ITS parity and reproducibility#
The xyz KSG implementation has been cross-checked against ITS toolbox
nearest-neighbor estimators (via Octave + TSTOOL MEX functions).
Compiled functions#
nn_preparenn_searchrange_search
These are used internally by ITS functions such as its_Eknn, its_BTEknn,
its_PTEknn and its_SEknn.
Reference benchmark#
Dataset: tests/r.csv, k=3, Chebyshev/max metric.
Measure |
Octave ITS |
Python |
|---|---|---|
|
|
|
|
|
|
|
|
|
|
|
|
Agreement is within floating-point precision.
Why parity mattered#
The critical correction was excluding self-neighbors in projected range counts
for KSG TE/PTE/SE, matching ITS range_search(..., past=0) semantics.