pomdp_py.representations.distribution package

pomdp_py.representations.distribution.histogram module

class pomdp_py.representations.distribution.histogram.Histogram

Bases: GenerativeDistribution

Histogram representation of a probability distribution.

__init__(self, histogram)

Parameters:

histogram (dict) – variable value to probability

__getitem__(self, value)

Returns the probability of value.

__setitem__(self, value, prob)

Sets probability of value to prob.

get_histogram(self)

Returns a dictionary from value to probability of the histogram

histogram
is_normalized(epsilon=1e-09)

Returns true if this distribution is normalized

mpe(self)

Returns the most likely value of the variable.

random(self)

Randomly sample a value based on the probability in the histogram

pomdp_py.representations.distribution.particles module

class pomdp_py.representations.distribution.particles.Particles

Particles is a set of unweighted particles; This set of particles represent a distribution \(\Pr(X)\). Each particle takes on a specific value of \(X\). Inherits WeightedParticles.

__init__(self, particles, **kwargs)

Parameters:

  • particles (list) – List of values.

  • kwargs – see __init__() of WeightedParticles.

add(self, particle)

particle: just a value

classmethod from_histogram(histogram, num_particles=1000)

Given a pomdp_py.Histogram return a particle representation of it, which is an approximation

get_abstraction(self, state_mapper)

feeds all particles through a state abstraction function. Or generally, it could be any function.

get_histogram(self)

Returns a mapping from value to probability, normalized.

particles

For unweighted particles, the particles are just values.

random()

Samples a value based on the particles

class pomdp_py.representations.distribution.particles.WeightedParticles

Represents a distribution \(\Pr(X)\) with weighted particles, each is a tuple (value, weight). “value” means a value for the random variable X. If multiple values are present for the same value, will interpret the probability at X=x as the average of those weights.

__init__(self, list particles, str approx_method=”none”, distance_func=None)

Parameters:

  • particles (list) – List of (value, weight) tuples. The weight represents the likelihood that the value is drawn from the underlying distribution.

  • approx_method (str) – ‘nearest’ if when querying the probability of a value, and there is no matching particle for it, return the probability of the value closest to it. Assuming values are comparable; “none” if no approximation, return 0.

  • distance_func – Used when approx_method is ‘nearest’. Returns a number given two values in this particle set.

  • frozen – if true, then this WeightedParticles object cannot be modified. This makes it hashable.

__getitem__()

Returns the probability of value; normalized

__setitem__()

The particle belief does not support assigning an exact probability to a value.

add(self, particle)

particle: (value, weight) tuple

condense()

Returns a new set of weighted particles with unique values and weights aggregated (taken average).

classmethod from_histogram(histogram, frozen=False)

Given a pomdp_py.Histogram return a particle representation of it, which is an approximation

frozen
get_histogram(self)

Returns a mapping from value to probability, normalized.

hist
hist_valid
mpe(self)

Returns the value of the variable that has the highest probability.

particles
random()

Samples a value based on the particles

values
weights

pomdp_py.representations.distribution.gaussian module

class pomdp_py.representations.distribution.gaussian.Gaussian

Bases: GenerativeDistribution

Note that Gaussian distribution is defined over a vector of numerical variables, but not a State variable.

__init__(self, mean, cov)

__getitem__(self, value)

Evaluate the probability of given value

Parameters:

value (list or array like) –

Returns:

The density of multivariate Gaussian.

Return type:

float

__setitem__(self, value, prob)

Not Implemented; It isn’t supported to arbitrarily change a value in a Gaussian distribution to have a given probability.

cov
covariance
mean
mpe(self)
random(self, n=1)
sigma