Ising/grplasso.R at master · hoenlab/Ising · GitHub

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
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
library(grplasso)
library(glmnet)
source("grplasso_supp.R")
netmat=function(A,width){
  ## this function conduct the logistics regression for each colomn of A on
  ## the other colomns that are not in the same group.
  ## A is the binary matrix representing the original data
  ##width is a vector representing the freedom of given taxa,width=M-1
  n=length(width) ##the number of taxa
  A=as.matrix(A)
  gamma=0.9
lam=0.1
  nlambda=200
  netmat=matrix(nrow = n,ncol = n)
  CoeVector=rep(0,n-1)
  for (i in 1:1){
    ## regression for the first taxa
    print(i)
    x=A[,-(1:width[1])]
    x=as.matrix(cbind(1,x))
    width1=width[-1]
    group1=rep(1:(n-1),width1)
    group1=c(NA,group1)

    for (j in 1:width1[1]) {
      y=as.vector(A[,j])
      #y=ifelse(y==-1,0,1)
      lambda=lambdamax(x,y=y,index=group1,penscale = sqrt,model = LogReg())*gamma^(0:nlambda)
      print(lambda)
      gg=grplasso(x,y=y,index=group1,lambda=lambda,model=LogReg(),penscale=sqrt,control=grpl.control(update.hess = "lambda",trace = 0))
      beta=gg$coefficients
      no=1:length(group1)
      beta=cbind(no,beta)
      aa=vector("list",length(lambda))
      dev=rep(0,(length(lambda)))
      eBIC=rep(0,(length(lambda)))
      for (k in 1:length(lambda)) {
        aa[[k]]=beta[beta[,k+1]!=0,1]
        dev[k]=glm.fit(x=x[,aa[[k]]],y=y,family = binomial(),intercept = T)$deviance
        eBIC[k]=dev[k]+length(aa[[k]])*(log(dim(A)[1])+lam*log(length(group1)))
      }
      k=which.min(eBIC)
      beta=beta[-1,]
      group=group1[-1]
      d=glength(beta[,k+1],group)
      CoeVector=c(1,d)
    }
  }
  netmat[1,]=CoeVector
  for (i in n:n){
    print(i)
    CoeVector=rep(0,n-1)
    cum=cumsum(width)
    x=A[,-(((cum[i-1]+1):cum[i]))]
    x=as.matrix(cbind(1,x))
    widthi=width[-i]
    group1=rep(1:(n-1),widthi)
    group1=c(NA,group1)
    for (j in 1:width[i]) {
      y=as.vector(A[,cum[i-1]+j])
      #y=ifelse(y==-1,0,1)
      lambda=lambdamax(x=x,y=y,index=group1,penscale = sqrt,model = LogReg())*gamma^(0:nlambda)
      print(lambda)
      gg=grplasso(x,y=y,index=group1,lambda=lambda,model=LogReg(),penscale=sqrt,control=grpl.control(update.hess = "lambda",trace = 0))
      beta=as.matrix(gg$coefficients)
      no=1:length(group1)
      beta=cbind(no,beta)
      aa=vector("list",length(lambda))
      dev=rep(0,(length(lambda)))
      eBIC=rep(0,(length(lambda)))
      for (k in 1:length(lambda)) {
        aa[[k]]=beta[beta[,k+1]!=0,1]
        dev[k]=glm.fit(x=x[,aa[[k]]],y=y,family = binomial(),intercept = T)$deviance
        eBIC[k]=dev[k]+length(aa[[k]])*(log(dim(A)[1])+lam*log(length(group1)))
      }
      k=which.min(eBIC)
      beta=beta[-1,]
      group=group1[-1]
      d=glength(beta[,k+1],group)
      CoeVector=CoeVector+d
    }
    CoeVector=c(CoeVector[1:(i-1)],1)
  }
  netmat[n,]=CoeVector
  for (i in 2:(n-1)){
    print(i)
    CoeVector=rep(0,n-1)
    ##regression for the remaining taxa
    cum=cumsum(width)
    x=A[,-(((cum[i-1]+1):cum[i]))]
    x=as.matrix(cbind(1,x))
    widthi=width[-i]
    group1=rep(1:(n-1),widthi)
    group1=c(NA,group1)
    for (j in 1:width[i]) {
      y=as.vector(A[,cum[i-1]+j])
      #y=ifelse(y==-1,0,1)
      lambda=lambdamax(x=x,y=y,index=group1,penscale = sqrt,model = LogReg())*gamma^(0:nlambda)
      print(lambda)
      gg=grplasso(x,y=y,index=group1,lambda=lambda,model=LogReg(),penscale=sqrt,control=grpl.control(update.hess = "lambda",trace = 0))

      beta=as.matrix(gg$coefficients)
      no=1:length(group1)
      beta=cbind(no,beta)
      aa=vector("list",length(lambda))
      dev=rep(0,(length(lambda)))
      eBIC=rep(0,(length(lambda)))
      for (k in 1:length(lambda)) {
        aa[[k]]=beta[beta[,k+1]!=0,1]
        dev[k]=glm.fit(x=x[,aa[[k]]],y=y,family = binomial(),intercept = T)$deviance
        eBIC[k]=dev[k]+length(aa[[k]])*(log(dim(A)[1])+lam*log(length(group1)))
      }
      k=which.min(eBIC)
      beta=beta[-1,]
      group=group1[-1]
      d=glength(beta[,k+1],group)
      CoeVector=CoeVector+d
    }
    CoeVector=c(CoeVector[1:(i-1)],1,CoeVector[i:(n-1)])
    netmat[i,]=CoeVector
  }
  return(netmat)
}
# TwoLevel=read.table("twolevel.cvs",sep=" ")
# M=scan("M.cvs")
# eBICnet=NetVector(A=TwoLevel,width = M-1,i=409)
# write(eBICnet,file="eBICnet.csv")