Description Usage Arguments Details Value Author(s) References Examples

Fit the Gaussian mixture model using either New-SP or New-JGL method. Sparsity and fusion penalty are imposed on the component-specific precision matrices with either non-convex (New-SP) or convex (New-JGL) function. The output contains the class assignment, the mixing proportion, the means and the component-specific precision matrices and the optimal tuning parameter (if a vector of tuning parameter is supplied).

1 2 | ```
pGMM(Y, k, method, lambda1, lambda2, ncluster, tau=0.01, threshold=1e-5,
MAX_iter=100, seed=1)
``` |

`Y` |
The input data as an |

`k` |
The fold for cross-validation. Default is 5. |

`method` |
Either "New-SP" or "New-JGL". "New-SP" uses the non-convex truncated lasso penalty (TLP). "New-JGL" uses the a convex lasso penalty. Default is "New-SP". |

`lambda1` |
Tuning parameter for the sparsity penalty, either a single value or a vector. If the input is a vector, cross-validation is applied to select the optimal value. |

`lambda2` |
Tuning parameter for the fusion penalty, either a single value or a vector. If the input is a vector, cross-validation is applied to select the optimal value. |

`ncluster` |
Number of components in the Gaussian mixture. |

`tau` |
The tuning paramter in the non-convex penalty function. Default is 0.01. |

`threshold` |
Threshold for convergence. Default is 1e-5. |

`MAX_iter` |
Maximum number of iterations. Default is 100. |

`seed` |
The seed used when splitting data for cross-validation. Default is 1. |

It aims to estimate multiple networks in the presence of sample heterogeneity, where the independent samples (i.e. observations) may come from different and unknown populations or distributions. Specifically, we consider penalized estimation of multiple precision matrices in the framework of a Gaussian mixture model. A major innovation is to take advantage of the commonalities across the multiple precision matrices through possibly nonconvex fusion regularization, which for example makes it possible to achieve simultaneous discovery of unknown disease subtypes and detection of differential gene (dys)regulations in functional genomics. We embed in the EM algorithm one of two recently proposed methods for estimating multiple precision matrices in Gaussian graphical models: one is based on the convex Lasso penalty (Danaher et al., 2014), and the other is based on the con-convex TLP penalty (Zhu et al., 2014); the former is faster, while the latter is less biased and will perform better with a relatively larger sample size.

The majority of the existing literature on mixture modeling focus on regularizing only the mean parameters with diagonal covariance matrices, though some have started considering regularization of the covariance parameters too, all of which, however, do not touch on the key issue of identifying both common and varying substructures of the precision matrices across the components of a mixture model. Since these methods always give different networks for different components unless a common network is assumed, they do not address the question of interest here: which parts of the networks change with the components. Since a fusion penalty is used to shrink multiple networks towards each other, the proposed methods not only are statistically more efficient with information borrowing, but also facilitate interpretation in identifying differential network substructures. In particular, due to the use of a non-convex penalty, the adopted method of Zhu et al. (2014) strives to uncover the commonalities among multiple networks while maintaining their unique substructures too.

`pie` |
A vector of the mixing proportion of each component in the model. |

`mu` |
A matrix that contains the mean of each component in the model in each row. |

`covinv` |
A list that contains the precision matrices for each component in the model. |

`membership` |
The class assignment for each observation. |

`par` |
The optimal value of lambda_1 and lambda_2. |

Chen Gao, Yunzhang Zhu

Gao, C., hu, Y., Shen, X., and Pan, W. (2016). Estimation of multiple networks in Gaussian mixture models, *Electronic Journal of Statistics*, **10**(1), 1133–1154.

http://projecteuclid.org/download/pdfview_1/euclid.ejs/1462192266

Zhu, Y., Shen, X., and Pan, W. (2014). Structural pursuit over multiple
undirected graphs, *Journal of the American Statistical Association*, **109**(508), 1683–1696.

http://www.tandfonline.com/doi/pdf/10.1080/01621459.2014.921182

Danaher, P., Wang, P., and Witten, D. M. (2014). The joint graphical lasso
for inverse covariance estimation across multiple classes, *Journal of the Royal
Statistical Society, Series B*, **76**(2), 373–397.

http://onlinelibrary.wiley.com/doi/10.1111/rssb.12033/epdf

1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 | ```
data(jglfit)
# simulate data
set.seed(12345)
Y1 = rmvnorm(1*sum(jglfit$membership==1), jglfit$mu[1,], solve(round(jglfit$covinv[[1]], digits=3)))
Y2 = rmvnorm(1*sum(jglfit$membership==2), jglfit$mu[2,], solve(round(jglfit$covinv[[2]], digits=3)))
Y3 = rmvnorm(1*sum(jglfit$membership==3), jglfit$mu[3,], solve(round(jglfit$covinv[[3]], digits=3)))
Y4 = rmvnorm(1*sum(jglfit$membership==4), jglfit$mu[4,], solve(round(jglfit$covinv[[4]], digits=3)))
Y=rbind(Y1,Y2,Y3,Y4)
p=ncol(Y)
# test
test = pGMM(Y, k=2, method="New-SP", lambda1=1, lambda2=1, ncluster=4)
``` |

pGMGM documentation built on May 30, 2017, 3:09 a.m.

Embedding an R snippet on your website

Add the following code to your website.

For more information on customizing the embed code, read Embedding Snippets.