update

Author: snoweye

DESCRIPTION

@@ -1,6 +1,6 @@
 Package: pmclust
 Version: 0.1-5
-Date: 2013-06-13
+Date: 2013-05-13
 Title: Parallel Model-Based Clustering
 Authors@R: c(person("Wei-Chen", "Chen", role = c("aut", "cre"), email =
         "[email protected]"), person("George", "Ostrouchov", role = "aut"))
@@ -11,20 +11,21 @@ LazyLoad: yes
 LazyData: yes
 Description: The pmclust aims to utilize model-based clustering (unsupervised)
         for high dimensional and ultra large data, especially in a distributed
-        manner. The package employs pbdMPI to perform a
+        manner. The package employs Rmpi to perform a
         expectation-gathering-maximization (EGM) algorithm
-        for finite mixture Gaussian models. The unstructured dispersion
-        matrices are assumed in the Gaussian models. The implementation is
-        default in the single program multiple data (SPMD) programming model.
-        The code can be executed through pbdMPI and independent to most MPI
-        applications. See the High Performance Statistical Computing (HPSC)
-        website for more information, documents and examples.
+        for finite mixture Gaussian
+        models. The unstructured dispersion matrices are assumed in the Gaussian
+        models. The implementation is default in the single program multiple
+        data (SPMD) programming model. The code can be executed through Rmpi and
+        independent to most MPI applications. See the High Performance
+        Statistical Computing (HPSC) website for more information, documents
+        and examples.
 License: GPL (>= 2)
 URL: http://r-pbd.org/
 BugReports: http://group.r-pbd.org/
 MailingList: Please send questions and comments regarding pbdR to
         [email protected]
-Packaged: 2013-05-31 14:21:37 UTC; snoweye
+Packaged: 2013-06-19 18:32:45 UTC; snoweye
 Author: Wei-Chen Chen [aut, cre],
   George Ostrouchov [aut]
 Maintainer: Wei-Chen Chen <[email protected]>

R/dmat_em_base.r

@@ -20,42 +20,42 @@ update.expectation.dmat <- function(PARAM, update.logL = TRUE){
   K <- PARAM$K
 
   ### WCC: original
-  .pmclustEnv$U.dmat <- sweep(.pmclustEnv$W.dmat, 2, PARAM$log.ETA, FUN = "+")
+  # .pmclustEnv$U.dmat <- sweep(.pmclustEnv$W.dmat, 2, PARAM$log.ETA, FUN = "+")
   ### WCC: temp dmat
   # tmp.1 <- sweep(.pmclustEnv$W.dmat, 2, PARAM$log.ETA, FUN = "+")
   # .pmclustEnv$U.dmat <- tmp.1
   ### WCC: temp spmd
-  # tmp.1 <- as.matrix(.pmclustEnv$W.dmat)
-  # tmp.2 <- sweep(tmp.1, 2, PARAM$log.ETA, FUN = "+")
-  # .pmclustEnv$U.dmat <- as.ddmatrix(tmp.2)
+  tmp.1 <- as.matrix(.pmclustEnv$W.dmat)
+  tmp.2 <- sweep(tmp.1, 2, PARAM$log.ETA, FUN = "+")
+  .pmclustEnv$U.dmat <- as.ddmatrix(tmp.2)
 
   ### WCC: original
-  .pmclustEnv$Z.dmat <- exp(.pmclustEnv$U.dmat)
+  # .pmclustEnv$Z.dmat <- exp(.pmclustEnv$U.dmat)
   ### WCC: temp dmat
   # tmp.1 <- exp(.pmclustEnv$U.dmat)
   # .pmclustEnv$Z.dmat <- tmp.1
   ### WCC: temp spmd
-  # tmp.1 <- as.matrix(.pmclustEnv$U.dmat)
-  # tmp.2 <- exp(tmp.1)
-  # .pmclustEnv$Z.dmat <- as.ddmatrix(tmp.2)
+  tmp.1 <- as.matrix(.pmclustEnv$U.dmat)
+  tmp.2 <- exp(tmp.1)
+  .pmclustEnv$Z.dmat <- as.ddmatrix(tmp.2)
 
   ### WCC: original
-  tmp.id <- rowSums(.pmclustEnv$U.dmat < .pmclustEnv$CONTROL$exp.min) == K |
-            rowSums(.pmclustEnv$U.dmat > .pmclustEnv$CONTROL$exp.max) > 0
-  tmp.id <- as.logical(as.vector(tmp.id))
+  # tmp.id <- rowSums(.pmclustEnv$U.dmat < .pmclustEnv$CONTROL$exp.min) == K |
+  #           rowSums(.pmclustEnv$U.dmat > .pmclustEnv$CONTROL$exp.max) > 0
+  # tmp.id <- as.logical(as.vector(tmp.id))
   ### WCC: temp dmat
-  # tmp.1 <- .pmclustEnv$U.dmat < .pmclustEnv$CONTROL$exp.min
-  #     tmp.1 <- as.matrix(tmp.1)
-  # tmp.2 <- rowSums(tmp.1)
-  # tmp.3 <- tmp.2 == K
-  # tmp.4 <- .pmclustEnv$U.dmat > .pmclustEnv$CONTROL$exp.max
-  #     tmp.4 <- as.matrix(tmp.4)
-  # tmp.5 <- rowSums(tmp.4)
-  # tmp.6 <- tmp.5 > 0
-  # tmp.7 <- tmp.3 | tmp.6
-  # tmp.8 <- as.vector(tmp.7)
-  # tmp.id <- tmp.8
-  # tmp.id <- as.logical(tmp.id)
+  tmp.1 <- .pmclustEnv$U.dmat < .pmclustEnv$CONTROL$exp.min
+      tmp.1 <- as.matrix(tmp.1)
+  tmp.2 <- rowSums(tmp.1)
+  tmp.3 <- tmp.2 == K
+  tmp.4 <- .pmclustEnv$U.dmat > .pmclustEnv$CONTROL$exp.max
+      tmp.4 <- as.matrix(tmp.4)
+  tmp.5 <- rowSums(tmp.4)
+  tmp.6 <- tmp.5 > 0
+  tmp.7 <- tmp.3 | tmp.6
+  tmp.8 <- as.vector(tmp.7)
+  tmp.id <- tmp.8
+  tmp.id <- as.logical(tmp.id)
 
   tmp.flag <- sum(tmp.id)
   if(tmp.flag > 0){

R/dmat_em_initial.r

@@ -24,7 +24,13 @@ initial.em.dmat <- function(PARAM, MU = NULL){
 # -1.3357516, 0.5336209, 1.2700404,
 # -1.3110521, 0.2632600, 1.7063794
 # )
-# PARAM$MU <- matrix(PARAM$MU, nrow = 4)
+PARAM$MU <- c(
+  0.2328901,  0.7281706, -1.0026075,
+ -0.6238820, -0.2584021,  0.8862744,
+  0.4944164,  0.7662080, -1.2993873,
+  0.4674663,  0.7461591, -1.2516524
+)
+PARAM$MU <- matrix(PARAM$MU, nrow = 4)
 
   e.step.dmat(PARAM)
   PARAM <- em.onestep.dmat(PARAM)

R/dmat_em_logdmvnorm.r

@@ -39,6 +39,15 @@ logdmvnorm.dmat <- function(PARAM, i.k){
     tmp.3 <- as.matrix(.pmclustEnv$W.dmat)
     tmp.3[, i.k] <- tmp.2
     .pmclustEnv$W.dmat <- as.ddmatrix(tmp.3)
+
+    # B <- sweep(X.dmat, 2, PARAM$MU[, i.k])
+    # C <- backsolve(U, diag(1, PARAM$p))
+    # B <- B %*% as.ddmatrix(C)
+    # distval <- rowSums(B * B)
+    # distval <- as.vector(distval)
+    # .pmclustEnv$W.dmat[, i.k] <- -(.pmclustEnv$p.times.logtwopi + logdet +
+    #                                distval) * 0.5
+
 #  }
   invisible()
 } # End of logdmvnorm.dmat().

R/pm_em_initial.r

@@ -55,7 +55,13 @@ initial.em.spmd <- function(PARAM, MU = NULL){
 # -1.3357516, 0.5336209, 1.2700404,
 # -1.3110521, 0.2632600, 1.7063794
 # )
-# PARAM$MU <- matrix(PARAM$MU, nrow = 4)
+PARAM$MU <- c(
+  0.2328901,  0.7281706, -1.0026075,
+ -0.6238820, -0.2584021,  0.8862744,
+  0.4944164,  0.7662080, -1.2993873,
+  0.4674663,  0.7461591, -1.2516524
+)
+PARAM$MU <- matrix(PARAM$MU, nrow = 4)
 
   e.step.spmd(PARAM)
   PARAM <- em.onestep.spmd(PARAM)