sketch demo

Small showcase of asap_sketchlib sketches, plus a benchmark + leaderboard.

Usage examples

cargo run --release -- all     # walk through every sketch with toy data
cargo run --release -- hll     # or: cms, cs, kll, dd

Benchmark + leaderboard

cargo run --release -- bench

Benchmarks every sketch against a shared workload — 1M items, Zipf(s=1.1) over 100k keys, deterministic seed — measuring:

• insert throughput (M items/s, mean ± stddev over 3 runs + 1 warmup)
• query throughput (each sketch's native query: estimate, quantile, ...)
• accuracy vs an exact baseline per category:
  • cardinality → relative error vs HashSet distinct count
  • frequency → mean relative error over the 1k most frequent keys vs HashMap
  • quantiles → tie-aware mean rank error over a 101-point grid vs sorted Vec

The exact baselines are benchmarked through the same pipeline, so each category shows the speed/accuracy trade-off directly.

Output:

• ranked tables printed to the terminal
• leaderboard.md — markdown tables
• leaderboard.html — self-contained dark-mode page with throughput bars (open directly in a browser, no server needed)