new_benchmark

[irlyngaas]

Matbench Pull Request Template

This is a new Benchmark submission

Benchmark submissions

This benchmark submission includes two subtasks

Brief description of your algorithm

Hydragnn is a pre-trained graph neural network model. These models are finetuned based using Matbench subtasks jdft_2d and mp_is_metal.

Included files

I do not include a notebook.ipynb because the commands to run the finetuning and/or evalutation are within bash scripts since the code does not run via the Matbench repository, but rather uses our own finetuning libraries. Installation instructions are provided and the finetuning can ran via running ./run_evaluate.sh if you want to use our finetuned models or ./run_full.sh if you want to go through the full finetuning training yourself. There are several .sh and .py included in the benchmark directory that facilitates running the workflow.

Example

-- benchmarks
---- matbench_v0.1_hydragnn
------ results.json.gz
------ info.json
------ run_full.sh 
------ run_evaluate.sh
------ preprocess_matbench.sh
------ preprocess_matbench.py
------ evaluate_our_finetune.sh
------ evaluate_your_finetune.sh
------ evaluate.py
------ finetune.sh
------ finetune.py
------ compile.sh
------ compile.py