遇见数据集

DFT_DatasetConstructionResults_BuildingInformativeMaterialsDatsetsBeyondTargetedObjectives

收藏
Zenodo2026-05-21 更新2026-05-26 收录
官方服务:

资源简介:

This repository contains CSV files from the active-learning dataset construction experiments performed for the paper “Building Informative Materials Datasets Beyond Targeted Objectives.” The files report model performance across acquisition iterations for different DFT candidate pools, construction strategies, target properties, models, and random seeds. DFT Candidate Pools The DFT experiments use four candidate pools: jarvis18: JARVIS 2018 jarvis22: JARVIS 2022 mp18: Materials Project 2018 mp21: Materials Project 2021 Target Properties The DFT target properties are: bandgap: electronic band gap bulk_modulus: bulk modulus e_form or eform: formation energy The two-target combinations are: bandgap_bulkmodulus bandgap_eform eform_bulk_modulus File Naming Conventions File names encode the dataset, target or target pair, acquisition strategy, model, and seed. random: random sampling QBC: query-by-committee QBCdiversity: QBC plus feature diversity MultiObjective_OptDiffModelsOnly2Var: QBC with two targets growing_EA_2obj_qbc_only: NSGA-II QBC with two targets, without feature diversity growing_EA_2obj_qbc_Diversity: NSGA-II QBC with two targets plus feature diversity rf: Random Forest xgb: XGBoost Example file names: {dataset}_all_random_rf{seed}.csv {dataset}_all_random_xgb{seed}.CSV {dataset}_{target}_QBC_rf.csv {dataset}_{target}_QBCdiversity_rf{seed}.csv {dataset}_{target_pair}MultiObjective_OptDiffModelsOnly2Var_rf.csv {dataset}_{target_pair}growing_EA_2obj_qbc_only_rf{seed}.csv {dataset}_{target_pair}growing_EA_2obj_qbc_Diversity_rf{seed}.csv File Content Each CSV file reports the prediction performance obtained during dataset construction. Results are provided for Random Forest and XGBoost models across different acquisition strategies and random initializations. These files allow comparison between random sampling, target-focused active learning, and diversity-aware active learning for both targeted and untargeted DFT properties.

提供机构:
Zenodo
创建时间:
2026-05-21
二维码
社区交流群
二维码
科研交流群
商业服务