The arms submodule

Contents:

class multibeep.arms.base

Bases: object

get_ident(self)

access to the possibly non-unique name of an arm

Returns:string – name of the arm
posterior(self)

access to the arm’s posterior

Returns:posterior_class – a (valid) posterior
provides_posterior(self)

to query if an arm provides a posterior

Returns:bool – True if the arm has a posterior, False if not
pull(self)

pulls the arm

Returns:float – recieved reward
real_mean(self)

the mean of the underlying distribution

Returns:float – mean of the underlying distribution, NaN if N/A
real_variance(self)

the variance of the underlying distribution

Returns:float – variance of the underlying distribution, NaN if N/A
class multibeep.arms.bernoulli(float_t p, rng_class rng)

Bases: multibeep.arms.base

The classic Bernoulli arm.

Provides the ‘default’ posterior using a Beta prior.

Parameters:
  • p (float) – the p parameter of the Bernoulli distribution
  • rng (multibeep.util.rng_class) – a random number generator object
class multibeep.arms.data(ndarray data, name, rng_class rng, bootstrap=False)

Bases: multibeep.arms.base

Parameters:
  • data (numpy.ndarray (1d)) – The data for this arme. The data is copied at least once, but if the data is not in C order, a second temporary copy is made.
  • name (string) – the name associated with this arm
  • rng (multibeep.util.rng_class) – a random number generator object
  • bootstrap (bool) –

    If false, the rewards are returned in sequential order starting with the first when the end is reached.

    If true, an entry is chosen uniformly at random (with replacement)

    Default is False.
class multibeep.arms.exponential(float_t l, rng_class rng)

Bases: multibeep.arms.base

An arm with the exponential reward distribution.

Provides a posterior using an inverse-Gamma prior.

Parameters:
  • l (float) – the lambda parameter of the exponential distribution
  • rng (multibeep.util.rng_class) – a random number generator object
class multibeep.arms.normal(float_t mean, float_t variance, rng_class rng)

Bases: multibeep.arms.base

An arm with a normal reward distribution

Provides a Bayesian posterior where the mean and variance are unknown!

Parameters:
  • mean (float) – the mean of the normal distribution
  • variance (float) – the variance of the normal distribution
  • rng (multibeep.util.rng_class) – a random number generator object
class multibeep.arms.python(obj, name='custom python arm')

Bases: multibeep.arms.base

deactivate(self)
posterior(self)
pull(self)
real_mean(self)
real_variance(self)