source: branches/PerformanceComparison/ProblemInstanceIdentifier/InstanceDescriptor.cs @ 15016

using System;
using System.Collections.Generic;
using System.Linq;
using HeuristicLab.Analysis.FitnessLandscape;
using HeuristicLab.Common;
using HeuristicLab.Data;
using HeuristicLab.Encodings.PermutationEncoding;
using HeuristicLab.Problems.QuadraticAssignment;

namespace ProblemInstanceIdentifier {
  public class InstanceDescriptor {
    public string Name { get; set; }
    public string Cls { get; private set; }
    public int Dimension { get; set; }

    public string[] FeatureNames { get; set; }
    public double[] FeatureValues { get; set; }
    
    private InstanceDescriptor() { }

    public InstanceDescriptor(string name, string cls, int dimension, string[] names, double[] values) {
      Name = name;
      Cls = cls;
      Dimension = dimension;
      FeatureNames = names;
      FeatureValues = values;
    }

    public InstanceDescriptor(InstanceDescriptor other) {
      Name = other.Name;
      Cls = other.Cls;
      Dimension = other.Dimension;
      FeatureNames = (string[])other.FeatureNames.Clone();
      FeatureValues = (double[]) other.FeatureValues.Clone();
    }

    public static InstanceDescriptor FromProblemOnly(QuadraticAssignmentProblem qap) {
      return new InstanceDescriptor() {
        Name = qap.Name,
        Cls = GetClass(qap.Name),
        Dimension = qap.Weights.Rows,
        FeatureNames = new string[0],
        FeatureValues = new double[0]
      };
    }

    public static InstanceDescriptor FromPaths(QuadraticAssignmentProblem qap, List<List<Tuple<Permutation, double>>> trajectories) {
      var result = PermutationPathAnalysis.GetCharacteristics(trajectories);

      return new InstanceDescriptor() {
        Name = qap.Name,
        Cls = GetClass(qap.Name),
        Dimension = qap.Weights.Rows,
        FeatureNames = result.GetNames(),
        FeatureValues = result.GetValues()
      };
    }

    public static string GetClass(string name) {
      var cls = name.Substring(0, 3);
      var subCls = name.Last();
      switch (cls) {
        case "lip":
          cls = name.Substring(0, 4) + "-" + subCls;
          break;
        case "RAN":
          cls = name.Substring(0, 4) + "-" + name[name.Length - 2] + name[name.Length - 1];
          break;
        case "tai":
          if (char.IsLetter(subCls)) cls += "-" + subCls;
          break;
      }
      return cls;
    }

    public double CalculateSimilarity(InstanceDescriptor other) {
      return FeatureValues.Select((v, i) => (v - other.FeatureValues[i]) * (v - other.FeatureValues[i])).Sum();
    }

    public override string ToString() {
      return Name;
    }
  }

  public class InstancesStandardizer {
    private double[] featureMeans;
    private double[] featureStdevs;

    public IEnumerable<double> GetMeans() {
      return featureMeans;
    }
    public IEnumerable<double> GetStdevs() {
      return featureStdevs;
    }

    private InstancesStandardizer() { }

    public static InstancesStandardizer Create(IList<InstanceDescriptor> instances) {
      var standardizer = new InstancesStandardizer();
      var featureLength = instances.First().FeatureValues.Length;
      standardizer.featureMeans =
        Enumerable.Range(0, featureLength)
        .Select(x => instances.Select(y => y.FeatureValues[x]).Average()).ToArray();
      standardizer.featureStdevs =
        Enumerable.Range(0, featureLength)
        .Select(x => instances.Select(y => y.FeatureValues[x]).StandardDeviation()).ToArray();

      return standardizer;
    }
    public static InstancesStandardizer CreateAndApply(IList<InstanceDescriptor> instances) {
      var standardizer = Create(instances);
      standardizer.Apply(instances);
      return standardizer;
    }

    public void Apply(IList<InstanceDescriptor> instances) {
      for (var i = 0; i < instances.Count; i++) {
        var inst = instances[i];
        for (var x = 0; x < featureMeans.Length; x++)
          inst.FeatureValues[x] = (inst.FeatureValues[x] - featureMeans[x]) / featureStdevs[x];
      }
    }
  }
}