mc.glmnet: Multiple Core Multiple Model glmnet calculations

In jandziak/mc.glmnet: Multiple Core Multiple Model glmnet calculations

Description

Package mc.glmnet performs fit of L1 penalised model to the data. It is introduction to solution for Graphical Gaussion Models precission matrix. There had been used parallel computing in order to reduce time of computations.

Details

For solving simple lasso problem, the number of predictors is usualy reasonably small so the time of computations is not crutial. For Graphical Gaussian Models number of parameters to be estimated is squared.

The main function of package mc.glmnet is mcmm.glmnet which allows to solve multiple Lasso problems using parallel computations. For large number of predictros solution presents significant better performance than usage of simple onee core computations.

We also provide function sm.glmnet that computes single model single L1 penalised model for generated inside data.

Note that Windows users for propper evaluation of mcmm.glmnet and reduction in computation time, script available at http://www.stat.cmu.edu/~nmv/setup/mclapply.hack.R

Version: 0.1.1

Author(s)

Jan Idziak

Maintainer: Jan Idziak [e-mail]

Examples

library(GeneNet)
graph.data.generate <- function(seed = 42048, fraction = 0.1, p = 500, n = 50, b.values = rep(2,50)){
set.seed(seed)
data <- list()
precision <- ggm.simulate.pcor(p, fraction)
data$X <- ggm.simulate.data(n, precision)
beta <- c(b.values, rep(0, times = p-length(b.values)))
probs <- round(1/(1+(exp(-data$X%*%beta))), digits = 4)
data$Y <- rbinom(n, 1, probs)
return(data)}
sm.glmnet(graph.data.generate)
sm.glmnet(graph.data.generate, p = 500, n = 50)
sm.glmnet(graph.data.generate, fraction = 0.05 ,p = 500, n = 100)
mcmm.glmnet(graph.data.generate)
mcmm.glmnet(graph.data.generate, p = 500, n = 50)
mcmm.glmnet(graph.data.generate, fraction = 0.05 ,p = 500, n = 100)

jandziak/mc.glmnet documentation built on May 18, 2019, 12:23 p.m.