lacerbi/lautils-mat

Luigi Acerbi's utility functions for MATLAB.

lautils-mat: Luigi Acerbi's utility functions for MATLAB

This is a collection of MATLAB utility functions that I wrote (unless otherwise specified), and that I share among a number of projects. Functions in the following list have been consistently used or at least somewhat tested. Undocumented functions in the repository might be untested/deprecated, so use only at your own risk.

Functions marked with (+) are pretty useful functions that I use regularly. Functions marked with (++) have been game-changers for several projects.

Functions are loosely divided in folders by category.

Math (./math)

Numerical and math-related functions.

• (++) bsxfunandsum.m: Element-by-element matrix operation followed by sum (prevents memory overflow with huge matrices)
• derivcheck.m: Check analytical vs numerical differentiation for a function
• fgrad.m: Estimation of function gradient via finite differences
• fhess.m: Estimation of function Hessian via finite differences
• intpower.m: Array integer power (faster than .^ with large matrices)
• (+) lininterp1.m: Linear 1D interpolation on a regular grid (like interp1.m, but faster)
• logsumexp.m: Compute log(sum(exp(X))) while avoiding numerical underflow
• nansumall.m: Sum of all elements, ignoring NaN values
• (+) qcumtrapz.m: Quick cumulative trapezoidal numerical integration
• (++) qtrapz.m: Quick trapezoidal numerical integration (3-4 times faster than trapz.m on large arrays)

Markov Chain Monte Carlo sampling (./mcmc)

MCMC samplers and related files.

• maxsample.m: Stochastic exploration of high-valued regions of a function (untested)
• mhsamplecon.m: Metropolis-Hastings with reflective constraints (untested)

Optimization (./optim)

Optimization algorithms and related functions.

• cmaes_wrapper.m: A wrapper with a simplified interface for the CMA-ES algorithm (requires cmaes.m)
• coordtransf.m: Coordinate transform
• fminfold.m: Multi-fold optimization (work in progress); uses fminfold_wrap.m
• fmingrid.m: Multidimensional minimization via grid search
• (+) fminmulti.m: Function minimization via recursive multi-start method (usable but needs improvement)
• (+) funlogger.m: Record log of calls to a specific function
• lhs.m: Latin hypercube sample
• (+) qargmax1: Quick and dirty numerical argmax via 1-D quadratic interpolation
• transvars.m: Another coordinate transform function

Plotting (./plot)

Plotting and graphics functions.

• graph.m: Prepare a boxed, annotated graph
• multigraph.m: Prepare a graph with multiple panels (usable, but needs rehauling)
• scatterbars.m: Scatter/bubble plot with error bars (unused?)

PsychToolbox Add-ons (./psych)

A bunch of functions for PsychToolbox that I wrote during my PhD. Nothing particularly useful.

Statistics (./stats)

Probability distributions and other statistics-related functions.

• binbuild.m: Build binned data
• binolike.m: Negative log-likelihood for the binomial distribution
• (++) bsxfun_normcdf.m, bsxfun_normlogpdf.m, bsxfun_normpdf.m: Vectorized normal cdf, log pdf and pdf
• (+) fisher2kappa.m: Fisher information to Von Mises concentration parameter
• gaussks1.m: 1-D Gaussian kernel smooth function approximation
• gpinterp1.m: 1-D interpolation using Gaussian process (requires GPML toolbox)
• moments.m: Central moments of discrete distribution
• (++) qrandvm.m: Quick Von Mises distributed pseudorandom numbers (faster than circ_vmrnd.m from the Circular Statistics Toolbox)
• randntrim.m: Normally distributed trimmed pseudorandom numbers (dumb rejection method)
• randx.m: Uniformly distributed non-overlapping random points
• samplemeanpdf.m: Sampling distribution of the mean (see also samplemeanpdf_test.m)
• sqmean.m, sqsum.m: Return root of squared mean/sum of data
• stderr.m: Standard error of the mean (ignores NaNs)

Standard and input/output functions (./stdio)

Functions that deal with format conversion, argument parsing, structure manipulation and related miscellanea.