Re: [Bug-gnubg] The race training and benchmark datasets

[Øystein Schønning-Johansen]

Yes! It is really a naïve parallel operation, that's why I tried to do this with OpenMP. (I've used OpenMP with great success in other projects).

I guess your suggestion is actually more pragmatic and doesn't need any more coding. I'll try that. Thanks.

-Øystein

On Mon, Jun 17, 2019 at 9:33 AM Joseph Heled <address@hidden> wrote:

You have many independent runs, right? Why worry about multi-threading? Divide the set into (say) 16 threads (or whatever makes sense for your CPU) and run each set on another thread.

-Joseph

On Mon, 17 Jun 2019 at 19:28, Øystein Schønning-Johansen <address@hidden> wrote:

Thanks for your input. I'm still thinking about this.

I really cannot decide which rollout code to use. :-)

- sagnubg is really good but I have to tweak it a bit to make it work. I'm also not sure it is multi-threading.

- GNU Backgammon, for sure it is good, but I probably have to build some scripts to run this through. Indeed multi-threading seems to work.

- My own little rollout code. It really works good, but when I add multi-threading with OpenMP, it really does not speed up at all. Getting this working is actually what I'm spending time on now. I'm really scratching my head.

As you mention, the tool I posted some weeks ago can indeed be used for some of the positions. That was actually the main reason I created the tool in the first place. I think you are right. I have to start with the positions with a lot of checkers born off. I can probably sort out in a simple way. I should also only handle positions where gammon and backgammon are not a subject. I can probably handle those separately.

-Øystein

On Sun, Jun 16, 2019 at 10:33 PM Philippe Michel <address@hidden> wrote:

On Mon, Jun 10, 2019 at 02:49:26PM +0200, Øystein Schønning-Johansen wrote:

> I will try re-rolling out these positions. Do you have any experience of
> how to do good rollouts of race positions? Good rollout settings for race
> positions?

When I re-rolled out the benchmarks I mostly used the settings that had
been used previously. I think I changed the rolloutLimit parameter
somehow (the number of alternatives included for checker plays). It is
currently set to 10 for doublets and 5 for other rolls. I don't remember
exactly what I did ; maybe it used to be 5 in all cases.

I previously wrote it would be useful to have variance reduction in
sagnubg, but this is not very important since it does 0-plys rollouts
(VR works for them, but it is slow and simply doing more trials is about
as good in terms of SD vs. time used). Doing 7776 trials instead of 1296
doesn't seem unrealistic.

>From the other parameters :

s version 1.93 weights 1.00 moves2plyLimit 20 rolloutLimit 5 nRollOutGames 1296 cubeAway 7 include0Ply 1 evalPlies 2 shortCuts 1 osrGames 1296 osrInRoll 1

experimenting with osrInRoll set to 0 may be interesting. I dont know if OSR
is used for speed or for accuracy...

Another interesting thing to try, if it is practical, would be to use the
software you mentionned a few weeks ago to calculate exact values.

For instance, sort the positions by leading player's pipcount. Start
from the smallest ones with your software ; that should tackle the
hypergammon-like positions with few checkers and ideally the very
unbalanced ones where the trailer can only try to save the gammon. The
latter may well be misplayed in the current rollouts.

_______________________________________________
Bug-gnubg mailing list
address@hidden
https://lists.gnu.org/mailman/listinfo/bug-gnubg