cma.optimization_tools.BinaryEvolutionPath

class documentation

class BinaryEvolutionPath(EvolutionPath):

Constructor: BinaryEvolutionPath(time_constant)

View In Hierarchy

Undocumented

Method binary_s how many increments for one decrement in stationary state
Property probability_larger_than_one_from_binary propability of path entries to be larger than one,
Property raw_binary_s return one of two possible values with expectation of zero.

Inherited from EvolutionPath:

Method __init__ Undocumented
Property path Undocumented

Inherited from ExponentialSmoothing (via EvolutionPath):

Method __call__ Undocumented
Method __getitem__ Undocumented
Instance Variable count Undocumented
Instance Variable normalizer Undocumented
Instance Variable time_constant Undocumented
Instance Variable values Undocumented
Method _init_ Undocumented
def binary_s(self, odds_of_increment=1):

how many increments for one decrement in stationary state

@property
probability_larger_than_one_from_binary =

propability of path entries to be larger than one,

given the input is sign(randn()). Check out:

n = int(1e4)
greater_than_one = []
ar_tc = [1.2, 1.5, 1.9, 2, 4, 8, 16, 32, 100]
for tc in ar_tc:
    p = cma.optimization_tools.EvolutionPath(tc)
    for i in range(int(10 * tc)):
        p(np.sign(np.random.randn(n)))
    # plot(*step_data(p.path))
    greater_than_one += [(np.mean(p.path > 1) + np.mean(p.path < -1)) / 2]
@property
raw_binary_s =

return one of two possible values with expectation of zero.

the maximum for the larger value is 1 - 0.15865525393145707 for tc to infty.