Installation

Supported systems

• *nix like OS (Linux, Unix, …)

• Mac OS

Other operating systems have not been tested and they likely require a more complex procedure for the installation (this includes the Microsoft Windows family..).

We reccommend to work in a virtual environment using virtualenv or Anaconda.

Installation requirements

• gcc (or an alternative C/C++ compiler)

• gfortran (or an alternative Fortran compiler)

• python3

• openMPI or mpich

Installing requirements on Ubuntu

sudo apt install gcc libc6-dev
sudo apt install gfortran libgfortran3
sudo apt install libopenmpi-dev

Installing requirements on Mac OS (using homebrew)

brew install llvm
brew install gfortran
brew install open-mpi

Installation

First of all make sure to have the latest version of pip installed

pip install --upgrade pip

The package and its python dependencies can be installed running the commands:

pip install --upgrade TransportMaps

The packages numpy and cython must be installed separately because some dependencies do not list then among the build requirements.

If one whish to enable some of the optional dependencies:

pip install --upgrade TransportMaps[option]

These options will install the following modules:

• DOLFIN – dolfin package for Partial Differential Equations

• SUITESPARSE – scikit-sparse

  • This requires scitik-sparse and the package libsuitesparse-dev

• HMC – Hamiltonian Monte Carlo pyhmc (it requires having numpy and cython already installed)

Manual installation

If anything goes wrong with the automatic installation you can try to install from source:

git clone git@bitbucket.org:dabi86/transportmaps.git
cd transportmaps
pip install --upgrade numpy cython
pip install .

Running the Unit Tests

Unit tests are available and can be run through the command:

>>> import TransportMaps as TM
>>> TM.tests.run_tests()

Or directly using the bash command:

$ tmap-run-tests

The Git repository also contains a docker-compose configuration file to test the whole suite on several versions of python.

There are >2000 unit tests, and it will take some time to run all of them.

FAQ

List of Frequently Asked Questions

The package mpi4py fail to install with error: missing <mpi.h>

One mpi backend must be installed on the machine (e.g. OpenMPI, MPICH) and find out where the corresponding header file <mpi.h> is located. Then the package can be installed by setting the environment variables CPLUS_INCLUDE_PATH or C_INCLUDE_PATH:

CPLUS_INCLUDE_PATH=$CPLUS_INCLUDE_PATH:<mpi_path> \
pip install --upgrade mpi4py