DeepDriveMD Documentation
DeepDriveMD: Deep-Learning Driven Adaptive Molecular Simulations.
- Release
0.0.2
- Date
May 17, 2022
Summary
deepdrivemd is a Python package for coupling molecular dynamics ensemble simulations to sampling agents guided by machine learning.
DeepDriveMD can support two modes of execution, one is synchronous and runs MD simulations, aggregation, training, and inference stages in a pipeline where each stage blocks the others and the stages communicate via the filesystem (DeepDriveMD-F).
The second, and more optimal, mode of execution is asynchronous where each of the previously mention stages run continously as independent components communicating via adios2 to stream data between concurrently running workflow components, enabling efficient feedback between simulations and learning algorithms (DeepDriveMD-S).
Both modes of execution are implemented using RADICAL-Ensemble Toolkit to enable support for large scale runs on high-performance computing platforms.
Additional information can be found on our website.
Getting involved
Please report bugs or enhancement requests through the Issue Tracker.
Installing DeepDriveMD
To install the latest release, clone the code from the main branch and use pip to install the package.
pip
Installation with pip and a minimal set of dependencies:
git clone https://github.com/DeepDriveMD/DeepDriveMD-pipeline
cd deepdrivemd
pip install -e .