json-config-expander

Expand multi optional configuration to multiple configurations.

Example 1

base_config = {'param_1*': [12, 13]}
expand_configs(base_config)

Returns:

[{'param_1': 12}, {'param_1': 13})

Example 2

base_config = {'param_1': {'param_2*': [12, 13]}}
expand_configs(base_config)

Returns:

[
    {'param_1': {'param_2': 12}}, 
    {'param_1': {'param_2': 13}}
]

Example 3

base_config = {'param_1*': [12, 13], 'param_2*': ['a', 'b']}
expand_configs(base_config)

Returns:

[
    {'param_1': 12, 'param_2': 'a'}, 
    {'param_1': 12, 'param_2': 'b'}, 
    {'param_1': 13, 'param_2': 'a'}, 
    {'param_1': 13, 'param_2': 'b'}
]

Example 4

base_config = {
    'param_1*': [
        {'param_2*': [20, 30, 50]},
        {'param_3*': ['Big', 'Small']}
    ]
}
expand_configs(base_config)

Returns:

[
    {'param_1': {'param_2': 20}}, 
    {'param_1': {'param_2': 30}},
    {'param_1': {'param_2': 50}},  
    {'param_1': {'param_3': 'Big'}},
    {'param_1': {'param_3': 'Small'}}
]

Motivation Scenario

You would like to run a classification task on multiple parameters of multiple classifier types, and see which one performs better:

base_config = {
    'classifier*': [
        {'name': 'logistic_regression', 'max_iter*': [100, 200, 300]},
        {'name': 'xgboost', 'n_estimators*': [50, 100, 200], 'max_depth*': [3,4,5]}
    ]
}

To returns all the possible configurations of your setting:

expand_configs(base_config)

Returns:

[
    {'classifier': {'name': 'logistic_regression', 'max_iter': 100}}, 
    {'classifier': {'name': 'logistic_regression', 'max_iter': 200}}, 
    {'classifier': {'name': 'logistic_regression', 'max_iter': 300}}, 
    {'classifier': {'name': 'xgboost', 'n_estimators': 50, 'max_depth': 3}}, 
    {'classifier': {'name': 'xgboost', 'n_estimators': 50, 'max_depth': 4}}, 
    {'classifier': {'name': 'xgboost', 'n_estimators': 50, 'max_depth': 5}}, 
    {'classifier': {'name': 'xgboost', 'n_estimators': 100, 'max_depth': 3}}, 
    {'classifier': {'name': 'xgboost', 'n_estimators': 100, 'max_depth': 4}}, 
    {'classifier': {'name': 'xgboost', 'n_estimators': 100, 'max_depth': 5}}, 
    {'classifier': {'name': 'xgboost', 'n_estimators': 200, 'max_depth': 3}}, 
    {'classifier': {'name': 'xgboost', 'n_estimators': 200, 'max_depth': 4}}, 
    {'classifier': {'name': 'xgboost', 'n_estimators': 200, 'max_depth': 5}}
]

If you want to run evaluation on each configuration, you need to pass evaluation_function:

def evaluation_function(config):
    ...

results = expand_configs(base_config, evaluation_function)

The results list would have all the evaluation results on each config, then you can select the best result for your needs.

License

This project is licensed under the MIT License - see the LICENSE.md file for details