Like Categorical but the data are counts, so _get_statistics is overridden (though _get_weighted_statistics can stay the same!). log_likeliho

This class represents a categorical distribution over labels, where the parameter is weights and the prior is a Dirichlet distribution. F

Categorical with resampling of the symmetric Dirichlet concentration parameter. concentration ~ Gamma(a_0,b_0) The Dirichlet pr

Implements a Gamma(a_0,b_0) prior over finite dirichlet concentration parameter. The concentration is scaled according to the weak-limit sequ

Multivariate von-Mises Fisher distribution class. NOTE: Only works for 2 or more dimensions. Uses the following prior. x

concentration ~ Gamma(a_0,b_0) [b_0 is inverse scale, inverse of numpy scale arg] rvs ~ CRP(concentration) This class models CRPs. The p

Product of normal-inverse-gamma priors over mu (mean vector) and sigmas (vector of scalar variances). The prior follows sigmas

Product of normal priors over mu and product of gamma priors over sigmas. Note that while the conjugate prior in DiagonalGaussian is of the f

Multivariate Gaussian distribution class. NOTE: Only works for 2 or more dimensions. For a scalar Gaussian, use one of the scalar classe

Geometric distribution with a conjugate beta prior. NOTE: the support is {1,2,3,...} Hyperparameters: alpha_0, beta_0 Param

Normal-Inverse-Gamma prior over mu (mean vector) and sigma (scalar variance). Essentially, all coordinates of all observations inform the

This class is for mixtures of other distributions.

This makes a Mixture act like a Distribution for use in other models

Nonconjugate Discrete+Beta prior r_discrete_distribution is an array where index i is p(r=i+1)

Poisson distribution with a conjugate Gamma prior. NOTE: the support is {0,1,2,...} Hyperparameters (following Wikipedia's notation):

Conjugate normal prior on mean.

Conjugate Normal-(Scaled-)Inverse-ChiSquared prior. (Another parameterization is the Normal-Inverse-Gamma.)

Non-conjugate separate priors on mean and variance parameters, via mu ~ Normal(mu_0,tausq_0) sigmasq ~ (Scaled-)Inverse-ChiSquared(sigmas

Models a uniform distribution over [low,high] for parameters low and high. The prior is non-conjugate (though it's conditionally conjugate ov

Models a uniform distribution over [low,high] for a parameter high. Low is a fixed hyperparameter (hence "OneSided"). See the Uniform class f

Negative Binomial distribution with a conjugate beta prior on p and a separate gamma prior on r. The parameter r does not need to be an integ