On 5/5/2024 10:54 AM, Philippe Michel wrote:
On Fri, May 03, 2024 at 04:21:51AM -0600, Murat K wrote:
I just remembered that someone had asked not so long ago
about the possibility of saving the moves during rollouts
and I had inquired about why he was wanting to do that.
Let me start by adding to my own comments first. For having
argued for years that bots' calculations, (including during
rollouts), weren't accurate, I had reasons to want to see a
log of moves during rollouts and I thought that the person
who asked about it may have had the same/similar reasons as
me, which could support my stance on the subject.
I searched the archives but couldn't find it. Maybe I was
using the wrong keywords. Does anyone remember?
After your below answer, I remembered you had given him the
same answer but even searching again with keywords like "set
rollout log on", I still can't find it even though now I did
find older posts about it. I wonder if entire threads do ever
get deleted?
Now that we are talking about this, I think it would be a
great unique feature for GnuBG to allow saving the moves
in trials during rollouts. That way we can scan them for
any berserk sequences of moves and other kinds of similar
occurrences that we may be curious about.
It is possible to save every game played in a rollout, but
the corresponding commands are only available from the CLI.
What another nice surprise to find out that GnuBG already had
it for a long time. Despite some of it bugs and shortcomings,
GnuBG is the best bot that ever was and is for offering such
unique features.
You would use something like:
set rollout logfile /tmp/test
set rollout log on ....
I saved your instructions in a text file. I recently ordered a
I7-12700K based PC which should be about four times as fast as
my current one and I'm looking forward to doing more of all
kinds of experiments.
Th distribution of game length is appended below. The longest
one was 156 moves ..... > The containment play was probably far from perfect
but not
ridiculous. A pretty normal long game.
Thank for the trouble of providing this info. It actually made
me realize that a game doesn't necessarily last many hundreds of
moves in order to contain weird moves; even under 150-200 moves
can have them. So, we can't know for sure without actually looking
at the logs but we may reasonably make that assumption, considering
the impracticability of looking for them in huge log files.
games length
6 5
.....
1 156
A trial of 1,296 games is not much but it's still very interesting
to see that the average number of moves in those games was 42.15
In all my experiments the average was 46-47 moves per game which is
considerably less than the 54 from other sources like the 6 million
games on https://zooescape.com/backgammon-stats.pl that I was using
in my mutant strategy based on my "games stages fartoffski formula".
That 54 most likely includes match games also which are apparently
longer than money games on the average. I already have done better
mutant experiments that I didn't get a chance to put on my web site
yet but now I think I need to do even more using shorter averages
to calculate my game stages intervals. I think that will cause my
mutant to play more conservatively but I'm not sure if it will win
more. I'll let you all know... ;)
MK
