Update tests
[valse.git] / pkg / R / constructionModelesLassoMLE.R
CommitLineData
0ba1b11c 1#' constructionModelesLassoMLE
3453829e
BA
2#'
3#' Construct a collection of models with the Lasso-MLE procedure.
0ba1b11c 4#'
3453829e
BA
5#' @param phiInit an initialization for phi, get by initSmallEM.R
6#' @param rhoInit an initialization for rho, get by initSmallEM.R
7#' @param piInit an initialization for pi, get by initSmallEM.R
8#' @param gamInit an initialization for gam, get by initSmallEM.R
9#' @param mini integer, minimum number of iterations in the EM algorithm, by default = 10
10#' @param maxi integer, maximum number of iterations in the EM algorithm, by default = 100
11#' @param gamma integer for the power in the penaly, by default = 1
12#' @param X matrix of covariates (of size n*p)
13#' @param Y matrix of responses (of size n*m)
14#' @param eps real, threshold to say the EM algorithm converges, by default = 1e-4
15#' @param S output of selectVariables.R
16#' @param ncores Number of cores, by default = 3
17#' @param fast TRUE to use compiled C code, FALSE for R code only
18#' @param verbose TRUE to show some execution traces
0ba1b11c 19#'
3453829e
BA
20#' @return a list with several models, defined by phi, rho, pi, llh
21#'
22#' @export
0ba1b11c 23constructionModelesLassoMLE <- function(phiInit, rhoInit, piInit, gamInit, mini,
fb3557f3 24 maxi, gamma, X, Y, eps, S, ncores, fast, verbose)
3453829e
BA
25{
26 if (ncores > 1)
27 {
28 cl <- parallel::makeCluster(ncores, outfile = "")
0ba1b11c
BA
29 parallel::clusterExport(cl, envir = environment(), varlist = c("phiInit",
30 "rhoInit", "gamInit", "mini", "maxi", "gamma", "X", "Y", "eps", "S",
3453829e
BA
31 "ncores", "fast", "verbose"))
32 }
33
34 # Individual model computation
35 computeAtLambda <- function(lambda)
36 {
0ba1b11c 37 if (ncores > 1)
3453829e
BA
38 require("valse") #nodes start with an empty environment
39
0ba1b11c 40 if (verbose)
3453829e
BA
41 print(paste("Computations for lambda=", lambda))
42
43 n <- nrow(X)
44 p <- ncol(X)
45 m <- ncol(Y)
46 k <- length(piInit)
47 sel.lambda <- S[[lambda]]$selected
48 # col.sel = which(colSums(sel.lambda)!=0) #if boolean matrix
49 col.sel <- which(sapply(sel.lambda, length) > 0) #if list of selected vars
0ba1b11c 50 if (length(col.sel) == 0)
3453829e
BA
51 return(NULL)
52
53 # lambda == 0 because we compute the EMV: no penalization here
fb3557f3
BA
54 res <- EMGLLF(array(phiInit[col.sel, , ], dim=c(length(col.sel),m,k)),
55 rhoInit, piInit, gamInit, mini, maxi, gamma, 0,
56 as.matrix(X[, col.sel]), Y, eps, fast)
3453829e
BA
57
58 # Eval dimension from the result + selected
59 phiLambda2 <- res$phi
60 rhoLambda <- res$rho
61 piLambda <- res$pi
62 phiLambda <- array(0, dim = c(p, m, k))
63 for (j in seq_along(col.sel))
64 phiLambda[col.sel[j], sel.lambda[[j]], ] <- phiLambda2[j, sel.lambda[[j]], ]
65 dimension <- length(unlist(sel.lambda))
66
67 ## Affectations
68 Gam <- matrix(0, ncol = length(piLambda), nrow = n)
69 for (i in 1:n)
70 {
71 for (r in 1:length(piLambda))
72 {
73 sqNorm2 <- sum((Y[i, ] %*% rhoLambda[, , r] - X[i, ] %*% phiLambda[, , r])^2)
74 Gam[i, r] <- piLambda[r] * exp(-0.5 * sqNorm2) * det(rhoLambda[, , r])
75 }
76 }
77 Gam2 <- Gam/rowSums(Gam)
78 affec <- apply(Gam2, 1, which.max)
79 proba <- Gam2
80 LLH <- c(sum(log(apply(Gam,1,sum))), (dimension + m + 1) * k - 1)
81 # ## Computation of the loglikelihood
82 # # Precompute det(rhoLambda[,,r]) for r in 1...k
83 # detRho <- sapply(1:k, function(r) gdet(rhoLambda[, , r]))
84 # sumLogLLH <- 0
85 # for (i in 1:n)
86 # {
87 # # Update gam[,]; use log to avoid numerical problems
88 # logGam <- sapply(1:k, function(r) {
89 # log(piLambda[r]) + log(detRho[r]) - 0.5 *
90 # sum((Y[i, ] %*% rhoLambda[, , r] - X[i, ] %*% phiLambda[, , r])^2)
91 # })
0ba1b11c 92 #
3453829e
BA
93 # #logGam <- logGam - max(logGam) #adjust without changing proportions -> change the LLH
94 # gam <- exp(logGam)
95 # norm_fact <- sum(gam)
96 # sumLogLLH <- sumLogLLH + log(norm_fact) - m/2* log(2 * base::pi)
97 # }
98 #llhLambda <- c(-sumLogLLH/n, (dimension + m + 1) * k - 1)
99 list(phi = phiLambda, rho = rhoLambda, pi = piLambda, llh = LLH, affec = affec, proba = proba)
100 }
101
102 # For each lambda, computation of the parameters
103 out <-
104 if (ncores > 1) {
105 parLapply(cl, 1:length(S), computeAtLambda)
106 } else {
107 lapply(1:length(S), computeAtLambda)
108 }
109
0ba1b11c 110 if (ncores > 1)
3453829e
BA
111 parallel::stopCluster(cl)
112
113 out
114}