Grid-Search

The basic type of optimization strategies is a defined Grid-Search. For every hyperparameter PHOTONAI creates a grid over one-dimensional intervals. The setting of these intervals is up to the user. Grid-Search then tests every point of the grid. PHOTONAI distinguishes between full Grid-Search and Random Grid-Search with stopping criteria after time or n_configurations.

Check out the implementation and add it simply to your Hyperpipe:

       
my_pipe = Hyperpipe('GS_pipeline',
                    optimizer='random_grid_search',
                    optimizer_params={'n_configurations': 25},
                    ...
                    )

       

GridSearchOptimizer

Set: optimizer = 'grid_search'. Without any parameters.

RandomGridSearchOptimizer

Set: optimizer = 'random_grid_search'

Parameter Type Description
n_configurations int Number of configurations to test in every outer_fold. Default is set to 25.

RandomGridSearchOptimizer

Set: optimizer = 'timeboxed_random_grid_search'

Parameter Type Description
n_configurations int Number of configurations to test in every outer_fold. Default is set to 25.
limit_in_minutes number Time in minutes to test configurations each outer_fold. Default is set to 60.