A framework for fully autonomous design of materials via multiobjective optimization and active learning: challenges and next steps

This repository contains several scripts for optimizing the production of the electrolyte 2,2,2-trifluoroethyl Methyl carbonate (TFMC) on a continuous-flow reactor using the ParMOO solver with the MDML to distribute experiment requests. We have also included experimental data and a script for relplaying this data and plotting it, using our ParMOO-MDML extension.

Setup and Running

The requirements for this directory are:

• parmoo (v 0.1.0),
• mdml-client, and
• matplotlib.

To try running these solvers your self, clone this directory and install the requirements, or use the included REQUIREMENTS.txt file.

python3 -m pip install -r REQUIREMENTS.txt

To replay the experiment and generate the results graph from Figure 1 in the paper, run the following.

cd parmoo-mdml-experiments && python3 cfr-tfmc-solver.py

To fully recreate our experiments, one would need to recreate the automated CFR/NMR setup described in Appendix B of the paper, create a valid MDML Host, then uncomment the solve command in our script.

Creating an MDML Host

An MDML host and its services can be quickly spun up using docker compose. Once the host is running, any mdml client created through the mdml_client python package will be able to connect and start streaming data. The compose files and directions for setup are contained within the mdml-minimal repo. Assuming docker-compose is installed on the host machine, MDML hosts only require a few environment variables be specified before starting.

Directory Structure

The base directory contains the README.rst and REQUIREMENTS.txt files.

The subdirectory parmoo-mdml-experiments contains:

• __init__.py (package setup);
• parmoo_mdml_extension.py (ParMOO ext using MDML to send experiments);
• results (directory containing data from our 41 experiment run);
• cfr-tfmc-solver.py (script for replaying the experiment in results and plotting); and
• tfmc-manufacture-config.json (config file for our TFMC-making solver).

Citing this work

To cite this work, use the following:

@inproceedings{iclr23-cfr-materials,
    author  = {Chang, Tyler H. and Elias, Jakob R. and Wild, Stefan M. and Chaudhuri, Santanu and Libera, Joseph A.},
    title   = {A framework for fully autonomous design of materials via multiobjective optimization and active learning: challenges and next steps},
    year    = {2023},
    booktitle = {ICLR 2023, Workshop on Machine Learning for Materials (ML4Materials)},
    url     = {https://openreview.net/forum?id=8KJS7RPjMqG}
}