Exploration

class optimas.explorations.Exploration(generator, evaluator, max_evals=inf, sim_workers=1, run_async=False, history=None, history_save_period=None, exploration_dir_path='./exploration', resume=False, libe_comms='local')

Class for launching an exploration.

Depending on the generator, the exploration can be an optimization, a parameter scan, etc.

Parameters:

generatorGenerator

The generator used to suggest new Trials.

evaluatorEvaluator

The evaluator that will execute the Trials.

max_evalsint, optional

Maximum number of trials that will be evaluated in the exploration. If not given, the exploration can run indefinitely.

sim_workersint, optional

Number of parallel workers performing simulations. By default, 1.

run_asyncbool, optional

Whether the evaluations should be performed asynchronously (i.e., without waiting for all workers to finish before staring a new evaluation). This is useful when the completion time of the evaluations is not uniform. By default, False.

historystr, optional

Path to a history file of a past exploration from which to restart the new one. By default, None.

history_save_periodint, optional

Periodicity, in number of evaluated Trials, with which to save the history file to disk. By default equals to sim_workers.

exploration_dir_pathstr, optional.

Path to the exploration directory. By default, './exploration'.

resumebool, optional

Whether the exploration should resume from a previous run in the same exploration_dir_path. If True, the exploration will continue from the last evaluation of the previous run until the total number of evaluations (including those of the previous run) reaches max_evals. There is no need to provide the history path (it will be ignored). If False (default value), the exploration will raise an error if the exploration_dir_path already exists.

libe_comms{‘local’, ‘threads’, ‘mpi’}, optional.

The communication mode for libEnseble. Determines whether to use Python multiprocessing (local), threading (threads) or MPI for the communication between the manager and workers. The 'threads' mode is only recommended when running in a Jupyter notebook if the default ‘local’ mode has issues (this can happen especially on Windows and Mac, which use multiprocessing spawn). 'threads' only supports FunctionEvaluator``s. If running in ``'mpi' mode, the Optimas script should be launched with mpirun or equivalent, for example, mpirun -np N python myscript.py. This will launch one manager and N-1 simulation workers. In this case, the sim_workers parameter is ignored. By default, 'local' mode is used.

Methods

attach_evaluations(evaluation_data[, ...])	Attach evaluations from external source.
attach_trials(trial_data[, ...])	Attach trials for future evaluation.
evaluate_trials(trial_data[, ...])	Attach and evaluate trials.
mark_evaluation_as_failed(trial_index)	Mark an already evaluated trial as failed.
run([n_evals])	Run the exploration.

Attributes

history	Get the exploration history.
is_manager	Get whether the current process is the manager.