f227455a |
1 | meanX = rep(0,6) |
2 | covX = 0.1*diag(6) |
3 | |
4 | covY = array(dim = c(5,5,2)) |
5 | covY[,,1] = 0.1*diag(5) |
6 | covY[,,2] = 0.2*diag(5) |
7 | |
8 | beta = array(dim = c(6,5,2)) |
9 | beta[,,2] = matrix(c(rep(2,12),rep(0, 18))) |
10 | beta[,,1] = matrix(c(rep(1,12),rep(0, 18))) |
11 | |
12 | n = 500 |
13 | |
14 | pi = c(0.4,0.6) |
15 | |
16 | source('~/valse/R/generateSampleInputs.R') |
17 | data = generateXY(meanX,covX,covY, pi, beta, n) |
18 | |
19 | X = data$X |
20 | Y = data$Y |
21 | |
22 | k = 2 |
23 | |
24 | n = nrow(Y) |
25 | m = ncol(Y) |
26 | p = ncol(X) |
27 | |
28 | Zinit1 = array(0, dim=c(n)) |
29 | betaInit1 = array(0, dim=c(p,m,k)) |
30 | sigmaInit1 = array(0, dim = c(m,m,k)) |
31 | phiInit1 = array(0, dim = c(p,m,k)) |
32 | rhoInit1 = array(0, dim = c(m,m,k)) |
33 | Gam = matrix(0, n, k) |
34 | piInit1 = matrix(0,k) |
35 | gamInit1 = array(0, dim=c(n,k)) |
36 | LLFinit1 = list() |
37 | |
38 | require(MASS) #Moore-Penrose generalized inverse of matrix |
39 | |
40 | distance_clus = dist(X) |
41 | tree_hier = hclust(distance_clus) |
42 | Zinit1 = cutree(tree_hier, k) |
43 | sum(Zinit1==1) |
44 | |
45 | for(r in 1:k) |
46 | { |
47 | Z = Zinit1 |
48 | Z_indice = seq_len(n)[Z == r] #renvoit les indices où Z==r |
49 | if (length(Z_indice) == 1) { |
50 | betaInit1[,,r] = ginv(crossprod(t(X[Z_indice,]))) %*% |
51 | crossprod(t(X[Z_indice,]), Y[Z_indice,]) |
52 | } else { |
53 | betaInit1[,,r] = ginv(crossprod(X[Z_indice,])) %*% |
54 | crossprod(X[Z_indice,], Y[Z_indice,]) |
55 | } |
56 | sigmaInit1[,,r] = diag(m) |
57 | phiInit1[,,r] = betaInit1[,,r] #/ sigmaInit1[,,r] |
58 | rhoInit1[,,r] = solve(sigmaInit1[,,r]) |
59 | piInit1[r] = mean(Z == r) |
60 | } |
61 | |
62 | for(i in 1:n) |
63 | { |
64 | for(r in 1:k) |
65 | { |
66 | dotProduct = tcrossprod(Y[i,]%*%rhoInit1[,,r]-X[i,]%*%phiInit1[,,r]) |
67 | Gam[i,r] = piInit1[r]*det(rhoInit1[,,r])*exp(-0.5*dotProduct) |
68 | } |
69 | sumGamI = sum(Gam[i,]) |
70 | gamInit1[i,]= Gam[i,] / sumGamI |
71 | } |
72 | |
73 | miniInit = 10 |
74 | maxiInit = 101 |
75 | |
76 | new_EMG = EMGLLF(phiInit1,rhoInit1,piInit1,gamInit1,miniInit,maxiInit,1,0,X,Y,1e-6) |
77 | |
78 | new_EMG$phi |
79 | new_EMG$pi |
80 | LLFEessai = new_EMG$LLF |
81 | LLFinit1 = LLFEessai[length(LLFEessai)] |
82 | |
83 | |
84 | b = which.max(LLFinit1) |
85 | phiInit = phiInit1[,,,b] |
86 | rhoInit = rhoInit1[,,,b] |
87 | piInit = piInit1[b,] |
88 | gamInit = gamInit1[,,b] |