Fix initializing clusters with random randoms
During the quantization process, an instance of a Quantizer is created.
When quantization is run, the Quantizer may need to create starting
points that are random. To keep results stable, meaning multiple runs of
a quantizer output the same results, those starting points must
be created using a Random with the same starting seed. Without
specifying a starting seed, or if an instance of Random is reused across
runs, the random starting points will vary across quantizer runs.
With this patch, the Random used to create starting points is no longer
used across multiple runs (i.e. it isn't a static variable anymore).
Additionally, a starting seed is specified.
Test: Add log dumps after the quantization process. Verify that results
from Kmeans remain consistent as long as the input is consistent.
Bug: 182333325
Change-Id: Ia0f984349b7885ae1b44536aaa1b286a28e50587
2 files changed