[Bug-gnubg] Re: Rollout settings?? What went wrong?

[Øystein O Johansen]

I'm posting this to the mailing list just to archive it. It's originally
posted by Jim to the GammOnLine forum, but GammOnLine is not archived so I
post it here for the archiving in the mainling list. This message can be
nice for the documentation.

-Øystein

--- 8< --- The message ---8< ---

If I had to guess, I'd guess that you'd done at least a short rollout of
this position, then decided to choose different settings to get a more
accurate rollout. When this happens, gnubg will try to extend the original
rollout and will use the settings from the original rollout, not the
current settings. If you want to restart a rollout using different
settings, use the 0 or 1 play button first to get an ordinary evaluation,
then starr ot the rollout. Just to be clear, if you want to change any of
the following, you must ensure that the current evaluation is *NOT* a
rollout:


Truncated/not truncated


Truncation number of moves


Stepped truncation (changing evaluation parameters after some number of
moves)


Cubeful/Cubeless


Quasi-random dice


Truncation at bearoff databases


RNG seed


Any form of changes of the evaluation settings for either player.


What isn't a problem when extending a rollout and doesn't require clearing
the old results:


Changing the number of trials Changeing the criterion for stopping the
rollout (e.g. if you previously said to stop if the j.s.d. of the best play
was 3 or more better than any other, you could change that to 4, or turn it
off as a stopping parameter and gnubg would "do the right thing"


The screeenshots and rollout results don't match - grandmaster setting
disables pruning on 1 ply, and does do 3 ply evaluations, the rollout
results have 1 ply pruning on andareonly 2 ply evaluations. The most likely
is that you had the settings listed in the rollout results, started a
rollout, decided to use a more strict setting, changed to grandmaster and
restarted the rollout, expecting gnubg to use the new settings.


--- 8< ---


Some discussion: Should the rollout compare the current rollout settings to
the previous rollout settings before extending an rollout?





-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.