public TemplateModelFeatureSelecting(
IFeatureSetSelectorDiscrete <DomainType, LabelType> selector,
ITemplateModelDiscrete <DomainType, LabelType> template)
{
this.selector = selector;
this.template = template;
}
public double Evaluate(ITemplateModelDiscrete <DomainType, LabelType> template, IDataSet <DomainType, LabelType> data_set, ISet <int> feature_set)
{
IDataSet <DomainType, LabelType> selected_data_set = data_set.SelectFeatures(new List <int>(feature_set));
double[] scores = new double[this.FoldCount];
Parallel.For(0, this.FoldCount, fold_index =>
{
Tuple <IDataSet <DomainType, LabelType>, IDataSet <DomainType, LabelType> > split = selected_data_set.Split(this.TrainingSetFraction);
ReportDiscrete <DomainType, LabelType> report = template.GenerateAndTestDiscrete(split.Item1, split.Item2);
scores[fold_index] = report.CorrectLabelRate;
});
return(ToolsMathCollection.Sum(scores) / ((double)this.FoldCount));
}
public Tuple <IModelDiscrete <DomainType, LabelType>, IList <int> > SelectFeatureSet(ITemplateModelDiscrete <DomainType, LabelType> template, IDataSet <DomainType, LabelType> data_set)
{
ISet <ISet <int> > black_list = new HashSet <ISet <int> >();
ISet <int> full_set = new HashSet <int>();
for (int feature_index = 0; feature_index < data_set.FeatureCount; feature_index++)
{
full_set.Add(feature_index);
}
ISet <int> current_set = new HashSet <int>();
double best_score = feature_set_evaluator.Evaluate(template, data_set, current_set);
bool improvement = true;
while (improvement)
{
List <ISet <int> > options = new List <ISet <int> >();
options.AddRange(RemoveFeature(current_set, black_list));
options.AddRange(AddFeature(current_set, black_list, full_set));
improvement = false;
foreach (ISet <int> option in options)
{
double score = feature_set_evaluator.Evaluate(template, data_set, option);
//System.out.print("scorein " + score);
//CollectionTools.print(option);
if (best_score.CompareTo(score) == -1)
{
//System.out.print("New best scorein " + score);
//CollectionTools.print(option);
best_score = score;
current_set = option;
improvement = true;
}
}
}
List <int> current_list = new List <int>(current_set);
current_list.Sort();
IModelDiscrete <DomainType, LabelType> model = template.GenerateModelDiscrete(data_set.SelectFeatures(current_list));
return(new Tuple <IModelDiscrete <DomainType, LabelType>, IList <int> >(model, current_list));
}
