Inversion framework of fault parameter using JAGURS version 3.4

Inversion framework of fault parameter using JAGURS version 3.4 Overview This repository provides a refactored and extensively documented version of the fault parameter inversion framework originally developed for tsunami source inversion using JAGURS simulations.The refactored implementation introduces: New Features Selectable acquisition mode Single-model Bayesian optimization (LogEI) or Ensemble acquisition using weighted mixtures of GP models Random-start capability Automatically generates and updates the CSV during optimization Cleaner architecture & documentation All user-editable parameters consolidated in a Config dataclass Modular structure for sampling, scoring, GP building, PI density estimation Improved safety Controlled directory operations Clearer lifecycle for generated case folders PI-based pseudo-density estimation for post-run visualization This code is intended for research in fault mechanism inversion, tsunami simulation workflows, and general Bayesian optimization in geophysical applications. Key Dependencies Python ≥ 3.9 PyTorch BoTorch GPyTorch NumPy / Pandas / SciPy netCDF4 JAGURS (external tsunami simulation code) 高性能津波計算コードの開発 | index How to Run Prepare a template directory (e.g., 1911Kikai-00/) including tsun.par and other necessary JAGURS files. Create criteria_RH.csv of an observation data including point as observed inundation elevation and above/below flag of inundation or non-inundation. Place criteria_RH.csv under root_dir. Adjust the configuration block in the script. Run: python inversion_JAGURS_fm_v3.4.py Input / Output Files The script automatically generates: File Description input_ .csv History of sampled parameter vectors score_ .csv Parameter vectors with associated scores result_ .csv Cached sampling results for each JAGURS output pi_density_ .csv PI-based pseudo-density over parameter domain Citation If you use this software in published research, please cite: Nakanishi, R. (2026). Inversion of fault mechanism using JAGURS version 3.4. Zenodo. https://doi.org/10.5281/zenodo.18521698 Acknowledgements This code builds upon prior tsunami simulation and inversion frameworks developed in the JAGURS project and related research activities. We deeply appreciate all contributors and maintainers of the open-source scientific Python ecosystem.