meta numerics

Meta.Numerics library features include advanced functions, function analysis (solvers), statistics and data analysis, linear algebra, and Fourier transforms.

Advanced Functions

The library defines a large number of simple and advanced mathematical functions on real numbers, Complex numbers, integers, and other specialized mathematical objects.

Advanced functions of real and complex numbers include:

Function Real Complex Notes
Gamma also ln Γ, incomplete Gamma
Psi (Digamma) also polygamma ψ(n)
Beta   also incomplete Beta
Error Function also erfc, erf-1, Faddeeva, Fresnel C and S
Exponential Integrals includes Ein, Ei, En, and trigonometric integrals Ci and Si
Bessel J and Y   also for non-integer orders, spherical Bessel j and y
Modified Bessel I and K   also for non-integer orders, Airy Ai and Bi
Coulomb Wave Functions F and G   accurate even in quantum tunneling region
Reimann Zeta also Dirichlet η
Dilogarithm Li2 (Spence's Function) also polylogarithm Lin
Orthogonal polynomials   Chebyshev T, Hermite H, Legendre P, Laguerre L, Zernike R
Elliptic Integrals   Legendre F, K, E; Carlson RF and RD, and RG
Elliptic Functions   Jacobi cn, sn, and dn
Hypergeometric function   2F1

Advanced functions of integers include:

Meta.Numerics also defines various specialized mathematical objects and associated functions:

Object Functions
Complex numbers arithmetic, basic and advanced functions
Vectors and matrices arithmetic, inversion, decompositions
Spinors Clebsch-Gordon coefficients, 3j and 6j symbols
Uncertain values arithmetic and basic functions with error propagation, confidence intervals
Polynomials arithmetic, evaluation, composition, integration and differentiation
Permutations generation, multiplication, inversion, cyclic decomposition and other properties
Integer partitions generation

Numerical Analysis (Solvers)

For arbitrary user-supplied functions, Meta.Numerics supports optimization (minimization and maximization), root-finding, integration, and the solution of ordinary differential equations. All operations are supported on multi-dimensional functions as well as functions of simple real numbers.

Functionality 1-d n-d Notes
optimization also global optimum search
root-finding  
integration advanced multi-grid Monte-Carlo integrator
ordinary differential equations also conservative integrator

Statistics and Data Analysis

Data Wrangling

The library provides a framework similar to the data frame systems familiar to R and Pandas users. Arrays of strongly typed data can be imported and exported from and to CSV and JSON formatted files. Null data entries are supported via .NET's Nullable structure. Data can be filtered, ordered, and transformed in many different ways. The new views produced copy underlying data only when necessary, so manipulations of even large data sets are memory efficient. Data can be passed in column-oriented form to our or any other statistical analysis APIs.

Statistical Analysis

The library provides specialized classes for working with various kinds of data, including:

Type Functionality
Univariate Sample sample and population statistics, transformations, percentile/score conversions, fits to distributions, parametric and non-parametric tests
Bivariate Sample sample and population statistics, regression (linear, polynomial, non-linear, logistic), parametric and non-parametric tests of association
Multivariate Sample sample and population statistics, regression (linear and logistic), cluster and component analysis
Data with Error Bars fit to line, constant, proportionality, polynomial, non-linear function, linear combination of functions
Contingency Table sample and population statistics, parametric and non-parametric tests of association
Histograms sample and population statistics, fits to distributions, parametric tests
Time Series sample and population statistics, power spectrum, difference and integrate, fit to AR and MA models

For each kind of data, methods allow you to evaluate descriptive statistics, fit models, and perform appropriate statistical tests. All fits produce not just the best-fit parameter set, but also error bars, a covariance matrix, and a goodness-of-fit test. Specialized methods make it easy to add, remove, update, and locate data.

Statistical Tests

Some of the many statistical tests supported by the library include:

Parametric Test Nonparametric Alternative Purpose
one-sample t-test sign test compare a sample's mean or median to a reference value
two-sample t-test Mann-Whitney U-test compare the means or medians of two samples
one-way and two-way ANOVA Kruskal-Wallis compare the means or medians of three or more samples
Pearson's r Spearman's rho, Kendall's tau detect association between two continuous variables
Pearson's χ2 test Kendall's exact test detect associated between two categorical variables
Shapiro-Francia Kolmogorov-Smirnov, Kuiper compare continuous sample data to a reference distribution
Ljung-Box test detect autocorrelation in time series

For all tests, we provide exact null distributions for small samples.

Distributions

Meta.Numerics defines a large number of probability distributions, both continuous and discrete:

Beta Cauchy Chi Square Exponential Fisher's F
Frechet Gamma Gumbel Kolmogorov Kuiper
Logistic Lognormal Normal Pareto Pearson's r
Rayleigh Student's t Triangular Uniform Wald
Weibull
Bernoulli Binomial Geometric
Hypergeometric Negative Binomial Poisson
Uniform

For all defined distributions, you can obtain:

  • Basic Descriptive Statistics: mean, median, variance, standard deviation, skewness, excess kurtosis
  • Probability Mass and Probability Density Function (PDF) values
  • Cumulative Distribution Function (CDF) values, integrated from the left or right
  • Inverse CDF values, i.e. percentile to score conversions
  • Arbitrary raw moments, central moments, and cumulants
  • Random deviates

You can also fit sample data to many of the distributions and perform maximum likelihood fits to any user-supplied distribution.

Matrix Algebra

The library defines a number of matrix classes: rectangular, square, symmetric, and tri-diagonal. Each class defines operations appropriate to that matrix type, implemented to exploit the matrix structure for optimum performance. The following table summarizes the available operations:

Operation Rectangular Square Symmetric Tri-diagonal
Arithmetic
Decomposition
Determinant  
Inverse  
Eigenvalues and Eigenvectors  

Available decompositions include LU, QR, and singular value decompositions (SVD).

Extended Precision

We supply a quad precision floating point type that tracks approximately 60 decimal digits.