SIMULE

This is an R implementation of the SIMULE algorithm proposed in the following paper:

"A constrained L1 minimization approach for estimating multiple Sparse Gaussian or Nonparanormal Graphical Models", accepted by Machine Learning @ URL

Please run demo(simuleDemo) to learn the basic functions provided by this package. For further details, please read the original paper @ URL or read the R-package Manual: @ URL

• CRAN R Library page: URL

• Presentation @ URL

Dependency

It depends on the following existing packages. To use them, simply

library('pcaPP')
library('lpSolve')
library('parallel')

If you don't have these packages, simply use

install.packages('packageNameFromAbove')

Usage

0. install the R "simule" package through R console:

install.packages('simule')

1. then load the library simule in R console, by running:

library(simule)

2. Then, simply run the function simule on your favorite datasets For example,

simule(CovarianceMatrixList, lambda = 0.05, epsilon = 0.5, parallel = TRUE)

This function will returns a list (a data structure in R) of graphs estimated by the SIMULE package.

Three possible types of inputs for the Argument CovarianceMatrixList

1. The argument CovarianceMatrixList can represent a list of data matrices directly: The i-th item CovarianceMatrixList[[i]] represents the i-th matrix in a list of data matrices CovarianceMatrixList. ** Please make sure the order of the feature variables are the same among all the data matrices in CovarianceMatrixList.**

2. If the input CovarianceMatrixList is Symmetric, the package automatically assumes that the data inputs belong to the following two types:

• The argument CovarianceMatrixList can a list of covariance matrices. Assuming X represents a list of data matrices, whose i-th item X[[i]] represents the data matrix of the i-th task.

We can use the following function to calculate the covariance matrices:

CovarianceMatrixList[[i]] = cov(X[[i]])

• The argument CovarianceMatrixList can represent a list of kendall's tau correlation matrices. The kendall's tau correlation matrices can be calculated by using the following command:

cor.fk(X[[i]])  

(by the 'pcaPP' package.)

The kendall's tau correlation matrices can also be calculated through the following R functions:

cor(X[[i]], method = 'kendall')

However the above way of calculating kendall's tau correlation matrix is very slow in R.

Other Arguments

• lambda

The parameter for the sparsity level of the estimated graphs. The larger lambda you choose, the sparser graphs you will estimate from the inputs.

• epsilon

The parameter reflects the differences of sparsity level between the shared subgraph versus the context-specific subgraphs. The larger epsilon you choose, the denser the shared subgraph is (while the context-specific subgraphs are sparser) and vice versa.

• covType This parameter controls SIMULE estimates the sparse Gaussian Graphical models (sGGM) or the sparse Nonparanormal Graphical Models from the input data. This parameter matters only when the input argument CovarianceMatrixList represents a list of data matrices directly: When covType = "cov" the package estimates sGGMs from the input. When covType == "kendall", the package estimates sNGMs from the input.

• parallel

Logic parameter for parallel implementation or not. If you have a multi-core machine, let parallel = TRUE.