model-class: Finmix 'model' class

In simonsays1980/finmix: An R package for Bayesian estimation of finite mixture distributions

Description

This class specifies a finite mixture model. Entities are created from it by calling its constructor model().

Details

A finite mixture model in the finmix package is defined by its number of components K, the component distributions dist, the data dimension r and an indicator defining, if the model has fixed or unknown indicators. Finite mixture models for the following distributions can be constructed:

• Poisson,

• Conditional Poisson,

• Exponential,

• Binomial,

• Normal,

• Multivariate Normal,

• Student-t,

• Multivariate Student-t.

Using the constructor model() a finite mixture model can be created, the default being a mixture model of Poisson distributions.

Fully defined finite mixture models

A fully defined finite mixture model contains next to the distribution and the components also weights and parameters. The weights are defined in slot weight and must be of class matrix with as many weights as there are components in the mixture model (dimension Kx1). Parameters are defined in a list named par. The elements of this list depend on the chosen distribution in slot dist:

• Poisson: A matrix named lambda of dimension Kx1 holding the rate parameters.

• Exponential: A matrix named lambda of dimension Kx1 holding the rate parameters.

• Binomial: A matrix of dimension Kx1 named p storing the probabilities.

Slots

dist

A character, defining the distribution family. Possible choices are binomial, exponential, normal, normult, poisson, student, and studmult.

r

An integer. Defines the vector dimension of a model. Is one for all univariate distributions and larger than one for normult and studmult.

K

An integer, defining the number of components in the finite mixture.

weight

A matrix, containing the weights of the finite mixture model. The matrix must have dimension 1 x K and weights must add to one must all be larger or equal to zero.

par

A list containing the parameter vectors for the finite mixture distribution. The list can contain more than one named parameter vector.

indicmod

A character defining the indicator model. So far only multinomial indicator models are possible.

indicfix

A logical. If TRUE the indicators are given and therefore fixed.

T

A matrix containing the repetitions in case of a "binomial" or "poisson" model.

See Also

• mixturemcmc() for performing MCMC sampling with a mixture model

• modelmoments() for compute theoretical moments of a finite mixture model

simonsays1980/finmix documentation built on Dec. 23, 2021, 2:25 a.m.