On 2017-08-24 02:59, Philippe Michel wrote:
On Wed, 23 Aug 2017, tchow wrote:
Something that might have a better chance of yielding superior results
is to use 1-ply or even 2-ply for the variance reduction computation
instead of 0-ply, at least for the first or second steps of a rollout
trial.
This is already the case. Variance reduction for n-ply rollouts (n >=
1) is done at (n-1)-ply.
Obviously this would slow things down
This is not an issue. The costs of one more ply and variance reduction
in number of evaluations are similar (x 21). The worst case would then
be a slowdown of 100% instead of 5% or less, but the same evaluations
are used for both features so the slowdown is very small in practice,
as long as the variance reduction ply is less than that of the main
evaluation.
O.K., so let me rephrase my suggestion: What happens if we use n-ply
variance reduction for an n-ply rollout, at least for the first or
second steps of a rollout trial?
Tim
_______________________________________________
Bug-gnubg mailing list
[email protected]
https://lists.gnu.org/mailman/listinfo/bug-gnubg
