avantgardnerio commented on code in PR #2294: URL: https://github.com/apache/datafusion-ballista/pull/2294#discussion_r3786701328
########## benchmarks/benches/quantile_sketch.rs: ########## @@ -38,9 +144,26 @@ use std::sync::Arc; use ballista_core::kll::KllSketch; +/// The shipped capacity, imported rather than restated: `kll_norm_u64` is +/// only comparable to `sort_key_sketch_f64` while the two sketches are +/// sized identically, and a local copy could drift out from under that +/// claim. Picked for worst-case rank-error parity with TDigest at +/// `TDIGEST_MAX_SIZE=100` on the uniform 1M stream this bench feeds: +/// +/// | sketch | worst rank err (9 deciles) | +/// |------------------------------|---------------------------:| +/// | TDigest max_size=100 | 0.0021 | +/// | KLL k=800 | 0.0016 | +/// +/// Rerun via `KLL_PARITY_CHECK=1 cargo bench --bench quantile_sketch`. +use ballista_core::sort_key::KLL_K; Review Comment: > **`KLL_K = 800` is duplicated** — private in `sort_key.rs`, redeclared in the bench Fixed. It is `pub` in `sort_key` now and the bench imports it, so the arms cannot be sized differently. The parity table moved onto the import. -- This is an automated message from the Apache Git Service. To respond to the message, please log on to GitHub and use the URL above to go to the specific comment. To unsubscribe, e-mail: [email protected] For queries about this service, please contact Infrastructure at: [email protected] --------------------------------------------------------------------- To unsubscribe, e-mail: [email protected] For additional commands, e-mail: [email protected]
