odatse-aenet

odatse-aenet

AENET (Atom-centered Neural Network) solver module for ODAT-SE (Open Data Analysis Tool for Science and Engineering).

This package provides:

• AenetSolver --- ODAT-SE solver that drives AENET predict.x for parameter optimization via grid search, Nelder-Mead, Bayesian optimization, replica exchange, and other algorithms.

Requirements

• Python >= 3.9
• numpy >= 1.22
• ODAT-SE >= 3.0
• AENET (predict.x, generate.x, train.x)

Optional (for training tutorial):

• Quantum ESPRESSO (pw.x)

Installation

pip install odat-se[all]       # Install ODAT-SE
pip install .                  # Install this package

After installation, the odatse-aenet command becomes available:

odatse-aenet --version

See the documentation for detailed installation instructions including AENET and QE.

Quick Start

CLI

odatse-aenet input.toml

MPI parallel

mpiexec -np 4 odatse-aenet input.toml

Tutorials

Directory	Description	Requirements
sample/aenet_training/	AENET training pipeline: structure generation → QE relaxation → fingerprint → training → prediction	QE, AENET
sample/aenet_mapper/	N2 dimer grid search using AENET + mapper	AENET, ODAT-SE
sample/aenet_minsearch/	N2 dimer optimization using AENET + Nelder-Mead	AENET, ODAT-SE

Each tutorial includes a run_all.sh script to execute the full workflow.

Project Structure

src/
  AenetSolver/               # ODAT-SE solver for AENET
    aenet.py                 # Solver class (extends odatse.solver.SolverBase)
    _main.py                 # CLI entry point
sample/                      # Tutorial samples
docs/
  ja/                        # Japanese documentation
  en/                        # English documentation
tests/                       # Unit tests (52 tests)

Documentation

Full documentation is available online:

• English
• 日本語

To build locally:

cd docs
sh make.sh       # Build both ja/en HTML docs

Developers

• 谷田秀哉 (Shuya Tanida), Tottori University --- Initial code development
• 星健夫 (Takeo Hoshi), National Institute for Fusion Science --- ODAT-SE co-development, research supervision
• 吉見一慶 (Kazuyoshi Yoshimi), ISSP, The University of Tokyo --- Code organization and packaging

Citation

If you use ODAT-SE in your research, please cite:

Y. Motoyama, K. Yoshimi, I. Mochizuki, H. Iwamoto, H. Ichinose, and T. Hoshi, "Data-analysis software framework 2DMAT and its application to experimental measurements for two-dimensional material structures", Computer Physics Communications 280, 108465 (2022).

License

MPL-2.0