Open ameyaacharya opened 10 years ago
In b3de91, most of the output makes sense (vectors dotted with itself will have 0 Mahalanobis distance, which makes sense). But, we still have anomalies like this:
[1 1 1 3] [1 0 0 4] Diff at 1 Diff at 2 Diff at 3 1.97020835811
[1 0 0 4] [1 2 0 4] Diff at 1 1.97563556343
The first pair of vectors is different in 3 positions, but they are "closer" than the second pair of vectors because the second pair of vectors has a "distance" of 2 in the second position.
Is this what we want? Maybe.
Actually, this is a better example of what's wrong:
[1 1 1 3] [1 0 0 4] Diff at 1 Diff at 2 Diff at 3 1.97020835811
[1 1 1 3] [1 0 0 3] Diff at 1 Diff at 2 1.98913558085
Why is the second pair of vectors farther than the first one?
This is dependent on #13.
Note: use python distance for now.
Currently, the calculation between our 12-vectors in do_mahal_distance in covariance.py is incorrect. When we return the sorted list of pairwise distances, there are some "dissimilar" vectors that are supposedly "more similar than" or "the same as" a vector against itself. For example: [3 1 0 4] [3 1 0 4] 79512674.1057 [3 1 0 4] [0 3 0 4] 79512674.1057.