cma.sigma_adaptation.CMAAdaptSigmaDistanceProportional

class documentation

class CMAAdaptSigmaDistanceProportional(CMAAdaptSigmaBase):

Constructor: CMAAdaptSigmaDistanceProportional(coefficient, direct_mode, **kwargs)

artificial setting of sigma proportional to ||m||,

specifically sigma = coefficient * mueff * norm(mean) / dimension / c_m or sigma = coefficient * norm(mean) in direct mode.

>>> import cma
>>> cma.evolution_strategy._redistribute_sigma_above = False
>>> es = cma.CMAEvolutionStrategy(4 * [1/2], 2, {'verbose': -9,
...         'AdaptSigma': cma.sigma_adaptation.CMAAdaptSigmaDistanceProportional(
...                           1.2, True)})
>>> # now we can call es.optimize(...)
>>> assert isinstance(es.adapt_sigma, cma.sigma_adaptation.CMAAdaptSigmaDistanceProportional
...                   ), es.adapt_sigma
>>> assert es.sigma == 1.2, (es.adapt_sigma.__dict__, es.sigma)

The optimal coefficient in infinite dimension is 1.253 = (pi/2)**0.5, the optimal mueff is lambda / pi, hence the optimal phi is pi/2 x lambda / pi / 2 = lambda / 4 where exp(-phi/n) is the (log-)expected convergence rate per iteration.

This is mainly useful for test purposes, e.g. to simulate optimal progress rates.

Details: setting cma.options_parameters.CMAOptions._stationary_sphere to True has on scaling invariant functions the "same effect". Instead of changing sigma, it resets norm(mean) at the end of tell.

Method	`__init__`	pass coefficient multiplier for normalized step-size
Method	`initialize`	same as update, set the correct step-size before the first iteration
Method	`update`	update `es.sigma` by calling `update2`.
Method	`update2`	return sigma update factor.
Instance Variable	`coefficient`	Undocumented
Instance Variable	`direct_mode`	experimental: when True, interpret coefficient as sigma / norm(mean)
Instance Variable	`is_initialized`	Undocumented

Inherited from CMAAdaptSigmaBase:

Method	`check_consistency`	make consistency checks with a `CMAEvolutionStrategy` instance as input
Method	`hsig`	return "OK-signal" for rank-one update.
Method	`initialize_base`	set parameters and state variable based on dimension, mueff and possibly further options.
Instance Variable	`cs`	Undocumented
Instance Variable	`delta`	cumulated effect of adaptation
Instance Variable	`is_initialized_base`	Undocumented
Instance Variable	`ps`	Undocumented
Method	`_update_ps`	update the isotropic evolution path.
Instance Variable	`_ps_updated_iteration`	Undocumented

def __init__(self, coefficient=1.2, direct_mode=False, **kwargs): ¶

overrides cma.sigma_adaptation.CMAAdaptSigmaBase.__init__

pass coefficient multiplier for normalized step-size

def initialize(self, es, *args, **kwargs): ¶

same as update, set the correct step-size before the first iteration

def update(self, es, **kwargs): ¶

overrides cma.sigma_adaptation.CMAAdaptSigmaBase.update

update es.sigma by calling update2.

def update2(self, es, **kwargs): ¶

overrides cma.sigma_adaptation.CMAAdaptSigmaBase.update2

return sigma update factor.

Uses attributes .N, .sp.weights.mueff, .mean, and .sp.cmean of input es.

coefficient = ¶

Undocumented

direct_mode = ¶

experimental: when True, interpret coefficient as sigma / norm(mean)

is_initialized: bool = ¶

Undocumented