Global Options

  • software [str]:

    Specify what software package to use when rsopt runs. See Software for a list of supported software.

  • method [str]:

    When software package chosen contains multiple algorithms the choice is specified by method. See Software for particulars on methods supported by the chosen software.

  • software_options [dict]:

    Can be used to pass configuration options directly to the chosen software. See Software for available options that can be set depending on chosen software.

  • nworkers [int] (default: 2):

    Number of workers used for running simulations, and in some cases a worker may be used run the governing software. In cases where the software option supports parallel evaluations then workers will each, independently run through a Job chain. For some software options the supported number of worker is fixed (e.g. local optimizers can only use two workers. One to manage the optimizer and one to manage simulation execution.) See Software for information on support for parallel evaluation in a particular software option.

  • exit_criteria [dict]: Required when there is not a deterministic stopping point (sampler software). For all optimizers this must be set even if some stop criteria has been provided in software_options. libEnsemble will pre-allocate a history array based on the settings of exit_criteria so it must fit into memory on whatever resource rsopt is being executed on. Allowed fields (At least one must be given):

    ‘sim_max’ [int]:

    Stop when this many new points have been evaluated by sim_f

    ‘gen_max’ [int]:

    Stop when this many new points have been generated by gen_f

    ‘elapsed_wallclock_time’ [float]:

    Stop when this time (since the manager has been initialized) has elapsed

    ‘stop_val’ [(str, float)]:

    Stop when H[str] < float for the given (str, float pair)

  • objective_function [list]: Required to use an optimizer if the final code in the job chain is not a serial Python execution. Should contain the name of the Python module defining the objective function and the name of that function. For example objective_function: [functions.py, obj_f]. If the final code in the job chain is serial Python then rsopt will try to directly pass its output to the optimizer if no objective_function is given.

  • record_interval [int]: If given the optimizer or sampler progress (from the libEnsemble history array) will be dumped every record_interval evaluations. If not given or 0 then the history will only be dumped at the end of the run. libEnsemble will also attempt to dump the history in the event of premature termination due to a worker crash.

  • output_file [str]: The name of a file without the extension where the history array will be dumped. File written will be: output_file.npy.

  • run_dir [str]: If given then all code evaluations will use run_dir as the top level for executing evaluations. If not given the default path is ./ensemble. If a relative path is given the path will be relative to the rsopt execution directory. Additional directories may be created within run_dir depending on the setting of use_worker_dirs and sim_dirs_make.

  • sym_links [list]: A list of file names that will be symlinked into every simulation run directory.

  • sim_dirs_make [bool]: If true then every simulation will be run in a separate directory under run_dir. This is always set to True if any code is using parallel or rsmpi execution regardless of setting in the configuration file. Otherwise input files might clash.

  • use_worker_dirs [bool]: If true then sim directories are organized under a common worker directory. This is used by default if parallel execution is being used.

  • copy_final_logs [bool]: Defaults to True. If True then ensemble.log, libE_stats.txt, the configuration file, and the final history file will be copied to the top level of run_dir. This copying does not occur if the run terminates unexpectedly.

  • executor_options [dict]: Options given here are passed directly to the libEnsemble Executor.

    • hosts [list] (rsmpi only): A list of rsmpi host indices that workers can use. To see your available host indices run rsmpi on the command line. If not set rsmpi will be allowed to assign work to all rsmpi nodes available to you.

  • seed [int or None or str]: Sets the seed to initialize the pseudo-random number generator used by the sampler. Behavior depends on the setting:

    • '': default If an empty string is given, or seed is not explicitly included then a fixed seed is set.

      Seed is based on worker ID number. Repeated runs with this setting will always be repeatable.

    • None: If seed is set to :code: None then a random seed will be used. The run will not be repeatable. IMPORTANT: To set a field to be None-type in YAML the field must be empty. So the options block should look like:

      options:
 software: lh_scan
 seed:
 batch_size: 42

    • int: If set be any integer between 0 and 2**32 - 1 inclusive then the integer is used as the seed initialize the pseudo-random number generator.