average

Releases

1. ▸▾0.17.015 Apr 2026

   ExpMovingMean: exponentially weighted moving average (EWMA). Constructors: with_alpha, with_span, with_half_life (the latter two require std or libm f

   Added

   • #[derive(Merge)] proc macro (requires derive feature): automatically derives average::Merge for structs whose fields all implement Merge.
   • #[concatenate] attribute macro (requires derive feature): generates new(), Default, add(), accessor methods, and FromIterator impls for a struct of estimator fields. Use #[statistics(method, ...)] on a field to expose multiple accessors; without the attribute a single accessor named after the field is generated.
   • ExpMovingMean: exponentially weighted moving average (EWMA). Constructors: with_alpha, with_span, with_half_life (the latter two require std or libm feature).
   • ExpMovingVariance: exponentially weighted moving variance. Exposes population_variance(), sample_variance(), standard_deviation() (the last one requires std or libm).
   • ExpMovingMean::append / ExpMovingVariance::append (requires std or libm): combine two estimators in time order so that the result is identical to feeding the concatenated stream to a single estimator. Merge is intentionally not implemented for these types because exponential weighting is order-dependent.
   • Skewness::sample_skewness() and Kurtosis::sample_excess_kurtosis(): the adjusted Fisher-Pearson bias-corrected estimators, matching Excel/SAS/scipy.

   Changed

   • Behavior change: sample_skewness() and sample_excess_kurtosis() on types produced by define_moments! (including the built-in Moments4) now return NaN when n < 3 and n < 4 respectively. Previous versions returned a finite (but mathematically incorrect) value for small samples via a method-of-moments fallback. The formulas have also been corrected and verified against scipy.
   • Skewness and Kurtosis docs: renamed "population variance of the sample" to "biased population variance" for clarity.