pyMAISE.init

pyMAISE.init(problem_type, **kwargs)[source]

Initialize pyMAISE global settings.

Parameters:

  • problem_type (pyMAISE.ProblemType, {'regression', 'classification'}, or {0, 1}) – Defines a regression or classification problem.

  • verbosity (int, default=0) – Level of output.

  • random_state (int or None, default=None) – Controls the randomness of all processes in pyMAISE.

  • num_configs_saved (int, default=5) – Number of top hyperparameter configurations saved for each type of model.

  • new_nn_architecture (bool, default=True) – Controls the hyperparameter tuning architecture used for tuning neural network models.

  • cuda_visible_devices (str or None, default=None) – Devices visible to tensorflow. Sets the CUDA_VISIBLE_DEVICES environment variable.

  • run_parallel (bool, default=False) – Controls NN hyperparameter tuning parallelization. If True then pyMAISE launches process within all available GPUs (depends on tuning strategy). By default pyMAISE attempts to approximate the memory footprint of each model to ensure a given GPU is fully utilized. This may be unstable. To control the maximum number of models allowed on a given GPU set the max_models_per_device argument. Parallel is only supported for GPUs and assumes at least one model will fit into one GPU.

  • max_models_per_device (int, default=numpy.inf) – The maximum number of NN models allowed on a single GPU during tuning when running in parallel (run_parallel = True).

Returns:

values – The settings class with the provided parameters changed.

Return type:

pyMAISE.Settings