Free cookie consent management tool by TermsFeed Policy Generator

Ignore:
Timestamp:
01/20/15 20:25:00 (9 years ago)
Author:
gkronber
Message:

#2283: separated value-states from done-states in GenericGrammarPolicy and removed disabling of actions from bandit policies

File:
1 edited

Legend:

Unmodified
Added
Removed
  • branches/HeuristicLab.Problems.GrammaticalOptimization/HeuristicLab.Algorithms.Bandits/BanditPolicies/ChernoffIntervalEstimationPolicy.cs

    r11792 r11806  
    2121      // select best
    2222      var myActionInfos = actionInfos.OfType<DefaultPolicyActionInfo>();
    23       int k = myActionInfos.Count(a => !a.Disabled);
    24       int totalTries = myActionInfos.Where(a => !a.Disabled).Sum(a => a.Tries);
     23      int k = myActionInfos.Count();
     24      int totalTries = myActionInfos.Sum(a => a.Tries);
    2525      double bestQ = double.NegativeInfinity;
    2626      var bestActions = new List<int>();
     
    2828      foreach (var aInfo in myActionInfos) {
    2929        aIdx++;
    30         if (aInfo.Disabled) continue;
    3130        double q;
    3231        if (aInfo.Tries == 0) {
     
    4645          bestActions.Clear();
    4746          bestActions.Add(aIdx);
    48         } else if (q == bestQ) {
     47        } else if (q.IsAlmost(bestQ)) {
    4948          bestActions.Add(aIdx);
    5049        }
Note: See TracChangeset for help on using the changeset viewer.