The aim is to discover high-figure–of-merit (
To date, there has been no statistically robust approach to simultaneously incorporate experimental and model error into machine learning models in a search space with high opportunity cost and high latency (i.e. large time between prediction and validation).
Consequently, searches have been unable to effectively guide experimentalists in the selection of exploring or exploiting new materials when the validation step is inherently low throughput and resource-intensive, as is the case for synthesizing new bulk functional materials like thermoelectrics. This project aims to implement a holistic pipeline to discover novel thermoelectrics: ML models predict the
To check out the code in this repo, reproduce results and start contributing to the project, clone the repo and create a conda
environment containing all dependencies by running the following command (assumes you have git
and conda
installed)
git clone https://github.com/janosh/thermo \
&& cd thermo \
&& pip install -r requirements.txt
&& pre-commit install
Run any of the files in src/notebooks
. The recommended way to work with those files is using VS Code and its Python extension. You'll see the results of running those files in an interactive window (similar to Jupyter).