Re: [Bug-gnubg] Problems with Stop rollout when one move appears to be best

[Christian Anthon]

Are you using the multithreaded code or not. There are some problems
in the non-threaded code?

Christian.

On Thu, May 28, 2009 at 9:05 PM, Neil Robins<address@hidden> wrote:
> I find that when rolling out more than two moves there are various problems
> that sometimes happen if this box is checked:
>
> 1. A move that has come back within the set JSDs only sometimes not always
> restarts rolling.
> 2. A move that has gone way beyond the set JSDs sometimes continues to roll.
> 3. The whole rollout can stop with some rolls neither rolled to maximum
> number of trials nor outside the set JSDs.
>
> I think these problems happen generally when the plays don't fall out of
> contention from the bottom of the list up. Stopping and restarting the
> rollout with them reordered gets the right moves rolling again.
>
> Try this Example:
>
>  GNUBgID: hucZAxDYzmDABg:8AmmAAAAIAAA
> Roll out the top five 2-ply moves, using :
> Full cubeful rollout with var.redn.
>         1296 games, Mersenne Twister dice gen. with seed 359697340 and
> quasi-random dice
>         Stop when best play is enough JSDs ahead: limit 3.1 (min. 217 games)
>         Play: 0-ply cubeful [expert]
>         Cube: 0-ply cubeful [expert]
>
> If you have more time even more problems can be seen rolling out the same
> moves in the same position using 2-ply with a different seed same box
> checked same JSDs (my initial rollout):
>  1296 games, Mersenne Twister dice gen. with seed 859697340 and quasi-random
> dice
>         Play:  2-ply cubeful prune [world class]
>         keep the first 0 0-ply moves and up to 8 more moves within equity
> 0.12
>         Skip pruning for 1-ply moves.
>         Cube: 2-ply cubeful prune [world class]
>
>
> _______________________________________________
> Bug-gnubg mailing list
> address@hidden
> http://lists.gnu.org/mailman/listinfo/bug-gnubg
>
>