Search the parameter space using parallel processing and plot the heatmap

gridsearch_basic

# source: http://statcompute.wordpress.com/2013/06/01/grid-search-for-free-parameters-with-parallel-computing/
library(MASS)
data(Boston)
X <- I(as.matrix(Boston[-14]))
st.X <- scale(X)
Y <- I(as.matrix(Boston[14]))
boston <- data.frame(X = st.X, Y)

# DIVIDE THE WHOLE DATA INTO TWO SEPARATE SETS
set.seed(2013)
rows <- sample(1:nrow(boston), nrow(boston) - 200)
set1 <- boston[rows, ]
set2 <- boston[-rows, ]

# LOAD FOREACH PACKAGE
library(doParallel)
registerDoParallel(cores = 8)
library(foreach)

# GRID SEARCH BASED ON THE MINIMUM SSE WITH PARALLEL COMPUTING
cv.sse <- 
  foreach(b = seq(0, 10, 2), .combine = rbind) %dopar% {
    library(foreach)
    foreach(n = 1:10, .combine = rbind) %dopar% {  
      # TRAIN A PROJECTION PURSUIT REGRESSION WITH VARIOUS SETTINGS AND TRAINING DATA
      ppreg <- ppr(Y ~ X, data = set1, nterms = n, sm.method = "supsmu", bass = b)
      # CALCULATE SSE WITH VALIDATION DATA
      test.sse <- sum((set2$Y - predict(ppreg, set2))^2)
      data.frame(bass = b, nterms = n, sse = test.sse)
  }
}
# PRINT OUT THE BEST SETTING BASED ON THE GRID SEARCH
print(best.setting <- cv.sse[cv.sse$sse == min(cv.sse$sse), ]) 

# OUTPUT WITH THE LOWEST SSE BY GRID SEARCH #
#    bass nterms     sse
# 17    2      7 2126.07

# GENERATE A HEAT MAP TO VISUALIZE THE GRID SEARCH OUTCOME 
library(ggplot2)
bass <- factor(cv.sse$bass)
nterms <- factor(cv.sse$nterms)
sse <- factor(floor(cv.sse$sse / 100) * 100)
jpeg('cv.jpeg', width = 800, height = 500, quality = 100)
qplot(x = bass, y = nterms, fill = sse, geom = 'tile')
dev.off()