`altar.cuda.distributions`

Submodules

Package Contents

Classes

Functions

class altar.cuda.distributions.distribution

Bases: altar.protocol

The protocol that all AlTar probability distributions must satisfy

parameters

doc = the number of model parameters that i take care of

offset

doc = the starting point of my parameters in the overall model state

initialize(self, **kwds): Initialize with the given random number generator

initializeSample(self, theta): Fill my portion of {theta} with initial random values from my distribution.

priorLikelihood(self, theta, prior): Fill my portion of {prior} with the likelihoods of the samples in {theta}

verify(self, theta, mask): Check whether my portion of the samples in {theta} are consistent with my constraints, and update {mask}, a vector with zeroes for valid samples and non-zero for invalid ones

sample(self): Sample the distribution using a random number generator

density(self, x): Compute the probability density of the distribution at {x}

vector(self, vector): Fill {vector} with random values

matrix(self, matrix): Fill {matrix} with random values

classmethod pyre_default(cls): Supply a default implementation

class altar.cuda.distributions.cudaDistribution(name, locator, **kwds)

Bases: altar.distributions.Base.Base

The base class for probability distributions

parameters

doc = the number of model parameters that belong to me

offset

doc = the starting point of my parameters in the overall model state

device

idx_range

precision

initialize(self, rng): Initialize with the given random number generator

verify(self, theta, mask): Check whether my portion of the samples in {theta} are consistent with my constraints, and update {mask}, a vector with zeroes for valid samples and non-zero for invalid ones

cuInitialize(self, application): cuda specific initialization

cuInitSample(self, theta): cuda process to initialize random samples

cuVerify(self, theta, mask): cuda process to verify the validity of samples

cuEvalPrior(self, theta, prior): cuda process to compute the prior

altar.cuda.distributions.uniform()

altar.cuda.distributions.gaussian()

altar.cuda.distributions.tgaussian()

altar.cuda.distributions.preset()