███████████      █████      ███████████    
                                    ███     ███    █████████    ███     ███    
                                    ███     ███   ████   ████   ███     ███    
                                    ███           ███     ███   ███     ███    
                                    ███           ███     ███   ███████████    
                 ██            ██   ███   █████   ███████████   ███            
    ██████████        ██████        ███     ███   ███     ███   ███            
    ██  ██  ██   ██   ██  ██   ██   ███     ███   ███     ███   ███            
    ██  ██  ██   ██   ██  ██   ██   ███████████   ███     ███   ███            
                                                                               

Introduction

miniGAP is a proxy application for molecular and materials property prediction using the Gaussian Process Approximation. This is code is meant to run in multiple architectures, such as many-core and accelerators.

Installation

This code could be installed within an conda enviroment as:

conda env create -f environment.yml

Then the new environment is activated as:

conda activate minigap

Creating an custom kernel in Jupiter

conda activate minigap
python -m ipykernel install --user --name "minigap"

Dependencies:

• python >= 3.6
• dscribe
• SYCL compiler
• Tensorflow
• Tensorflow-probability
• GPflow
• scikit-learn

What is inside?

• data: Initial XYZ, sample trajectories, and downloaded material.
• code: Repo specific modules for training and creating the models.
• notebooks:
• results: Figures and models
• media: Assorted Images

Contributors

Contributions are always welcome. Contributors should fork this repository and submit a merge request for review of the code.

References

Dscribe

GAP

Copyright 2021 Argonne UChicago LLC