`bolero.optimizer`.CMAESOptimizer¶

class bolero.optimizer.CMAESOptimizer(initial_params=None, variance=1.0, covariance=None, n_samples_per_update=None, active=False, bounds=None, maximize=True, min_variance=9.8607613152626476e-32, min_fitness_dist=4.4408920985006262e-16, max_condition=10000000.0, log_to_file=False, log_to_stdout=False, random_state=None)[source]¶

Covariance Matrix Adaptation Evolution Strategy.

See Wikipedia for details.

Plain CMA-ES [1] is considered to be useful for

non-convex,
non-separable,
ill-conditioned,
or noisy

objective functions. However, in some cases CMA-ES will be outperformed by other methods:

if the search space dimension is very small (e.g. less than 5), downhill simplex or surrogate-assisted methods will be better
easy functions (separable, nearly quadratic, etc.) will usually be solved faster by NEWUOA
multimodal objective functions require restart strategies

Parameters:

Parameters:	initial_params : array-like, shape = (n_params,), optional (default: 0s) Initial parameter vector. variance : float, optional (default: 1.0) Initial exploration variance. covariance : array-like, optional (default: None) Either a diagonal (with shape (n_params,)) or a full covariance matrix (with shape (n_params, n_params)). A full covariance can contain information about the correlation of variables. n_samples_per_update : integer, optional (default: 4+int(3log(n_params))) Number of roll-outs that are required for a parameter update. active* : bool, optional (default: False) Active CMA-ES (aCMA-ES) with negative weighted covariance matrix update bounds : array-like, shape (n_params, 2), optional (default: None) Upper and lower bounds for each parameter. maximize : boolean, optional (default: True) Maximize return or minimize cost? min_variance : float, optional (default: 2 * np.finfo(np.float).eps 2) Minimum variance before restart min_fitness_dist** : float, optional (default: 2 * np.finfo(np.float).eps) Minimum distance between fitness values before restart max_condition : float optional (default: 1e7) Maximum condition of covariance matrix log_to_file: boolean or string, optional (default: False) Log results to given file, it will be located in the $BL_LOG_PATH log_to_stdout: boolean, optional (default: False) Log to standard output random_state : int or RandomState, optional (default: None) Seed for the random number generator or RandomState object.

initial_params : array-like, shape = (n_params,), optional (default: 0s): Initial parameter vector.
variance : float, optional (default: 1.0): Initial exploration variance.
covariance : array-like, optional (default: None): Either a diagonal (with shape (n_params,)) or a full covariance matrix (with shape (n_params, n_params)). A full covariance can contain information about the correlation of variables.
n_samples_per_update : integer, optional (default: 4+int(3*log(n_params))): Number of roll-outs that are required for a parameter update.
active : bool, optional (default: False): Active CMA-ES (aCMA-ES) with negative weighted covariance matrix update
bounds : array-like, shape (n_params, 2), optional (default: None): Upper and lower bounds for each parameter.
maximize : boolean, optional (default: True): Maximize return or minimize cost?
min_variance : float, optional (default: 2 * np.finfo(np.float).eps ** 2): Minimum variance before restart
min_fitness_dist : float, optional (default: 2 * np.finfo(np.float).eps): Minimum distance between fitness values before restart
max_condition : float optional (default: 1e7): Maximum condition of covariance matrix
log_to_file: boolean or string, optional (default: False): Log results to given file, it will be located in the $BL_LOG_PATH
log_to_stdout: boolean, optional (default: False): Log to standard output
random_state : int or RandomState, optional (default: None): Seed for the random number generator or RandomState object.

References

[1]	(1, 2) Hansen, N.; Ostermeier, A. Completely Derandomized Self-Adaptation in Evolution Strategies. In: Evolutionary Computation, 9(2), pp. 159-195. https://www.lri.fr/~hansen/cmaartic.pdf

__init__(initial_params=None, variance=1.0, covariance=None, n_samples_per_update=None, active=False, bounds=None, maximize=True, min_variance=9.8607613152626476e-32, min_fitness_dist=4.4408920985006262e-16, max_condition=10000000.0, log_to_file=False, log_to_stdout=False, random_state=None)[source]¶

get_args()¶

Get parameters for this estimator.

Returns:	params : mapping of string to any Parameter names mapped to their values.

get_best_fitness()[source]¶

Get the best observed fitness.

Returns:	best_fitness : float Best fitness (sum of feedbacks) so far. Corresponds to the parameters obtained by get_best_parameters(method=’best’). For maximize=True, this is the highest observed fitness, and for maximize=False, this is the lowest observed fitness.

get_best_parameters(method='best')[source]¶

Get the best parameters.

Parameters:	method : string, optional (default: ‘best’) Either ‘best’ or ‘mean’
Returns:	best_params : array-like, shape (n_params,) Best parameters

get_next_parameters(params)[source]¶

Get next individual/parameter vector for evaluation.

Parameters:	params : array_like, shape (n_params,) Parameter vector, will be modified

init(n_params)[source]¶

Initialize the behavior search.

Parameters:	n_params : int dimension of the parameter vector

is_behavior_learning_done()[source]¶

Check if the optimization is finished.

Returns:	finished : bool Is the learning of a behavior finished?

set_evaluation_feedback(feedback)[source]¶

Set feedbacks for the parameter vector.

Parameters:	feedback : list of float feedbacks for each step or for the episode, depends on the problem

bolero.optimizer.CMAESOptimizer¶

`bolero.optimizer`.CMAESOptimizer¶