Hyperopt

Ver. 1.1.1 (2022-10-17)

This module provides a wrapper class for hyperparameter tuning by reusing the Hyperopt framework.

See also: https://pypi.org/project/hyperopt/

class mlpro.wrappers.hyperopt.WrHPTHyperopt(p_logging=True, p_algo='RND', p_ids=None)

Bases: HyperParamTuner, Wrapper, ScientificObject

This class is a ready to use wrapper class for Hyperopt framework. Objects of this type can be treated as a hyperparameter tuner object.

Parameters:

  • p_logging – Log level (see constants for log levels)

  • p_algo (str, optional) – Selection of a hyperparameter tuning algorithm, default: C_ALGO_RAND

  • p_ids (list of str, optional) – List of hyperparameter ids to be tuned, otherwise all hyperparameters, default: None

C_NAME

Name of the class.

Type:

str

C_ALGO_TPE

Refer to Tree of Parzen Estimators (TPE) algorithm.

Type:

str

C_ALGO_RAND

Refer to Random Grid Search algorithm.

Type:

str

C_NAME = 'Hyperopt'
C_WRAPPED_PACKAGE = 'hyperopt'
C_ALGO_TPE = 'TPE'
C_ALGO_RAND = 'RND'
C_SCIREF_TYPE = 'Proceedings'
C_SCIREF_AUTHOR = 'James Bergstra, Dan Yamins, David D. Cox'
C_SCIREF_TITLE = 'Hyperopt: A Python Library for Otimizing the Hyperparameters of Machine Learning Algorithms'
C_SCIREF_CONFERENCE = 'Proceedings of the 12th Python in Science Conference'
C_SCIREF_YEAR = '2013'
C_SCIREF_PAGES = '13-19'
C_SCIREF_DOI = '10.25080/Majora-8b375195-003'
C_SCIREF_EDITOR = 'Stefan van der Walt, Jarrod Millman, Katy Huff'
C_LOG_SEPARATOR = '------------------------------------------------------------------------------'
_maximize() → TrainingResults

This method is a place to setup a hp tuner based on hp structure of the model and run the hp tuner.

Returns:

best_result – The best result after a number of evaluations.

Return type:

float

_ofct_hyperopt(p_params)

This method is a place to run the evaluations by getting next set of hps from the tuner, inducting hps to the model, and running the the objective function.

Returns:

result – The result of an evaluations.

Return type:

float

setup_spaces()

This method is used to setup the hyperparameter spaces, including the tuning boundaries and a suitable discrete value. The hyperparameter should be bounded both above and below. We are using the “quantized” continuous distributions for natural and integer numbers. Meanwhile the real numbers are not quantized. For different parameter expressions, please redefined this method and check http://hyperopt.github.io/hyperopt/getting-started/search_spaces/! For big data handling, please redifined this method!

Returns:

spaces – List of parameter expressions.

Return type:

list

Cross References