figwig benchmark: random 2k-window reads up to ~40x faster than bigwig for ML training
jmschreiber91 · x · 2026-10-28
@jmschreiber91 benchmarked the figwig genomics data format for ML training workloads (e.g. Cherimoya/ChromBPNet), reading a random 2k window per file, single thread. figwig took 0.0029 ms (6 count tracks) and 0.0020 ms (48 tracks) per window per track, versus 0.036/0.014 ms for bigwig, 12.3-12.8 ms for d4tools view, and 272-273 ms for pyd4 — up to 40x faster than bigwig. He also noted Claude flagged potential correctness issues in figwig, unverified so far.
More from coding & agent
- Row-Bot: a local 27B model runs a shop's morning ops entirely on one PC via MCP — Acceptable-Object390 · 2026-10-28
- Offboarding AI agents: revoke sessions first, deleting the container isn't enough — ComprehensiveShake76 · 2026-10-28
- The compounding error problem: 95% per-step accuracy means 36% over 20 steps — KelvenSevero0 · 2026-10-28
- Dev shares AI game project's review list: laser bugs, podracer handling fixes queued — jasonkneen · 2026-10-28
- Dev builds online multiplayer castle-storming game with GPT 6.1 Terra, playable on mobile — chongdashu · 2026-10-28
- Self-distillation explained: how a model teaches itself with richer token-level training signals — arpit_bhayani · 2026-10-28