Install Machine Learning Plugins

How to install machine learning based plugins (GraPPA and HAT) and their dependencies

Warning

The following plugins are not yet publicly available. As such, this guide is only for internal use and will be updated once the plugins are ready.

Complete Installation with ML Plugins

conda is required for this installation to access the tensorflow and openmm packages. Solving the dependencies of the environment can overflow memory on the original conda solver. So we change the solver to the new libmamba. We also add the conda-forge channel to access the openmm package.

conda install -n base conda-libmamba-solver
conda config --set solver libmamba
conda config --append channels conda-forge

Create a new environment.

conda create -n kimmdy_full python=3.10 tensorflow==2.10 openmm
conda activate kimmdy_full

Clone all repositories and install them in editable mode for development.

git clone git@github.com:graeter-group/kimmdy.git
git clone git@github.com:graeter-group/kimmdy-reactions.git
git clone git@github.com:graeter-group/kimmdy-grappa.git
git clone git@github.com:graeter-group/HAT_reaction_plugin.git
git clone git@github.com:graeter-group/grappa.git

pushd kimmdy
# installs kimmdy, kimmdy-reactions and kimmdy-grappa
pip install -r requirements.txt
popd

pushd HAT_reaction_plugin
pip install -r requirements.txt
popd

pushd grappa
pip install -e .
popd

Potential problems when running

  • openmm can’t find the correct libstdc++.so.6
    • symlink system libstdc++.so.6 to the conda libstdc++.so.6
pushd $CONDA_PREFIX/lib
mv libstdc++.so.6 libstdc++.so.6.old
ln -s /usr/lib/x86_64-linux-gnu/libstdc++.so.6 libstdc++.so.6
popd
Back to top