integrate it in
a MCTS program and see? Michi is also written in python...
Gonçalo
On 12/01/2016 21:30, Justin .Gilmer wrote:
Hey everyone,
I recently trained
Quick question: When using this mailing list, how to I explicately reply to
a thread, so far I've just been editing the subject and sending it to
computer-go@computer-go.org.
Regarding use in a MTCS engine, I strongly suspect it would perform poorly
in its current form. It is quite poor at life
Hi Justin,
that sounds interesting. WOPuld it be possible for you
to train an analogous CNN for 9x9 Go?
Ingo.
Gesendet: Dienstag, 12. Januar 2016 um 22:30 Uhr
Von: "Justin .Gilmer" <jmgil...@gmail.com>
An: computer-go@computer-go.org
Betreff: [Computer-go] Board
11 matches
Mail list logo