Implement population diversity analysis

[swagner]

Several commits by swinkler between r4402 and r4701.

[swagner]

Merged into trunk in r4703.

[swagner]

Updated GA & TSP sample in r4705.

[swagner]

Worked on in r4715.

[swagner]

Removed branch in r4738.

[swagner]

Worked on in r4739.

[swagner]

Worked on in r4748.

[gkronber]

Reviewed classes HeatMap, HeatMapHistory, PopulationDiversityAnalyzer, DataTableHistory, !TSPPopulationDiversityAnalyzer and made minor changes with r4848.

[gkronber]

Edge similarity for TSP is a bit difficult to understand but seems to be OK.

private double CalculateSimilarity(int[] edgesA, int[] edgesB) {
      // calculate relative number of identical edges
      int identicalEdges = 0;
      for (int i = 0; i < edgesA.Length; i++) {
        if ((edgesA[i] == edgesB[i]) || (edgesA[edgesB[i]] == i))
          identicalEdges++;
      }
      return ((double)identicalEdges) / edgesA.Length;
    }

[gkronber]

Please review my changes and assign the ticket for release.

[gkronber]

Additional remarks: in the PopulationDiversityAnalyzer it is implicitly assumed that the similarities calculated by specialized diversity analyzers are scaled to the range [0..1]. This assumption is never checked explicitly and implementers of specialized diversity analyzers must know this.

It would be possible to implement the abstract PopulationDiversityAnalyzer class in a way that it works correctly for arbitrary similarity ranges.

[gkronber]

Reviewed classes DataRowVisualProperties, DataRow, DataTable, HeatMapView, HeatMapHistoryView and DataTableHistoryView

[swagner]

Replying to gkronber:

Additional remarks: in the PopulationDiversityAnalyzer it is implicitly assumed that the similarities calculated by specialized diversity analyzers are scaled to the range [0..1]. This assumption is never checked explicitly and implementers of specialized diversity analyzers must know this.

It would be possible to implement the abstract PopulationDiversityAnalyzer class in a way that it works correctly for arbitrary similarity ranges.

Yes, that's true. I added an additional check in PopulationDiversityAnalyzer to ensure that the similarities are in the intervall [0;1] (see r4862).