Re: Using partial plys

[Vincent Diepeveen <diep@xxxxxxxxx>]

At 09:43 PM 12/22/98, Darren Cook wrote:
>A search question. In minimax or alpha/beta you search to a fixed ply, so
>the leaf values that are passed up the tree are comparable because the same
>number of moves will have been played.
>
>But if you are using partial ply's or quiescence search then it is possible
>that top-level move A went to a node where black got last move, and move B
>went to a node where white got last move. Is it fair to compare them?
>
>When using these algorithms do you normally add an extra move if black got
>the last move, so all leaf nodes ended with white getting the last move (or
>vice versa)? Or does the fact that you searched deeply to a quiet position
>mean that it doesn't really matter who got last move?
>
>A go example: if you play out two joseki and compare the end results, is it
>fair to compare them if black got an extra move in one of them? Or could
>you just add a bonus to the evaluation function for the side that has
>sente. If so, isn't calculating the worth of sente going to be very inexact?
>
>Darren
>
>PS. And how do you think humans handle this? Reading moves out to different
>depths is a very human thing to do.

a human is not busy with 'depths'. A human sees a move and a plan, and
searches till he thinks that is needed to search. This is called bestfirst
search without any depth limitation. However so far no one managed to 
implement that... :)
>
>