v0.4.0

Updated benchmarks using MODNet v0.1.12, using genetic algorithm hyperparameter optimization for all tasks.

v0.3.1

Minor bug fixes to the way our workflow/scripts were disseminated, with no associated changes to the benchmark results.

v0.3.0

Final results for matbench submission, including results on the larger matbench_perovskites, matbench_mp_gap, matbench_mp_is_metal, matbench_mp_eform tasks.

v0.2.2

A fix to the published data that does not change the aggregate results. KFold and not StratifiedKFold was used for classification splits; as the datasets were balanced, this had a limited effect on the results.

v0.2.1

Minor repository reorganisation relative to v0.2. Added an MIT license.

v0.2

Results as reported in final paper at 10.1088/1361-648X/ac1280.

NB: the MODNet version used was v0.1.10, and NOT v0.1.9 as reported in the requirements.txt.

v0.1

Data as first reported in https://arxiv.org/abs/2102.02263/v1