Re: [Bug-gnubg] bots roll out reliability

[Øystein O Johansen]

Hi,

This is a position type where I won't trust a rollout of any bot. As a
human player I would say JellyFish is the bot on the right track. The plan
must be to force O to open his prime. Slotting the acepoint does the trick.
It buys some timing if hit and it doesn't reduce the number of root numbers
(It's just 2-2 anyway). With this play X may be able to force O to break
the prime in the future. Trying to close the home board, loses right away.
I don't think these kind of positions shows up to often in TD-training
session, and I would suspect that the bots can misplay this in a rollout
and thereby get a very wrong result. I've seen rollouts of different bots
differ more than 0.2 in equity. I think such differences must come from
misplaying.

> what are GNUbg's equivalent settings for jf level 6 roll out ?

Level 6 rollouts uses variance reduction and 1-ply lookahead. Not sure if
it is cubeful or cubeless, or according to score or not according to score.

> what about GNUbg rolling this position out ?

I won't trust that rollout either.

> Is there any interest ,for Gnubg team developers ,
> to report here such positions (main bots disagree by a lot )?

Yes, Yes and Yes!!

-Øystein



-------------------------------------------------------------------
The information contained in this message may be CONFIDENTIAL and is
intended for the addressee only. Any unauthorised use, dissemination of the
information or copying of this message is prohibited. If you are not the
addressee, please notify the sender immediately by return e-mail and delete
this message.
Thank you.