Datasets:
metadata
language:
- en
pretty_name: Matbench
tags:
- multi-task
- datasets
- pytorch
- tar
configs:
- config_name: default
data_files:
- split: matbench_dielectric
path: matbench_dielectric.tar.gz
- split: matbench_phonons
path: matbench_phonons.tar.gz
- split: matbench_perovskites
path: matbench_perovskites.tar.gz
- split: matbench_jdft2d
path: matbench_jdft2d.tar.gz
- split: matbench_log_gvrh
path: matbench_log_gvrh.tar.gz
- split: matbench_mp_e_form
path: matbench_mp_e_form.tar.gz
- split: matbench_log_kvrh
path: matbench_log_kvrh.tar.gz
- split: matbench_mp_is_metal
path: matbench_mp_is_metal.tar.gz
- split: matbench_mp_gap
path: matbench_mp_gap.tar.gz
matbench
This repository contains copy of the matbench dataset, a multi-task dataset for materials property prediction.
NOTE: not all tasks are uploaded here. This is an unfinished repo.
Available tasks (splits)
matbench_dielectricmatbench_phononsmatbench_perovskitesmatbench_jdft2dmatbench_log_gvrhmatbench_mp_e_formmatbench_log_kvrhmatbench_mp_is_metalmatbench_mp_gap
Each <task>.tar.gz archive corresponds to a split named <task> and contains the original files
(e.g., .pt tensors and a summary.json) from the corresponding task directory.
the summary.json contains fold/split indexing.