Compiling diffusioncma

If you want to install the code (and are not planning to make any changes), you can use our anaconda binaries:

conda install -c conda-forge diffusioncma
dcma --help

If this works, you can stop reading here and just use diffusioncma.

If you want to make changes to the python part of the code (the dcma package), install the code from conda first, then clone the github repository, make changes, and python setup.py install. This will keep the C++ core in place and apply the changes to the Python part.

If you want to make changes to the C++ part of the code, continue reading. The core of diffusioncma is written in C++, which means that it needs to be compiled. It is highly recommended to use a conda environment, so you do not have to bother with compiling any of the third-party dependencies.

The recommended way of doing this is as follows

1. Install Requirements: Download and install anaconda from https://www.anaconda.com/download/. I recommend to set up a conda environment which contains all requirements, like so:

   conda create --name dcma_py36 python=3.6
   

   You can also set it up with python 2.7, if you prefer python 2 over python 3.

2. Activate the environment:

   source activate dcma_py36
   

   Note that you have to activate the environment each time you open a new terminal, before the installed packages can be used. (To deactivate a conda environment, type source deactivate).

3. Add the conda-forge channel:

   conda config --add channels conda-forge
   

   and install the required third-party packages:

   conda install "cmake>=3.12.1" "libcmaes>=0.9.6" eigen pybind11 "pytest>=3.4" pytest-runner "click>=6.0" six scipy numpy matplotlib git
   

4. Fork the code on gitlab and clone to your local machine. (Alternatively, if you are not planning to make changes to the code, just clone from my gitlab repo):

   git clone --recursive git@gitlab.com:Your_Gitlab_Handle/diffusioncma.git
   

   or:

   git clone --recursive git@gitlab.com:Olllom/diffusioncma.git
   

   Note the –recursive keyword that forces git to pull the L-BFGS code as a subpackage.

5. Enter the directory and run the setup script:

   cd diffusioncma
   python setup.py compile
   

   this will compile the c++ code (diffusioncma) and install the python wrapper (pydiffusioncma). To install the dcma package:

   python setup.py install  # this will install the dcma package
   

6. To test that the code was installed correctly, try the commands:

   dcma --version
   dcma --help
   

   If you want to be 100% sure, feel free to run the test suite:

   py.test src/python/dcma/pytest/
   

   which should run through in a couple of seconds.

Note

The setup.py supports two different commands for compiling and installing, so you don’t have to recompile the c++ core (diffusioncma) every time you make changes to the python modules (dcma).