find cutoff by simulation

src/find_by_simulation.R

@@ -0,0 +1,55 @@
+# check the BlockMissing() function, inspect the missing patter
+library(multicore)
+# HeatStruct(BlockMissing(), high.col = "darkgreen", missing.col = "red") 
+
+n.sim <-100  # number of simulations
+set.seed(100)
+sim.cutoff <- vector("list", n.sim)
+for (i in 1:n.sim){
+  sim.cutoff[[i]] <- BlockMissing()
+  sim.cutoff[[i]]$date <- date.month[325:900]
+}
+
+
+
+Rmse <- function(cut.value = 0.75, ...){
+  # 
+  com <- mclapply(sim.cutoff, FUN = Cut, cutoff = cut.value)  
+  RMSE <- numeric(n.sim)
+
+  for (i in seq_len(n.sim)){
+    err <- as.matrix(com[[i]] - complete.chunk)
+    n <- nmissing(sim.cutoff[[i]])
+    RMSE[i] <- sqrt(sum(err^2) / n)
+  }
+  return(RMSE)
+}
+
+Rmse()
+# try
+# set.seed(100)
+# mean(cutoff.rmse(0.75))
+cut <- seq(0.45, 0.95, by=0.001)
+try <- mapply(cutoff.rmse, cut.value = cut)
+cutoff.cv <- apply(try, 2, mean)
+cutoff.sd <- apply(try, 2, sd)
+
+x <- as.data.frame(cbind(cut, cutoff.cv))
+x <- transform(x, ymax =  cutoff.cv + cutoff.sd, ymin = cutoff.cv - cutoff.sd)
+limits <- aes(ymax=ymax, ymin=ymin)  # for errorbar plot
+minimal <- subset(x, cutoff.cv == min(cutoff.cv))
+p <- ggplot(x, aes(x = cut, y = cutoff.cv)) + geom_line() +
+  geom_point(data = minimal, size = 3, colour = "red") +
+  geom_errorbar(limits, col = "grey", lty = 5)
+p <- p + geom_abline(intercept = 0.6433, slop = 0, lty = 2, size = 0.9, colour = "darkgreen")
+p <- p + xlab("CUTOFF value") + ylab("Mean-RMSE") + theme_grey(base_size = 12)
+p <- p + annotate("text", x = 0.68, y = 0.66, label= "RMSE(SVD) = 0.6433", col= "darkgreen")
+p + scale_x_continuous(breaks = seq(0.45, 0.95, by = 0.05))
+# p <- p + annotate("text", x = 0.65, y = 0.545, label= "min-RMSE(CUTOFF) = 0.5367", col= "blue")
+# p <- p+annotate("text", x= 0.763, y = 0.53, label="0.763")
+# p <- p+annotate("rect", xmin = 0.768, xmax = 0.9, ymin = 0.537, ymax = 0.59, alpha=0.2)
+p
+
+
+
+