Releases: cwida/PDX
Releases · cwida/PDX
v0.2
- New feature: Scalar Quantization with 8 bits on PDX+ADSampling. We calculate distances in 8-bits (symmetric calculations) instead of decoding to floats.
- New algorithm: IVF2, A Two-level IVF index that reduces the time to find the nearest centroids in larger collections. Achieves extremely high throughputs when paired with SQ8. Check some benchmarks here.
- Improvements to the PDX data layout to tackle the poor performance gains when K > 10. Read more here. This new layout substantially reduce I/O on the PRUNE phase of the algorithm.
- Added new datasets from the VIBE benchmark.
- Created a playground to test the raw performance of PDX vs SIMD kernels.
- Removing unnecessary code from SIGMOD experiments.
- Added MIT LICENCE.
- Improved README.
- Improved examples.