Pruning, (as opposed to reductions)
a name for every heuristic that removes completely certain branches of the search tree, assuming they have no bearing to the search result. Alpha-Beta may be considered as backward pruning, because we found a refutation after searching [1]. Forward pruning always involves some risks to overlook something, with influence on the root score. Some pruning techniques rely on a specialized, but reduced search, others rely only on static move properties. Shannon Type B programs only consider some plausible mores, but all other moves are pruned.
Forward pruning techniques are either applied at expected Cut-Nodes or All-Nodes and return either beta as lower bound or alpha as upper bound. The none recursiveProbCut is applied at both types of none PV-nodes. Rather than immediately returning alpha at strong expected All-Nodes after a reduced or quiescence search, pruning techniques near the horizon skip moves, which are very unlikely to exceed alpha, or in case of the quiescence search itself, all (most) quiet moves as well as losing captures.
Peter W. Frey (1983). The Alpha-Beta Algorithm: Incremental Updating, Well-Behaved Evaluation Functions, and Non-Speculative Forward Pruning. Computer Game-Playing (ed. Max Bramer), pp. 285-289. Ellis Horwood Limited Publishers, Chichester.
Yngvi Björnsson, Tony Marsland (1998). Multi-Cut Pruning in Alpha-Beta Search. CG 1998, pp. 15-24. See also Yngvi Björnsson, Tony Marsland (2001). Multi-cut Alpha-Beta Pruning in Game Tree Search. Theoretical Computer Science, Vol. 252, pp. 177-196 for an expanded version.
Qian Liang (2003). The Evolution of Mulan: Some Studies in Game-Tree Pruning and Evaluation Functions in the Game of Amanons. M.Sc. thesis, Electronic Engineering, The University of New Mexico, pdf
a name for every heuristic that removes completely certain branches of the search tree, assuming they have no bearing to the search result. Alpha-Beta may be considered as backward pruning, because we found a refutation after searching [1]. Forward pruning always involves some risks to overlook something, with influence on the root score. Some pruning techniques rely on a specialized, but reduced search, others rely only on static move properties. Shannon Type B programs only consider some plausible mores, but all other moves are pruned.
Table of Contents
Backward pruning
sound pruning, not affecting the search resultForward pruning techniques
Forward pruning techniques are either applied at expected Cut-Nodes or All-Nodes and return either beta as lower bound or alpha as upper bound. The none recursive ProbCut is applied at both types of none PV-nodes. Rather than immediately returning alpha at strong expected All-Nodes after a reduced or quiescence search, pruning techniques near the horizon skip moves, which are very unlikely to exceed alpha, or in case of the quiescence search itself, all (most) quiet moves as well as losing captures.Returning Beta
at expected Cut-Nodes:Returning Alpha
or below at expected All-Nodes or expected Cut-Nodes which turn out to become All-Nodes:Skipping Moves
likely at All-Nodes:Extended Futility Pruning
Deep Futility Pruning
Move Count Based Pruning
See also
Late Move Reductions aka History Pruning [3]
Publications
1960 ...
1970 ...
1980 ...
1990 ...
2000 ...
2010 ...
Forum Posts
1995 ...
Re: Forward pruning by Andrew Williams, CCC, October 16, 1999
2000 ...
2005 ...
2010 ...
2015 ...
External Links
Pruning Search Trees
Pruning Trees
References
What links here?
Up one level