Installation

Source Code

To obtain the CLIFF source code, clone from the main repository with

git clone https://github.com/jeffschriber/cliff.git

Dependencies

To handle the dependencies in CLIFF, it is recommended to make a conda environment,

conda create --name cliff python=3.6

Next we will install the required libraries:

conda install numpy
conda install scipy
conda install qcelemental

and the pytest library for running the tests:

conda install pytest

There exist two sources of ML models for prediction atomic properties. A KRR-based one (from the article) uses the external library QML to handle the computation of descriptors for machine learning models. Documentation of QML can be found at https://www.qmlcode.org/. The simplest way to install QML is with pip,

pip install qml --user -U

Note that QML also requires a Fortran compiler which is available from conda. As a final piece of setup, all Kernel-Ridge Regression model files located in cliff/models/large need to be untarred.

Alternatively, the properties can be obtained from APNET, available at https://github.com/zachglick/apnet. Note that this is a private repository, so you must have permission to use it. Correspondingly, APNET-derived atomic properties is an experimental feature, so be cautious.

Running the Tests

Once CLIFF and all dependencies are obtained, it is recommended to run the test suite. This can be done by going into the tests directory and calling pytest:

cd cliff/tests
pytest