Tutorials#

In these tutorials, how to perform inverse problem analyses using ODAT-SE is explained by examples taken from minimization of analytical functions. The following algorithms are covered in these tutorials. global_search (global optimization with scipy.optimize) is also available; see Search algorithms for the complete list of the algorithms.

  • minsearch

    Nelder-Mead method.

  • mapper

    Exhaustive search over a grid of the given parameters.

  • random_search

    Random search.

  • bayes

    Bayesian optimization.

  • ttopt

    Tensor train optimization.

  • exchange

    Sampling by the replica exchange Monte Carlo method.

  • pamc

    Sampling by the population annealing Monte Carlo method.

In the following sections, the procedures to run these algorithms are provided. In addition, the usage of [runner.limitation] to apply limitations to the search region will be described. At the end of the section, a description of how to implement a direct problem solver is provided, as well as tutorials describing some applications.

Preparation#

To run the tutorials, the sample files are required in addition to the installation of ODAT-SE (see Installation of ODAT-SE). The sample files are included in the source package. Clone the repository and move into the obtained directory as follows:

$ git clone https://github.com/issp-center-dev/ODAT-SE.git
$ cd ODAT-SE

The steps in each tutorial are assumed to start from this ODAT-SE directory.

Note

To run the tutorials that use MPI parallelization (mapper, exchange, pamc, and so on), mpi4py (python3 -m pip install 'ODAT-SE[all]') is required in addition to a working MPI implementation providing the mpiexec / mpirun command.

On this page