Skip to content

0.1.3

Choose a tag to compare

@andreaderuvo andreaderuvo released this 02 Apr 11:12
· 35 commits to main since this release

What's Changed

Performance

  • 6.5x faster CPU distance calculation (numpy vectorized kernel replacing numba triple-loop)
  • 123x faster with GPU (optimized CUDA kernel with 32x32 thread blocks)
  • Vectorized loci/sample completeness filtering
  • Optimized matrix I/O with pandas to_csv and buffered writes

Benchmarks (5,000 samples × 3,000 loci)

Method Calc Time Speedup
v0.1.1 CPU (8 threads) 55.5s 1x
v0.1.3 CPU (8 threads) 8.5s 6.5x
v0.1.3 GPU (NVIDIA L4) 0.45s 123x

Other Changes

  • Added CHANGELOG.md
  • Updated Dockerfile to Python 3.13-slim
  • Added GitHub Actions CI/CD (tests + Docker publish to ghcr.io on release)
  • Unpinned dependency versions for broader compatibility
  • Fixed mixed-dtype issue with INF- prefixed allele calls