A new tool for solving complex science problems
The Open Data Analysis Tool for Science and Engineering helps solve inverse problems using repeated simulations.
Researchers developed version 4.0 of the Open Data Analysis Tool for Science and Engineering (ODAT-SE). This open-source Python framework solves inverse problems through repeated forward simulations.
The framework connects optimization and sampling algorithms to solvers. It uses a common interface for parameter mapping, constraints, and objective evaluation. This allows any existing simulation code to be used with any algorithm.
ODAT-SE provides two-level MPI parallelism. This distributes candidate evaluations over processes and each evaluation over a parallel solver. It also includes checkpoint and continuation facilities.
The tool includes a Bayesian workflow. This workflow estimates partition-function ratios, model evidence, effective noise levels, and marginal posterior distributions. It also features tensor-train optimization and global optimizers from SciPy.
The framework was tested using positron-diffraction. The workflow identified correlated structural solutions and an effective noise scale. It also exposed incomplete convergence of symmetry-related mode weights.
Key facts
- ODAT-SE is an open-source Python framework for inverse problems.
- The framework provides two-level MPI parallelism.
- Version 4.0 includes a Bayesian workflow and tensor-train optimization.