DaviesBouldinEvaluation: incorrect DB index due to stale loop variable

[beingamanforever]

evaluate() computes the Davies–Bouldin index using Dij(i, :) after the loop, where i is the stale loop variable (ends at k-1). This uses only one row instead of all clusters.

Current code:

this.CriterionValues(iter) = sum (max (Dij(i, :), [], 2)) / ...
                                    this.InspectedK(iter);

Impact

DB index is underestimated/incorrect
Only one term of the sum is used instead of all k clusters. I would be working on this issue

[pr0m1th3as]

It is important to add an example comparing Octave and MATLAB output, when opening an issue about incorrect output.
Additionally, before opening a PR, please make sure you test the output against MATLAB. Cluster criterion classes are not well tested, so possible changes might introduce regressions.

[beingamanforever]

Oh sure, i will keep this in mind, letme do the needful, thanks for pointing out!

[beingamanforever]

This is after fix:

Before fix it was outputing: Case 1: 0.013333, Case 2: 0.1943

MATLAB output:

The issue was that the code was doing sum(max(Dij(i, :), [], 2)) / k but it should do, sum(max(Dij, [], 2)) / k, hence it uses only one cluster’s value and ignores others