Introduction for R package Diffnet

Differential network is an important tool to capture the changes of conditional correlations under two sample cases. We develop an efficient fista algorithm via the symmetric quadratic loss for differential matrix estimation. The computation complexity of our algorithm is linear in the sample size and the number of parameters, which is optimal in the sense that it is of the same order as computing two sample covariance matrices.

References

Zhou Tang and Zhangsheng and Cheng Wang. "An efficient ADMM algorithm for high dimensional precision matrix estimation via penalized quadratic loss", 2019+. (arxiv).

Getting Started

These instructions will give you a toy example for implementing the package.

Prerequisites

What things you need to install the software and how to install them. The key functions of the package is writing in C++. So, make sure your OS can complies C++ code. For example, you should install Rtools under Windows and Xcode under MacOS. After that, the following R packages are also necessary.

install.packages("devtools")
install.packages("MASS")
install.packages("Matrix")

Install Diffnet

Install from source.

1. Download the Diffnet and open it with the project mode in the Rstudio.
2. Main Menu -> Build -> Install and Restart.

Install from binary prebuild release .

1. Download the tar.gz prebuild file in release tag and save in YOUR_ADDRESS.

2. type the following command in the Rstudio:

   install.packages("YOUR_ADDRESS", repos = NULL, type = "source") 
   

Toy example

library(diffnet)
rm(list = ls())
set.seed(123)
library('MASS')
library('Matrix')
library('Diffnet')
## ---------------------------- preprocess -----------------------------------------
n_X = 100; n_Y = n_X;
p_X = 100; p_Y = p_X;
nlambda = 50
tuning = "none"
case = "case1"
stop.tol = 1e-5
perturb = F
correlation = F
max.iter = 800
lambda.min.ratio = 0.5

# ------------ data generating -----------------
data = diffnet.case(n=n_X, p=p_X, method = case)
X = data$X
Y = data$Y
diff.Omega = data$diff.Omega
print(sum(diff.Omega!=0))

start = proc.time()[3]
result = diffnet(X, Y, verbose = F, nlambda = nlambda, max.iter = max.iter, lambda.min.ratio = lambda.min.ratio, stop.tol=stop.tol, method = "lasso", perturb =perturb, correlation = correlation)
elapse = proc.time()[3] - start

print(result$path[[50]][1:5,1:5])