This crate provides estimators for statistics on a sequence of numbers. Thetypical workflow looks like this:

1. If necessary, build your custom estimator usingconcatenate ordefine_moments.
2. Initialize the estimator of your choice withnew().
3. Add some subset (called “sample”) of the sequence of numbers (called“population”) for which you want to estimate the statistic, usingadd()orcollect().
4. Calculate the statistic withmean() or similar.

You can run several estimators in parallel and merge them into one withmerge().

Everything is calculated iteratively in a single pass using constant memory,so the sequence of numbers can be an iterator. The used algorithms try toavoid numerical instabilities.

If you wantSerde support,include"serde" in your list of features.

Note that deserializing does not currently check for all invalid inputs.For example, if you deserialize a corruptedVariance it may returna negative value for variance, even though that is mathematically impossible.In a future minor release some of these checks may be added.

§Example

useaverage::{MeanWithError, Estimate};letmuta: MeanWithError = (1..6).map(f64::from).collect();a.add(42.);println!("The mean is {} ± {}.", a.mean(), a.error());

§Estimators

• Mean (Mean) and its error (MeanWithError).
• Weighted mean (WeightedMean) and its error(WeightedMeanWithError).
• Variance (Variance), skewness (Skewness) and kurtosis(Kurtosis).
• Arbitrary higher moments (define_moments).
• Quantiles (Quantile).
• Minimum (Min) and maximum (Max).

§Estimating several statistics at once

The estimators are designed to have minimal state. The recommended way tocalculate several of them at once is to create a struct with all theestimators you need. You can then implementadd for your struct byforwarding to the underlying estimators. Everything is inlined, so thereshould be no overhead.

You can avoid the boilerplate code by using theconcatenate macro.

Note that calculating moments requires calculating the lower moments, so youonly need to include the highest moment in your struct.

§Calculating histograms

Thedefine_histogram macro can be used to define a histogram struct thatuses constant memory. SeeHistogram10 (defined usingdefine_histogram!(..., 10)) and the extension traitHistogramfor the methods available to the generated struct.