Original experimental data for the manuscript “Machine learning-based source term joint probability region inversion for uncertainty quantification in plutonium explosion accidents”
Description
This dataset contains the original experimental data supporting the manuscript “Machine learning-based source term joint probability region inversion for uncertainty quantification in plutonium explosion accidents”. It includes the original JPSTI simulation dataset, held-out test-set results, joint Monte Carlo sample arrays, Optuna hyperparameter-optimization records, cross-validation records, and source data for manuscript Figures 2-5. The package also includes file-level metadata, checksums, a data dictionary, a dataset inventory, an NPZ array index, and English/Chinese file-structure guides for reviewer verification. The dataset supports the final comparison between JPSTI and three retained direct joint baselines: MultiTargetXGBoostLSS_MVN, MultiTargetXGBoostLSS_MVT, and MultivariateNGBoost_MVN. The JPSTI independent joint baseline, trained model artifacts, and source code are not included in this review data package.
Files
Steps to reproduce
Use read_me.md as the entry point. The original simulation data are in 01_raw_dataset/, where parts 1-5 support training, cross-validation, model selection, and final retraining, and part 6 is the held-out final test set. Final test-set prediction diagnostics, joint statistics, coverage/area curves, JSON result summaries, and compressed NumPy NPZ joint-sample arrays are in 02_final_test_results/. Cross-validation and hyperparameter-optimization records are in 03_cv_and_hyperparameter_records/. Source data for manuscript Figures 2-5 are in 04_figure_source_data/. File checksums, field definitions, dataset inventory, NPZ array index, and English/Chinese file-structure guides are in 05_metadata_and_manifests/. Large Monte Carlo joint samples are stored as NPZ files and can be read with numpy.load. In each joint_samples array, the dimension order is test_sample x Monte_Carlo_draw x target, with target index 0 for airborne plutonium source term_kg and target index 1 for DebrisCloudTop_m.
Institutions
- PLA Rocket Force University of EngineeringShaanxi, Xi'an