update
authorBenjamin Auder <benjamin@auder>
Fri, 20 Sep 2019 13:22:30 +0000 (15:22 +0200)
committerBenjamin Auder <benjamin@auder>
Fri, 20 Sep 2019 13:22:30 +0000 (15:22 +0200)
reports/accuracy.R
reports/genfig.R [new file with mode: 0644]
reports/run_accu_cl.sh

index 2f35792..2381524 100644 (file)
@@ -2,7 +2,7 @@ optimBeta <- function(N, n, K, p, beta, b, link, ncores)
 {
        library(morpheus)
        res <- multiRun(
-               list(n=n,p=p,beta=beta,b=b,optargs=list(K=K,link=link)),
+               list(n=n, p=p, beta=beta, b=b, optargs=list(K=K, link=link)),
                list(
                        # morpheus
                        function(fargs) {
@@ -11,25 +11,39 @@ optimBeta <- function(N, n, K, p, beta, b, link, ncores)
                                M <- computeMoments(fargs$X, fargs$Y)
                                fargs$optargs$M <- M
                                mu <- computeMu(fargs$X, fargs$Y, fargs$optargs)
-                               op <- optimParams(K,link,fargs$optargs)
-                               x_init <- list(p=rep(1/K,K-1), beta=mu, b=rep(0,K))
-                               do.call(rbind, op$run(x_init))
-                       },
-                       # flexmix
-                       function(fargs) {
-                               library(flexmix)
-                               source("../patch_Bettina/FLXMRglm.R")
-                               K <- fargs$optargs$K
-                               dat <- as.data.frame( cbind(fargs$Y,fargs$X) )
-                               fm <- flexmix( cbind(V1, 1-V1) ~ .-V1, data=dat, k=K,
-                                       model = FLXMRglm(family = binomial(link = link)) )
-                               p <- mean(fm@posterior[["scaled"]][,1])
-                               out <- refit(fm)
-                               beta_b <- sapply( seq_len(K), function(i) {
-                                       as.double( out@components[[1]][[i]][,1] )
-                               } )
-                               rbind(p, beta_b[2:nrow(beta_b),], beta_b[1,])
-                       } ),
+                               res2 <- NULL
+                               tryCatch({
+                                       op <- optimParams(K,link,fargs$optargs)
+                                       x_init <- list(p=rep(1/K,K-1), beta=mu, b=rep(0,K))
+                                       res2 <- do.call(rbind, op$run(x_init))
+                               }, error = function(e) {
+                                       res2 <- NA
+                               })
+                               res2
+                       }
+#                      ,
+#                      # flexmix
+#                      function(fargs) {
+#                              library(flexmix)
+#                              source("../patch_Bettina/FLXMRglm.R")
+#                              K <- fargs$optargs$K
+#                              dat <- as.data.frame( cbind(fargs$Y,fargs$X) )
+#                              res2 <- NULL
+#                              tryCatch({
+#                                      fm <- flexmix( cbind(V1, 1-V1) ~ .-V1, data=dat, k=K,
+#                                              model = FLXMRglm(family = binomial(link = link)) )
+#                                      p <- mean(fm@posterior[["scaled"]][,1])
+#                                      out <- refit(fm)
+#                                      beta_b <- sapply( seq_len(K), function(i) {
+#                                              as.double( out@components[[1]][[i]][,1] )
+#                                      } )
+#                                      res2 <- rbind(p, beta_b[2:nrow(beta_b),], beta_b[1,])
+#                              }, error = function(e) {
+#                                      res2 <- NA
+#                              })
+#                              res2
+#                      }
+               ),
                prepareArgs = function(fargs, index) {
                        library(morpheus)
                        io = generateSampleIO(fargs$n, fargs$p, fargs$beta, fargs$b, fargs$optargs$link)
@@ -40,8 +54,14 @@ optimBeta <- function(N, n, K, p, beta, b, link, ncores)
                        fargs
                }, N=N, ncores=ncores, verbose=TRUE)
        p <- c(p, 1-sum(p))
-       for (i in 1:2)
+       for (i in 1:length(res)) {
+               for (j in N:1) {
+                       if (is.null(res[[i]][[j]]) || is.na(res[[i]][[j]]))
+                               res[[i]][[j]] <- NULL
+               }
+               print(paste("Count valid runs for ",i," = ",length(res[[i]]),sep=""))
                res[[i]] <- alignMatrices(res[[i]], ref=rbind(p,beta,b), ls_mode="exact")
+       }
        res
 }
 
@@ -97,4 +117,4 @@ mr <- optimBeta(N, n, K, p, beta, b, link, ncores)
 mr_params <- list("N"=N, "n"=n, "K"=K, "d"=d, "link"=link,
        "p"=c(p,1-sum(p)), "beta"=beta, "b"=b)
 
-save("mr", "mr_params", file=paste("multirun_",d,"_",link,".RData",sep=""))
+save("mr", "mr_params", file=paste("multirun_",n,"_",d,"_",link,".RData",sep=""))
diff --git a/reports/genfig.R b/reports/genfig.R
new file mode 100644 (file)
index 0000000..2edd546
--- /dev/null
@@ -0,0 +1,45 @@
+nvals <- c(5000,10000,100000,500000,1000000)
+for (link in c("logit","probit"))
+{
+       for (d in c(2,5,10))
+       {
+               par(mfrow=c(2,2), lwd=2, cex.axis=2, cex=2)
+               res <- list()
+               for (n in nvals)
+               {
+                       load(paste("multirun_",n,"_",d,"_",link,".RData",sep=""))
+                       res <- c(res, mr)
+               }
+               for (i in 1:2) #our, fm
+               {
+                       for (j in 1:2) #beta, p,b
+                       {
+                               if (j == 1) #beta
+                               {
+                                       for (dim in 1:d)
+                                       {
+                                               ypts <- c()
+                                               for (k in 1:5)
+                                                       ypts <- c(ypts, mean(sapply(1:length(res[[k]][[i]]), function(x) mean((res[[k]][[i]][[x]][2:(d+1),dim] - mr_params$beta[,dim])^2))))
+                                               plot(nvals, ypts)
+                                               if (dim < d)
+                                                       par(new=TRUE)
+                                       }
+                               }
+                               else #p + b
+                               {
+                                       for (rowidx in c(1,d+2))
+                                       {
+                                               ypts <- c()
+                                               ref <- if (rowidx==1) { mr_params$p } else { mr_params$b }
+                                               for (k in 1:5)
+                                                       ypts <- c(ypts, mean(sapply(1:length(res[[k]][[i]]), function(x) mean((res[[k]][[i]][[x]][rowidx,] - ref)^2))))
+                                               plot(nvals, ypts)
+                                               if (rowidx==1)
+                                                       par(new=TRUE)
+                                       }
+                               }
+                       }
+               }
+       }
+}
index d1d8e37..50b2844 100644 (file)
@@ -17,3 +17,11 @@ for d in 2 5 10 20; do
                R --slave --args N=1000 n=1e5 nc=15 d=$d link=$link <accuracy.R >out$d$link 2>&1
        done
 done
+
+#for d in 2 5; do
+#      for n in 5000 10000 100000 500000 1000000; do
+#              for link in "logit" "probit"; do
+#                      R --slave --args N=1000 n=$n nc=64 d=$d link=$link <accuracy.R >out_$n$link$d 2>&1
+#              done
+#      done
+#done