Insert rstats = "^1" in the Cargo.toml file, under [dependencies].
Use in source files any of the following structs, as needed:
use rstats::{MinMax,Med,Mstats};
and any of the following helper functions:
use rstats::{i64tof64,tof64,here,wi,wv,wsum,printvv,genvec,genvecu8};
and any of the following traits:
use rstats::{Stats,MutStats,Vecu8,Vecf64,Vecg,MutVecg,VecVec,VecVecg};
It is highly recommended to read and run tests/tests.rs, which shows examples of usage.
To run all the tests, use single thread in order to produce the results in the right order:
cargo test --release -- --test-threads=1 --nocapture --color always
Rstats is primarily about characterising multidimensional sets of points, with applications to Machine Learning and Big Data Analysis. It begins with basic statistical measures and vector algebra, which provide self-contained tools for the multidimensional algorithms but can also be used in their own right.
Our treatment of multidimensional sets of points is constructed from the first principles. Some original concepts, not found elsewhere, are introduced and implemented here. Specifically, the new multidimensional (geometric) median algorithm. Also, the comediance matrix as a replacement for the covariance matrix. It is obtained simply by supplying covar with the geometric median instead of the usual centroid (mean vector).
Zero median vectors are generally preferable to the commonly used zero mean vectors.
Most authors 'cheat' by using quasi medians (1-d medians along each axis). Quasi medians are a poor start to stable characterisation of multidimensional data. In a highly dimensional space, they are not even any quicker to compute.
Specifically, all 1-d measures are sensitive to the choice of axis and thus are affected by rotation.
In contrast, analyses based on the true geometric median (gm) are axis (rotation) independent. They are computed here by the novel methods smedian and gmedian and their weighted versions wsmedian and wgmedian.
The main constituent parts of Rstats are Rust traits, grouping together methods applicable to a single vector (of numbers) - Stats, two vectors - Vecg, n vectors - VecVec or n vectors and another generic argument - VecVecg. End type f64 is most commonly used for the results, whereas inputs to the generic methods can be vectors (or slices) of any numeric end types.
To see more detailed comments, plus some examples, see the source.
struct Med to hold median and quartiles
struct MStats to hold mean and standard deviation
struct MinMax reexported from crate indxvec to hold min and max values of a vector and their indices. It is returned by function indxvec::merge::minmax.
auxiliary functions: i64tof64, tof64, here, wsum, wi, wv, printvv, genvec, genvecu8.
One dimensional statistical measures implemented for all numeric end types.
Its methods operate on one slice of generic data and take no arguments.
For example, s.amean() returns the arithmetic mean of the data in slice s.
Some of these methods are checked and will report all kinds of errors, such as an empty input. This means you have to call .unwrap() or something similar on their results.
Included in this trait are:
A few of the Stats methods are reimplemented under this trait
(only for f64), so that they mutate self in-place.
This is more efficient and convenient in some circumstances, such as in
vector iterative methods.
Vector algebra operations between two slices &[T], &[U] of any length (dimensionality):
This trait is unchecked (for speed), so some caution with data is advisable.
A handful of primitive methods as in Vecg but operating on an argument of known end type f64 or &[f64].
Mutable vector addition, subtraction and multiplication.
Mutate self in-place.
This is for efficiency and convenience. Specifically, in
vector iterative methods.
MutVecf64 is to be used in preference, when the end type of self is known to be f64. Beware that these methods work by side-effect and do not return anything, so they can not be functionally chained.
Relationships between n vectors (in d dimensions). This is the original contribution of this library. True geometric median is found by fast and stable iteration, using improved Weiszfeld's algorithm boosted by a secant method. These measures solve Weiszfeld's convergence problems in the vicinity of existing set points.
Trait VecVec is entirely unchecked, so check your data upfront.
Methods which take an additional generic vector argument, such as a vector of weights for computing the weighted geometric medians.
Centroid\Centre\Mean is the (generally non member) point that minimises the sum of squares of distances to all member points. Thus it is susceptible to outliers. Specifically, it is the n-dimensional arithmetic mean. By drawing physical analogy with gravity, it is sometimes called 'the centre of mass'. Centroid can also sometimes mean the member of the set which is the nearest to the Centre. Here we follow the common (if somewhat confusing) usage: Centroid = Centre = Arithmetic Mean.
Quasi\Marginal Median is the point minimising sums of distances separately in each dimension (its coordinates are 1-d medians along each axis). It is a mistaken concept which we do not use here.
Tukey Median is the point maximising Tukey's Depth, which is the minimum number of (outlying) points found in a hemisphere in any direction. Potentially useful concept but not yet implemented here, as its advantages over gm are not clear.
Medoid is the member of the set with the least sum of distances to all other members.
Outlier is the member of the set with the greatest sum of distances to all other members.
Median or the true geometric median (gm), is the point (generally non member), which minimises the sum of distances to all members. This is the one we want. It is much less susceptible to outliers than centroid. In addition, unlike quasi median, gm is rotation independent.
Zero median vector is obtained by subtracting the geometric median. This is a proposed alternative to the commonly used zero mean vector, obtained by subtracting the centroid.
Comediance is similar to covariance, except zero median vectors are used to compute it, instead of zero mean vectors.
Version 1.0.3 Better implementations of joint probability and joint entropy. Code style and testing improvements.
Version 1.0.2 Updated the dependency indxvec to version 1. A few minor changes to this document.
Version 1.0.1 Minor change: sortedeccs and wsortedeccs now take gm as an argument for more efficient repeated use. Vecvec test improved.
Version 1.0.0 Rstats reaches stability (of sorts)! Some code simplifications: smedian and wsmedian are now just slightly more accurate than gmedian and wgmedian respectively, otherwise their performance is very similar. Sometimes a bit slower, at other times, especially on 'difficult' data, they can be a bit faster.
Version 0.9.5 Improved smedian. Also added a weighted version of it: wsmedian.
Version 0.9.4 Organisation improvements. Added trait Vecf64 and moved into it relevant methods from Vecg. Added a few functions to MutVecf64 trait. Simplified gmedian.
Version 0.9.3 Added hwmeanstd - harmonic weighted mean and standard deviation. Tidied up readme badges and some tests. Simplified random number generation. Weights for the weighted means are now ascending (more intuitive).
Version 0.9.2 Fixed some tests broken by moved functions. Added harmonic standard deviation, harmonic centroid and more tests.
Version 0.9.1 Made the auxiliary functions more visible by moving them to lib.rs (the top level of the crate).
Version 0.9.0 Added kron and outer products to Vecg trait.