Common and Individual Structure Explained for Multiple Graphs

Specific dimension reduction methods for replicated graphs (multiple undirected graphs repeatedly measured on a common set of nodes). The package contains efficient procedures for estimating a shared baseline propensity matrix and graph-specific low rank matrices. The algorithm uses block coordinate descent algorithm to solve the model, which alternatively performs L2-penalized logistic regression and multiple partial eigenvalue decompositions, as described in the paper Wang et al. (2017) .

CISE is an R package that fits the multiple-gragh factorization (M-GRAF) model ( Wang et al., 2017) for separating common and individual low rank structure for multiple graphs with a common set of nodes.



# Or the development version from GitHub:
# install.packages("devtools")


The package contains a dataset of binary brain networks of 212 subjects. After loading the data, we can apply 3 variants of M-GRAF model to decompose the data. Model checking is needed to evaluate which model provides the best fit.

# load data
# A is an array of adjacency matrices of multiple graphs

# specify low rank K=5
res = MGRAF2(A = A, K=5, tol=0.01, maxit=5)

# check the common structure - baseline propensities of edges between pairwise nodes
# check low rank component for each graph
# check joint log-likelihood across iterations


Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.


0.1.0 by Lu Wang, a year ago

Browse source code at

Authors: Lu Wang [aut, cre]

Documentation:   PDF Manual  

GPL-2 license

Imports far, gdata, glmnet, MASS, Matrix, rARPACK

Suggests knitr, rmarkdown, testthat

See at CRAN