Rstats is aimed at characterization of multidimensional sets of points, with applications to Machine Learning and Data Analysis. It begins with basic statistical measures and vector algebra, which provide self-contained tools for the more interesting algorithms but can also be used in their own right.
Our treatment of multidimensional sets of points is constructed from the first principles. Thus some original concepts, unlikely to be found elsewhere, are introduced and implemented here.
Going beyond one dimension, other people mostly cheat by using centroids or 'quasi medians' (1-d medians along each axis). They may be quicker to compute but they are a poor start to characterising multidimensional clouds of points reliably. Specifically, all such 1-d measures depend on the choice of axis. Such dependence has to be later removed by Principle Components Analysis or similar methods. In contradistinction to this, our methods based on true Geometric Median, (computed here by nmedian), are axis (or rotation) independent.
RStats is a lean minimalistic library that only depends on anyhow (for its error handling).
Trait RStats is carefully checked and will report all kinds of errors, such as empty input.
Trait Vectors is sometimes unchecked for speed, so some caution is advisable.
One dimensional statistical measures implemented for &[i64] and &[f64].
All these methods operate on one vector of data and take no arguments.
For example, s.amean() computes the arithmetic mean of slice s of either type.
Included are:
&[f64] slices of any length (vector dimensionality).Some of the methods are for memory efficiency reasons reimplemented in this trait so that they mutate self in place instead of creating a new Vec. They are useful in vector iterative methods. Beware that they do not return anything, so they can not be chained.
Version 0.5.0 Introduces VecVec trait for all multi-point methods, now implemented for type Vec<Vec<f64>>. This is a breaking change but it did allow streamlining of the code and a clean separation of the traits. Main benefit to the user is in no longer having to explicitly pass around the dimensionality d.
Version 0.4.15 Added setsub similar to mutzeromd but it subtracts any given vector from a whole set.
Moved all functions into new module functions.rs. The other modules are now exclusively trait implementations.
Version 0.4.14 Added mutzeromd - or mutated zero median data, which transforms mutable self to zero-median form.
Version 0.4.13 Added trend between two sets of points. More comments, tests and examples.
Version 0.4.12 Some more utilities.
Version 0.4.10 Moved unimportant helper functions out of the main module.
Version 0.4.9 Streamlining, introduced distances and eccentricities, with speedups up to 50%.
Version 0.4.8 Added generics to emphasise print vectors and items of various types.
Version 0.4.7 Added Spearman's Rho Correlation. Removed some spurious indirections.
Version 0.4.6 Made eccentricity measure continuous. Added Kendall's Tau (rank) correlation. Moved all correlations to Vectors trait. Improved readme and doc comments.
Version 0.4.4 Medoid now finds the Outlier as well. Improved tests. Defined and added MOE = median of eccentricities (spread) of multivariate data.
Version 0.4.3 Introduced computation of nmedian residual errors. Improved tests and benchmarks.
Version 0.4.2 Added eccentricity measures.
Version 0.4.1 Tidier testing and benchmarks.
Version 0.4.0 Cleanup. Changed the implementation types from Vecs to slices everywhere for consistency. You may need more .as_slice() conversions here and there. Made some subsidiary functions private.