Learn

Example:

import probpy as pp

prior = pp.normal.med(mu=1.0, sigma=2.0)
likelihood = pp.normal.med(sigma=1.0)

data = pp.normal.sample(mu=5.0, sigma=2.0, size=1000)

prior = pp.parameter_posterior(data, likelihood=likelihood, priors=prior)

probpy.learn.posterior.posterior.parameter_posterior(data: typing.Union[numpy.ndarray, typing.Tuple[numpy.ndarray]], likelihood: typing.Union[probpy.core.RandomVariable, typing.Callable[[typing.Tuple[numpy.ndarray]], numpy.ndarray]], prior: probpy.core.RandomVariable, mode='mcmc', **kwargs) → probpy.core.RandomVariable

Estimate the posterior distribution of some likelihood and priors. This function uses conjugate priors, mcmc or ga. If a likelihood is given conjugate priors then the mode argument will be ignored and a conjugate update will be done, because it is much faster.

Parameters:

• data – data for likelihood
• likelihood – likelihood function / distribution
• priors – prior or list of priors
• mode – mcmc or search
• kwargs – arguments passed to mcmc / ga

Returns:

RandomVariable

probpy.learn.posterior.mcmc.mcmc(data: typing.Tuple[numpy.ndarray], likelihood: typing.Union[probpy.core.RandomVariable, typing.Callable[[typing.Tuple[numpy.ndarray]], numpy.ndarray]], prior: probpy.core.RandomVariable, samples: int = 1000, mixing: int = 0, energy: float = 0.5, batch: int = 5, match_moments_for: probpy.core.Distribution = None, normalize: bool = True, density: probpy.core.Density = None)

Don’t call this function directly, always use parameter_posterior with mode=”mcmc”

Parameters:

• data – data passed to likelihood
• likelihood – likelihood function / distribution
• prior – prior distribution
• samples – number of mcmc samples to generate
• mixing – number of initial samples to ignore
• energy – variance in exploration
• batch – number of particles to run concurrently
• match_moments_for – distributions to force posterior into using moment matching
• normalize – normalize the resulting density
• density – density estimator

Returns:

RandomVariable

probpy.learn.posterior.search.search(data: typing.Tuple[numpy.ndarray], likelihood: typing.Union[probpy.core.RandomVariable, typing.Callable[[typing.Tuple[numpy.ndarray]], numpy.ndarray]], prior: probpy.core.RandomVariable, samples: int = 1000, energy: float = 0.5, batch=5, volume=10.0, normalize: bool = False, density: probpy.core.Density = None, **ubrk_args)

Don’t call this function directly, always use parameter_posterior with mode=”search”

Parameters:

• volume – volume of elements
• data – data passed to likelihood
• likelihood – likelihood function
• prior – prior / priors
• samples – samples in search estimate
• energy – energies in exploration
• batch – number of samples run concurrently
• normalize – normalize posterior
• density – density estimator
• ubrk_args – arguments to ubrk (default density estimator)

Returns:

RandomVariable