Principal variation search (or alpha-beta) wants the probability that a side line will become the main line, after the principal variation is rejected. Certainly evaluation difference from the principal variation is one heuristic to determine probabilty; are there other useful features? Discover them by machine learning.
The goal is to use parallel processors to speculatively evaluate these side lines. Schedule thread priority according to probability (but by what mapping?)
Chess
No comments :
Post a Comment