[Computer-go] How well does pure monte carlo play?

[Don Dailey]

If he means pure MC the strength is severely bounded without a tree search.
  It would not matter if you did a billion playouts,  it would not play
much stronger than if you did 1 million.    One you have "enough" playouts
there is a severe point of diminishing returns.      You can drop
all-moves-as-first to get a minor improvement once you are doing a lot, but
that will add almost nothing.     There is all sorts of systematic errors
that playouts (pure MC)  will never correct without a tree search.    The
most obvious errors are caused by moves which are very good against "random
play" but in reality are not good moves,  more playouts doesn't fix that.


Don




On Sat, Jan 21, 2012 at 3:45 PM, Michael Williams <
michaelwilliams75 at gmail.com> wrote:

> Do you really mean pure MC or MCTS?
>
> On Sat, Jan 21, 2012 at 2:31 PM, Eric Baum <ebaum at fastmail.fm> wrote:
>
>> If you don't do rave, or use patterns to predict likely moves, or any of
>> that, but just do pure monte carlo,
>> parallelized on a cluster,how well would that play Go? Anyone know?
>> ______________________________**_________________
>> Computer-go mailing list
>> Computer-go at dvandva.org
>> http://dvandva.org/cgi-bin/**mailman/listinfo/computer-go<http://dvandva.org/cgi-bin/mailman/listinfo/computer-go>
>>
>
>
> _______________________________________________
> Computer-go mailing list
> Computer-go at dvandva.org
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://computer-go.org/pipermail/computer-go/attachments/20120121/987cea0b/attachment.html>