Commit | Line | Data |
---|---|---|
228ee602 | 1 | #' valse |
2 | #' | |
3 | #' Main function | |
4 | #' | |
5 | #' @param X matrix of covariates (of size n*p) | |
6 | #' @param Y matrix of responses (of size n*m) | |
7 | #' @param procedure among 'LassoMLE' or 'LassoRank' | |
8 | #' @param selecMod method to select a model among 'DDSE', 'DJump', 'BIC' or 'AIC' | |
9 | #' @param gamma integer for the power in the penaly, by default = 1 | |
10 | #' @param mini integer, minimum number of iterations in the EM algorithm, by default = 10 | |
11 | #' @param maxi integer, maximum number of iterations in the EM algorithm, by default = 100 | |
12 | #' @param eps real, threshold to say the EM algorithm converges, by default = 1e-4 | |
13 | #' @param kmin integer, minimum number of clusters, by default = 2 | |
14 | #' @param kmax integer, maximum number of clusters, by default = 10 | |
15 | #' @param rank.min integer, minimum rank in the low rank procedure, by default = 1 | |
16 | #' @param rank.max integer, maximum rank in the low rank procedure, by default = 5 | |
17 | #' @param ncores_outer Number of cores for the outer loop on k | |
18 | #' @param ncores_inner Number of cores for the inner loop on lambda | |
19 | #' @param thresh real, threshold to say a variable is relevant, by default = 1e-8 | |
9cb34faf | 20 | #' @param grid_lambda, a vector with regularization parameters if known, by default numeric(0) |
228ee602 | 21 | #' @param size_coll_mod (Maximum) size of a collection of models |
22 | #' @param fast TRUE to use compiled C code, FALSE for R code only | |
23 | #' @param verbose TRUE to show some execution traces | |
24 | #' | |
25 | #' @return a list with estimators of parameters | |
26 | #' | |
27 | #' @examples | |
28 | #' #TODO: a few examples | |
29 | #' @export | |
30 | valse <- function(X, Y, procedure = "LassoMLE", selecMod = "DDSE", gamma = 1, mini = 10, | |
31 | maxi = 50, eps = 1e-04, kmin = 2, kmax = 3, rank.min = 1, rank.max = 5, ncores_outer = 1, | |
9cb34faf BA |
32 | ncores_inner = 1, thresh = 1e-08, grid_lambda = numeric(0), size_coll_mod = 10, |
33 | fast = TRUE, verbose = FALSE, plot = TRUE) | |
228ee602 | 34 | { |
35 | n <- nrow(X) | |
36 | p <- ncol(X) | |
37 | m <- ncol(Y) | |
38 | ||
39 | if (verbose) | |
40 | print("main loop: over all k and all lambda") | |
41 | ||
42 | if (ncores_outer > 1) { | |
43 | cl <- parallel::makeCluster(ncores_outer, outfile = "") | |
44 | parallel::clusterExport(cl = cl, envir = environment(), varlist = c("X", | |
45 | "Y", "procedure", "selecMod", "gamma", "mini", "maxi", "eps", "kmin", | |
46 | "kmax", "rank.min", "rank.max", "ncores_outer", "ncores_inner", "thresh", | |
47 | "size_coll_mod", "verbose", "p", "m")) | |
48 | } | |
49 | ||
50 | # Compute models with k components | |
51 | computeModels <- function(k) | |
52 | { | |
53 | if (ncores_outer > 1) | |
54 | require("valse") #nodes start with an empty environment | |
55 | ||
56 | if (verbose) | |
57 | print(paste("Parameters initialization for k =", k)) | |
58 | # smallEM initializes parameters by k-means and regression model in each | |
59 | # component, doing this 20 times, and keeping the values maximizing the | |
60 | # likelihood after 10 iterations of the EM algorithm. | |
61 | P <- initSmallEM(k, X, Y, fast) | |
9cb34faf | 62 | if (length(grid_lambda) == 0) |
228ee602 | 63 | { |
64 | grid_lambda <- computeGridLambda(P$phiInit, P$rhoInit, P$piInit, P$gamInit, | |
65 | X, Y, gamma, mini, maxi, eps, fast) | |
66 | } | |
67 | if (length(grid_lambda) > size_coll_mod) | |
68 | grid_lambda <- grid_lambda[seq(1, length(grid_lambda), length.out = size_coll_mod)] | |
69 | ||
70 | if (verbose) | |
71 | print("Compute relevant parameters") | |
72 | # select variables according to each regularization parameter from the grid: | |
73 | # S$selected corresponding to selected variables | |
74 | S <- selectVariables(P$phiInit, P$rhoInit, P$piInit, P$gamInit, mini, maxi, | |
75 | gamma, grid_lambda, X, Y, thresh, eps, ncores_inner, fast) | |
76 | ||
77 | if (procedure == "LassoMLE") { | |
78 | if (verbose) | |
79 | print("run the procedure Lasso-MLE") | |
80 | # compute parameter estimations, with the Maximum Likelihood Estimator, | |
81 | # restricted on selected variables. | |
82 | models <- constructionModelesLassoMLE(P$phiInit, P$rhoInit, P$piInit, | |
83 | P$gamInit, mini, maxi, gamma, X, Y, eps, S, ncores_inner, fast, verbose) | |
84 | } else { | |
85 | if (verbose) | |
86 | print("run the procedure Lasso-Rank") | |
87 | # compute parameter estimations, with the Low Rank Estimator, restricted on | |
88 | # selected variables. | |
89 | models <- constructionModelesLassoRank(S, k, mini, maxi, X, Y, eps, rank.min, | |
90 | rank.max, ncores_inner, fast, verbose) | |
91 | } | |
92 | # warning! Some models are NULL after running selectVariables | |
93 | models <- models[sapply(models, function(cell) !is.null(cell))] | |
94 | models | |
95 | } | |
96 | ||
97 | # List (index k) of lists (index lambda) of models | |
98 | models_list <- | |
99 | if (ncores_outer > 1) { | |
100 | parLapply(cl, kmin:kmax, computeModels) | |
101 | } else { | |
102 | lapply(kmin:kmax, computeModels) | |
103 | } | |
104 | if (ncores_outer > 1) | |
105 | parallel::stopCluster(cl) | |
106 | ||
107 | if (!requireNamespace("capushe", quietly = TRUE)) | |
108 | { | |
109 | warning("'capushe' not available: returning all models") | |
110 | return(models_list) | |
111 | } | |
112 | ||
113 | # Get summary 'tableauRecap' from models | |
114 | tableauRecap <- do.call(rbind, lapply(seq_along(models_list), function(i) | |
115 | { | |
116 | models <- models_list[[i]] | |
117 | # For a collection of models (same k, several lambda): | |
118 | LLH <- sapply(models, function(model) model$llh[1]) | |
119 | k <- length(models[[1]]$pi) | |
120 | sumPen <- sapply(models, function(model) k * (dim(model$rho)[1] + sum(model$phi[, | |
121 | , 1] != 0) + 1) - 1) | |
122 | data.frame(model = paste(i, ".", seq_along(models), sep = ""), pen = sumPen/n, | |
123 | complexity = sumPen, contrast = -LLH) | |
124 | })) | |
125 | tableauRecap <- tableauRecap[which(tableauRecap[, 4] != Inf), ] | |
126 | if (verbose == TRUE) | |
228ee602 | 127 | print(tableauRecap) |
228ee602 | 128 | modSel <- capushe::capushe(tableauRecap, n) |
129 | indModSel <- if (selecMod == "DDSE") | |
130 | as.numeric(modSel@DDSE@model) else if (selecMod == "Djump") | |
131 | as.numeric(modSel@Djump@model) else if (selecMod == "BIC") | |
132 | modSel@BIC_capushe$model else if (selecMod == "AIC") | |
133 | modSel@AIC_capushe$model | |
134 | ||
135 | mod <- as.character(tableauRecap[indModSel, 1]) | |
136 | listMod <- as.integer(unlist(strsplit(mod, "[.]"))) | |
137 | modelSel <- models_list[[listMod[1]]][[listMod[2]]] | |
138 | ||
139 | ## Affectations | |
140 | Gam <- matrix(0, ncol = length(modelSel$pi), nrow = n) | |
141 | for (i in 1:n) | |
142 | { | |
143 | for (r in 1:length(modelSel$pi)) | |
144 | { | |
145 | sqNorm2 <- sum((Y[i, ] %*% modelSel$rho[, , r] - X[i, ] %*% modelSel$phi[, , r])^2) | |
146 | Gam[i, r] <- modelSel$pi[r] * exp(-0.5 * sqNorm2) * gdet(modelSel$rho[, , r]) | |
147 | } | |
148 | } | |
149 | Gam <- Gam/rowSums(Gam) | |
150 | modelSel$affec <- apply(Gam, 1, which.max) | |
151 | modelSel$proba <- Gam | |
152 | modelSel$tableau <- tableauRecap | |
153 | ||
154 | if (plot) | |
155 | print(plot_valse(X, Y, modelSel, n)) | |
156 | ||
157 | return(modelSel) | |
158 | } |