Releases: seijikun/kmean-rs
kmeans v2.0.0
Major version bump according to semantic versioning standards, due to API change.
Changes
- Got rid of an unnecessary copy for the samples buffer, in case it is used multiple times. Thanks to @asm-volatile
Contributors
Assets 2
kmeans v1.1.0
Well, that didn't go as planned. Right after v1.0.0, another API change was required to fix a soundness bug.
So instead of going for v2.0.0 right away, I yanked v1.0.0 to replace it with this one.
Changes
- Previously, striding and padding logic leaked into the user API via
KMeansStatestruct. Padded/strided buffers now use theStrideBufferstruct that still allows accessing the raw memory inside kmeans, but only allows unpadded access outside of kmeans - Critical Fix: Replace broken custom aligned buffer allocation with
aligned-veccrate
kmeans v1.0.0
Major version bump according to semantic versioning standards, due to API change.
Changes
- Added ability to specify distance metric (Also with custom external ones). Big thanks to @reinismu
- Fixed unit-tests to use deterministic random
Contributors
Assets 2
kmeans v0.11.0
Added new initialization method for precomputed values (Thanks to @mfigurski80 !)
Contributors
Assets 2
kmeans v0.10.0
Changes:
- Since
packed_simdis broken now, kmeans was ported to using the official nightly featureportable_simd. (Big thanks to @wzh4464 !) - To use the
KMeansstruct, one now has to explicitly select the amount of SIMD lanes to try to use (KMeans<f64, 8>)
Contributors
Assets 2
Fix build with new packed_simd
Thanks goes out to @stelzo
Contributors
Assets 2
kmeans v0.2.0
Finally, all required crates were updated and I was able to publish a new kmeans version to crates.io.
The packed_simd dependency was (at least temporarily) revived with packed_simd_2.
Though in the future, kmeans will probably have to be ported to whatever the new Portable SIMD Project Group will come up with.
This version now finally has proper documentation and successfully builds ;).