Difference between revisions of "RankCut"

From Chessprogramming wiki
Jump to: navigation, search
Line 1: Line 1:
'''[[Main Page|Home]] * [[Search]] * [[Selectivity]] * [[Reductions]] * RankCut'''
'''[[Main Page|Home]] * [[Search]] * [[Selectivity]] * [[Reductions]] * RankCut'''
[[FILE:Rank badge.jpg|border|right|thumb| [https://en.wikipedia.org/wiki/Korea Korean]  [https://en.wikipedia.org/wiki/Mandarin_square Rank badge] <ref>[https://en.wikipedia.org/wiki/Korea Korean]  [https://commons.wikimedia.org/wiki/File:Rank_badge.jpg Rank badge],  1850-1900, [https://en.wikipedia.org/wiki/Victoria_and_Albert_Museum V&A Museum] (no. FE.272-1995), [https://en.wikipedia.org/wiki/Wikimedia_Commons Wikimedia Commons]</ref> ]]

Latest revision as of 22:00, 6 November 2019

Home * Search * Selectivity * Reductions * RankCut

a probability based depth reduction technique introduced by Yew Jin Lim and Wee Sun Lee in 2006 [2]. It estimates the probability of discovering a better move later in the search by using the relative frequency of such cases for various states during the search. These probabilities are pre-computed off-line using several self-play games. RankCut can then reduce search effort by performing a shallow search when the probability of a better move appearing is below a certain threshold. RankCut requires good move ordering and fail-soft to work well. Further elaborated by Yew Jin Lim in his 2007 Ph.D. thesis [3], RankCut was successfully implemented with Crafty and Toga II.

RankCut Pseudocode


RankCutReSearch = false;

int RankCut(State & state, int α, int β, int depth) {
  if ((depth == 0) || isTerminal(state))
    return Evaluate(state);
  pruneRest = false;
  score = −∞;
  while (move = NextMove(state) ) {
    r = 0;
    features = determineFeatures(state);
    if (pruneRest || (probability(features) < threshold) ) {
      r = depthReduction(state);
      pruneRest = true;
    score = −RankCut(successor(state, move), −β, −α, depth−1−r);
    if (RankCutReSearch && (score > α) && pruneRest)
      score = −RankCut(successor(state, move), −β, −α, depth−1);
    if (score ≥ β )
    if (score > α) {
      pruneRest = false;
      α = score;
  return score;


In the case of Crafty 19.19, The probability computation considers following features:

See also


External Links


  1. Korean Rank badge, 1850-1900, V&A Museum (no. FE.272-1995), Wikimedia Commons
  2. Yew Jin Lim, Wee Sun Lee (2006). RankCut - A Domain Independent Forward Pruning Method for Games. AAAI 2006
  3. Yew Jin Lim (2007). On Forward Pruning in Game-Tree Search. Ph.D. thesis, National University of Singapore
  4. based on pseudocode pp. 90 in Yew Jin Lim (2007). On Forward Pruning in Game-Tree Search. Ph.D. thesis, National University of Singapore, pdf

Up one level