Skip to content

Latest commit

 

History

History
72 lines (44 loc) · 2.4 KB

README.md

File metadata and controls

72 lines (44 loc) · 2.4 KB
Raw
                                                                              
                                    ███████████      █████      ███████████    
                                    ███     ███    █████████    ███     ███    
                                    ███     ███   ████   ████   ███     ███    
                                    ███           ███     ███   ███     ███    
                                    ███           ███     ███   ███████████    
                 ██            ██   ███   █████   ███████████   ███            
    ██████████        ██████        ███     ███   ███     ███   ███            
    ██  ██  ██   ██   ██  ██   ██   ███     ███   ███     ███   ███            
    ██  ██  ██   ██   ██  ██   ██   ███████████   ███     ███   ███

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