Type: Package
Title: Modelling Multivariate Binary Data with Blocks of Specific One-Factor Distribution
Version: 1.1
Date: 2016-12-15
Author: Matthieu Marbac and Mohammed Sedki
Maintainer: Mohammed Sedki <mohammed.sedki@u-psud.fr>
Description: Modelling Multivariate Binary Data with Blocks of Specific One-Factor Distribution. Variables are grouped into independent blocks. Each variable is described by two continuous parameters (its marginal probability and its dependency strength with the other block variables), and one binary parameter (positive or negative dependency). Model selection consists in the estimation of the repartition of the variables into blocks. It is carried out by the maximization of the BIC criterion by a deterministic (faster) algorithm or by a stochastic (more time consuming but optimal) algorithm. Tool functions facilitate the model interpretation.
License: GPL-2 | GPL-3 [expanded from: GPL (≥ 2)]
Imports: methods, mgcv, parallel
Depends: R (≥ 3.0.2)
Repository: CRAN
Repository/R-Forge/Project: mvbinary
Repository/R-Forge/Revision: 31
Repository/R-Forge/DateTimeStamp: 2016-12-14 14:16:05
Date/Publication: 2016-12-15 16:46:38
NeedsCompilation: no
Packaged: 2016-12-15 14:20:18 UTC; sedki

MvBinary a package for Multivariate Binary data

Description

MvBinary is a tool for fitting the distribution of correlated multivariate binary data.

Details

Package: MvBinary
Type: Package
Version: 1.0.0
Date: 2015-11-03
License: GPL-2
LazyLoad: yes

Author(s)

Author: Marbac M., and Sedki S.

References

Matthieu Marbac, Mohammed Sedki (2015). A Family of Blockwise One-Factor Distributions for Modelling High-Dimensional Binary Data. arXiv:1511.01343

Examples

# Package loading
rm(list=ls())
require(MvBinary)

# Data loading
data(MvBinaryExample)

# Parameter estimation by the HAC-based algorithm on 2 cores
# where the EM algorithms are initialized 10 times
res.CAH <- MvBinaryEstim(MvBinaryExample, 2, nbinit.EM = 10)

# Summary of the estimated model
summary(res.CAH)

# Print the parameters of the estimated model
print(res.CAH)

Computation of the Empiric Cramer'v.

Description

This function computes the Empiric Cramer's V for a binary data set.

Usage

ComputeEmpiricCramer(x)

Arguments

x

a binary matrix.

Value

Return the matrix of the Empiric Cramer's V.

Computation of the model Cramer'v.

Description

This function computes the model Cramer's V for a binary data set.

Usage

ComputeMvBinaryCramer(results)

Arguments

results

an instance of S4 class MvBinaryResult (provided by the function MvBinaryEstim)

Value

Return the matrix of the Empiric Cramer's V.

Create an instance of the [MvBinaryResult] class

Description

This function performs the model selection and the parameter inference.

Usage

MvBinaryEstim(x, nbcores = 1, algorithm = "HAC", modelslist = NULL,
  tol.EM = 0.01, nbinit.EM = 40, nbiter.MH = 50, nbchains.MH = 10)

Arguments

x

matrix of the binary observation.

nbcores

number of cores used for the model selection (only for Linux). Default is 1.

algorithm

algorithm used for the model selection ("HAC": deterministic algorithm based on the HAC of the variables, "MH": stochastic algorithm for optimizing the BIC criterion, "List": comparison of the models provided by the users). Default is "HAC".

modelslist

list of models provided by the user (only used when algorithm="List"). Default is NULL

tol.EM

stopping criterion for the EM algorithm. Default is 0.01

nbinit.EM

number of random initializations for the EM algorithm. Default is 40.

nbiter.MH

number of successive iterations without finding a model having a better BIC criterion which involves the stopping of the Metropolis-Hastings algorithm (only used when algorithm="MH"). Default is 50.

nbchains.MH

number of radom initializations for the stochastic algorithm (only used when algorithm="MH"). Default is 10.

Value

Returns an instance of the [MvBinaryResult] class.

Examples

# Data loading
data(MvBinaryExample)

# Parameter estimation by the HAC-based algorithm on 2 cores
# where the EM algorithms are initialized 10 times
res.CAH <- MvBinaryEstim(MvBinaryExample, 2, nbinit.EM = 10)

# Parameter estimation for two competing models
res.CAH <- MvBinaryEstim(MvBinaryExample, algorithm="List",
 modelslist=list(c(1,1,2,2,3,4), c(1,1,1,2,2,2)), nbinit.EM = 10)

# Summary of the estimated model
summary(res.CAH)

# Print the parameters of the estimated model
print(res.CAH)

Simulated binary data: MvBinaryExample

Description

The file MvBinaryExample.rda describes 400 individuals by 6 binary variables.

Format

A matrix with 400 observations on the 6 variables.

Details

This data set has been simulated from the MvBinary model. The first three variables are dependent. The last three variables are dependent.

Examples

data(MvBinaryExample)

Computation of the model Cramer'v.

Description

This function computes the model Cramer's V for a binary data set.

Usage

MvBinaryProbaPost(x, param)

Arguments

x

a binary matrix.

param

an instance of S4 class MvBinaryResult (provided by the function MvBinaryEstim)

Value

Return the logprobability for each row of matrix x conditionally on the model defined by param.

Constructor of [MvBinaryResult] class

Description

This S4 class contains the results from the function MvBinaryEstim.

Details

alpha

numeric. It indicates the marginal probability of that each variables are equal to 1.

epsilon

numeric. It indicates the dependency strength of each variables (between 0 and 1) with the other block variables.

delta

binary. It indicates the kind of dependency: two variables affiliated into the same block are positively correlated if their delta are equal and they are negatively correlated otherwise.

blocks

numeric. It indicates the block of each variable.

nbparam

numeric. It indicates the number of continuous parameters.

loglike

numeric. The model likelihood.

bic

numeric. The model BIC.

Examples

getSlots("MvBinaryResult")

Real binary data: Plants

Description

The file plants.rda describes 35583 plants by indicating if they occur (1) or not (2) in 69 states of the Norht America.

Format

A matrix with 35583 observations on the 69 variables.

Details

This data set been extracted from the USA plants database, July 29, 2015.

Examples

data(plants)

Summary function.

Description

This function prints the parameters resulting from MvBinaryEstim.

Usage

## S4 method for signature 'MvBinaryResult'
print(x)

Arguments

x

output object from MvBinaryEstim.

Summary function.

Description

This function gives the summary of output from MvBinaryEstim.

Usage

## S4 method for signature 'MvBinaryResult'
summary(object)

Arguments

object

output object from MvBinaryEstim.