-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathREADME.Rmd
111 lines (77 loc) · 3.41 KB
/
README.Rmd
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
---
title: "ADMMsigma"
output: github_document
---
```{r setup, include=FALSE}
knitr::opts_chunk$set(echo = TRUE, tidy = TRUE)
```
[](https://travis-ci.org/MGallow/ADMMsigma)
[](https://cran.r-project.org/package=ADMMsigma)
## Overview
`ADMMsigma` is an R package that estimates a penalized precision matrix via the alternating direction method of multipliers (ADMM) algorithm. It currently supports a general elastic-net penalty that allows for both ridge and lasso-type penalties as special cases.
<p align="center">
<img src = "https://github.com/MGallow/ADMMsigma/raw/master/vignettes/images/gif.gif"/>
</p>
A (possibly incomplete) list of functions contained in the package can be found below:
* `ADMMsigma()` computes the estimated precision matrix (ridge, lasso, and elastic-net type regularization optional)
* `RIDGEsigma()` computes the estimated ridge penalized precision matrix via closed-form solution
* `plot.ADMMsigma()` produces a heat map or line graph for cross validation errors
* `plot.RIDGEsigma()` produces a heat map or line graph for cross validation errors
See package [website](https://mgallow.github.io/ADMMsigma/) or [manual](https://github.com/MGallow/ADMMsigma/blob/master/ADMMsigma.pdf).
## Installation
```{r, eval = FALSE}
# The easiest way to install is from CRAN
install.packages("ADMMsigma")
# You can also install the development version from GitHub:
# install.packages("devtools")
devtools::install_github("MGallow/ADMMsigma")
```
If there are any issues/bugs, please let me know: [github](https://github.com/MGallow/ADMMsigma/issues). You can also contact me via my [website](https://mgallow.github.io/). Pull requests are welcome!
## Usage
```{r, message = FALSE}
library(ADMMsigma)
set.seed(1234)
# generate data from a sparse oracle precision matrix
# we will try to estimate this matrix from the data
# first compute the oracle covariance matrix
S = matrix(0.7, nrow = 5, ncol = 5)
for (i in 1:5){
for (j in 1:5){
S[i, j] = S[i, j]^abs(i - j)
}
}
# print oracle precision matrix
# because its sparse, some shrinkage might be useful
(Omega = round(qr.solve(S), 3))
# generate 100 x 5 data matrix with rows drawn from iid N_p(0, S)
set.seed(123)
Z = matrix(rnorm(100*5), nrow = 100, ncol = 5)
out = eigen(S, symmetric = TRUE)
S.sqrt = out$vectors %*% diag(out$values^0.5) %*% t(out$vectors)
X = Z %*% S.sqrt
# print sample precision matrix from the data
# this is perhaps a bad estimate (its not sparse)
sample = (nrow(X) - 1)/nrow(X)*cov(X)
round(qr.solve(sample), 5)
# now use ADMMsigma to provide estimates
# elastic-net type penalty (set tolerance to 1e-8)
ADMMsigma(X, tol.abs = 1e-8, tol.rel = 1e-8)
# lasso penalty (default tolerance)
ADMMsigma(X, alpha = 1)
# elastic-net penalty (alpha = 0.5)
ADMMsigma(X, alpha = 0.5)
# ridge penalty
ADMMsigma(X, alpha = 0)
# ridge penalty (using closed-form solution)
RIDGEsigma(X, lam = 10^seq(-8, 8, 0.01))
# produce CV heat map for ADMMsigma
ADMM = ADMMsigma(X, lam = 10^seq(-5, 5, 0.1), alpha = seq(0, 1, 0.1))
plot(ADMM, type = "heatmap")
# produce line graph for CV errors for ADMMsigma
plot(ADMM, type = "line")
# produce CV heat map for RIDGEsigma
RIDGE = RIDGEsigma(X, lam = 10^seq(-8, 8, 0.01))
plot(RIDGE, type = "heatmap")
# produce line graph for CV errors for RIDGEsigma
plot(RIDGE, type = "line")
```