On 2/19/07, Winter <[EMAIL PROTECTED]> wrote:
> At what point during the simulation of a game should I switch from 2
> ply to many ply?
>
> I notice that if I do it too early, the results with many ply often
> seem counterintuitive or nonsensical (to me).

If you use many plies early, the results will eventually converge, but
it takes a long time. Take the saying "measure twice, cut once." Many
plies at the beginning of the game is like "measure 80,000 times, each
time adding a random length between -200 and 200 inches, average the
results, cut once." It works, but it's not the best way. I kinda wish
we'd explained this better or put a warning message in there.

Use many plies when the game is far enough along that a 2 ply sim's
iterations would mix pre-endgame and endgame or endgame and game-over
situations. If there's a chance that the last move in an iteration
will empty the bag, use many plies. Then the game always ends for
every iteration, so it's apples to apples all around.

John O'Laughlin

Reply via email to