[algorithm] section

The name determines the type of algorithm. Each parameter is defined for each algorithm.

• name

  Format: String

  Description: Algorithm name. The following algorithms are available.

  • minsearch : Minimum value search using Nelder-Mead method

  • global_search : Global optimization using the scipy.optimize routines (differential evolution, shgo, direct, dual annealing)

  • mapper : Grid search

  • random_search : Random search

  • bayes : Bayesian optimization

  • ttopt : Optimization based on the tensor-train decomposition

  • exchange : Replica Exchange Monte Carlo method

  • pamc : Population Annealing Monte Carlo method

• label_list

  Format: List of strings. The length should match the value of dimension. (default: ["x1", "x2", ...])

  Description: Names of the parameter axes, used e.g. in the headers of the output files.

• seed

  Format: Integer

  Description: A parameter to specify the seed of the pseudo-random number generator used for random generation of initial values, Monte Carlo updates, etc. For each algorithm process, the value of seed + algrank * seed_delta is used as the seed, where algrank is the process rank in the algorithm layer (independent of the solver parallelism nsolve). If omitted, the initialization is done by NumPy’s prescribed method.

• seed_delta

  Format: Integer (default: 314159)

  Description: A parameter to calculate the seed of the pseudo-random number generator for each MPI process. For details, see the description of seed.

• checkpoint

  Format: Boolean (default: false)

  Description: A parameter to specify whether the intermediate states are periodically stored to files. The final state is also saved. If the execution is terminated, it can be resumed from the latest checkpoint.

• checkpoint_steps

  Format: Integer (default: 16,777,216)

  Description: A parameter to specify the iteration steps between the previous and next checkpoints. One iteration step corresponds to one evaluation of a grid point in the mapper algorithm, one evaluation of Bayesian search in the bayes algorithm, and one local update in the Monte Carlo (exchange and PAMC) algorithms. The default value is a sufficiently large number of steps. To use checkpointing, set checkpoint = true and specify at least one of checkpoint_steps and checkpoint_interval.

• checkpoint_interval

  Format: Floating point number (default: 31,104,000)

  Description: A parameter to specify the execution time between the previous and next checkpoints in unit of seconds. The default value is a sufficiently long period (360 days). To use checkpointing, set checkpoint = true and specify at least one of checkpoint_steps and checkpoint_interval.

• checkpoint_file

  Format: String (default: "status.pickle")

  Description: A parameter to specify the name of the output file to which the intermediate state is written. The files are generated in the output directory of each process. The past three generations are kept with the suffixes .1, .2, and .3.

See Search algorithms for details of the various algorithms and their input/output files.