MParT

Monotone Parameterization Toolkit (MParT): A core library for constructing and using transport maps.

MParT: A Monotone Parameterization Toolkit

A CPU/GPU performance-portable library for parameterizing and constructing monotone functions in the context of measure transport and regression.

Documentation

See measuretransport.github.io/MParT/ for more extensive documentation.

Installation

MParT uses a C++ core that needs Kokkos and Eigen directly as dependencies and Catch2 as a test dependency. MParT uses CMake to handle dependencies and compiler configurations. A basic build of MParT that should work on most operating systems can be obtained by running this in the MParT folder:

mkdir build

cd build

cmake \

-DCMAKE_INSTALL_PREFIX=<your/MParT/install/path> \

-DKokkos_ENABLE_PTHREAD=ON \

-DKokkos_ENABLE_SERIAL=ON \

..

make install

Usage

Once MParT is installed a simple usage to create a monotone component from a custom multi-index set is:

% Add path to the MParT installation folder
addpath(genpath('<your/MParT/install/path>'))

% Set number of threads used by KoKKos
num_threads = 8;
KokkosInitialize(num_threads);

% Create multi-index set in 2D
multis = [0,1;2,0;1,1];
mset = MultiIndexSet(multis);
fixedSet = mset.Fix();

% Create monotone component
opts = MapOptions();
mapComponent = CreateComponent(fixedSet,opts);

% Set coefficients
nc = mapComponent.numCoeffs;
mapComponent.SetCoeffs(ones(1,nc));

% Evaluate map at 10 random points
mapComponent.Evaluate(randn(2,10));

Tutorials

Many tutorials and examples using MParT are available on the Matlab folder of the MParT-examples repo. Examples include:

• Monotone regression
• Density estimation
• Bayesian inference

Running these examples will require the following matlab toolboxes:

• Statistics and Machine Learning Toolbox
• Optimization Toolbox