| 1 | constructionModelesLassoMLE = function(phiInit,rhoInit,piInit,gamInit,mini,maxi,gamma,glambda,X,Y,seuil,tau,selected){ |
| 2 | #get matrix sizes |
| 3 | n = dim(X)[1] |
| 4 | p = dim(phiInit)[1] |
| 5 | m = dim(phiInit)[2] |
| 6 | k = dim(phiInit)[3] |
| 7 | L = length(glambda) |
| 8 | |
| 9 | #output parameters |
| 10 | phi = array(0, dim = c(p,m,k,L)) |
| 11 | rho = array(dim = c(m,m,k,L)) |
| 12 | pi = array( dim = c(k,L)) |
| 13 | lvraisemblance = array( dim = c(L,2)) |
| 14 | |
| 15 | for (lambdaIndex in 1:L){ |
| 16 | # Procedure Lasso-MLE |
| 17 | a = selected[,1,lambdaIndex] |
| 18 | a(a==0) = c() |
| 19 | if (length(a) != 0){ |
| 20 | res_EM = EMGLLF(phiInit[a,,],rhoInit,piInit,gamInit,mini,maxi,gamma,0,X[,a],Y,tau) |
| 21 | phiLambda = res_EM$phi |
| 22 | rhoLambda = res_EM$rho |
| 23 | piLambda = res_EM$pi |
| 24 | for (j in 1:length(a)){ |
| 25 | phi[a[j],,,lambdaIndex] = phiLambda[j,,] |
| 26 | } |
| 27 | rho[,,,lambdaIndex] = rhoLambda |
| 28 | pi[,lambdaIndex] = piLambda |
| 29 | |
| 30 | dimension = 0 |
| 31 | for (j in 1:p){ |
| 32 | b = A2[j,2:end,lambdaIndex] |
| 33 | b[b==0] = c() |
| 34 | if (length(b) > 0){ |
| 35 | phi[A2[j,1,lambdaIndex],b,,lambdaIndex] = 0.0 |
| 36 | } |
| 37 | c = A1[j,2:end,lambdaIndex] |
| 38 | c[c==0] = c() |
| 39 | dimension = dimension + length(c) |
| 40 | } |
| 41 | |
| 42 | #on veut calculer l'EMV avec toutes nos estimations |
| 43 | densite = array(n,L) |
| 44 | for (i in 1:n){ |
| 45 | for (r in 1:k){ |
| 46 | delta = Y[i,]*rho[,,r,lambdaIndex] - X[i,a]*phi[a,,r,lambdaIndex] |
| 47 | densite[i,lambdaIndex] = densite[i,lambdaIndex] + pi[r,lambdaIndex]*det(rho[,,r,lambdaIndex])/(sqrt(2*base::pi))^m*exp(-delta %*% delta/2.0) |
| 48 | } |
| 49 | } |
| 50 | |
| 51 | lvraisemblance(lambdaIndex,1) = sum(log(densite[,lambdaIndex])) |
| 52 | lvraisemblance(lambdaIndex,2) = (dimension+m+1)*k-1 |
| 53 | } |
| 54 | } |
| 55 | } |