Blog

Playing with Sonar {mlbench}

A bit of gymnastics in RStudio, R Markdown and knitr.

This example also lives in an open repository in GitHub: https://github.com/Eclectikus/playingsonar

knitr quick reference: knitr in a knutshell - Author’s site

mlbench: Machine Learning Benchmark Problems -CRAN- A collection of artificial and real-world machine learning benchmark problems, including, e.g., several data sets from the UCI repository.

Usage (remember install library first: install.packages("mlbench"))

library('mlbench')
data("Sonar")

Format:

A data frame with 208 observations on 61 variables, all numerical and one (the Class) nominal.

Below, a couple of samples of this data frame:

b <- sample(c(1:208), size = 9, replace = FALSE, prob = NULL)
kable(head(Sonar[b, 1:10]), format = "markdown")

kable(head(Sonar[b, 52:61]), format = "markdown")

Physical base

Exploratory amusements

After loading libraries and data,

## Libraries&data
    library(mlbench)
    library(ggplot2)
    library(ggthemes)
    library(caret)
    library('knitr')
    data("Sonar")

and creating training and testing sets,

## Train&test
  set.seed(107)
  inTrain <- createDataPartition(y = Sonar$Class, p = .75, list = FALSE)
  training <- Sonar[ inTrain,]
  testing <- Sonar[-inTrain,]

I made a quick approach to data by running a few graphical routines on previous, arbitrary, training set.

# 
for(i in names(training)){   # loop columns
  a <- ggplot(data = training, aes(training[,i],y = Class, coord_fixed(),
                                   col = Class))
  a <- a + geom_point(size=3)
  a <- a + xlab("Signal") + ylab("Mine or Rock") 
  a <- a + ggtitle("Signal by spectral sample and outcome (mine or rock)")
  a <- a + theme_solarized(light = FALSE)+scale_colour_solarized("blue")
  a <- a + xlim(0,1)
  a <- a + annotate("text", x = 0.50, y = 1.2, label = "Mine", colour = "white")
  a <- a + annotate("text", x = 0.50, y = 2.2, label = "Rock", colour = "white")
  a <- a + annotate("text", x = 0.94, y = 2.4, label = i, colour = "yellow", size = 7.5)
                    blue.bold.text <- element_text(face="bold", color="white")
  a <- a + theme(axis.title = blue.bold.text, legend.position="none")
## plot to file
WD0 <- "C:/MyDATA/Routput/SONAR1"   # (use your local disk)
filename=paste(i,".png",sep="")
ggsave(filename, plot = a, path=WD0,width = 16, height = 9, dpi = 120)
print(a)
}

Signal by spectral sample and “outcome” (mine or rock)

And also,

for (j in 1:157) {  # loop observations
      Xvalues <- as.data.frame(1:60)
      Yvalues <- as.data.frame(t(training[j,1:60])) # transpose, row to column
      Signal <- cbind(Xvalues,Yvalues) # two-columns data for observation 'j'
    colnames(Signal) <- c("Spectral sample", "Normalized energy")
    a <- ggplot(data = Signal, aes(x = Xvalues, y = Yvalues, coord_fixed()))
    a <- a + geom_line(lty = 1, color = "gray", size = 0.3, na.rm = TRUE)
    a <- a + xlab("Spectral sample") + ylab("Normalized energy") 
    a <- a + ggtitle("Signal by spectral sample")
    a <- a + theme_solarized(light = FALSE) + scale_colour_solarized("blue")
    # a <- a + theme_base() + scale_colour_few()
    a <- a + xlim (0,60) + ylim(0,1)
        ## outcomes by color
          if (training[j,61] == "M") {
              ColorCode <- "blue"
              outcome <- "Mine"
          }
          else {
              ColorCode <- "orange"
              outcome <- "Rock"
          }
    a <- a + annotate("text", x = 5, y = 0.92, label = outcome , colour = ColorCode, size = 5.4)
    a <- a + annotate("text", x = 55, y = 0.92, label = paste("Return", j),
                      colour = "gray", size = 4.5)
    blue.bold.text <- element_text(face="bold", color="white")
    a <- a + theme(axis.title = blue.bold.text, legend.position="none")
    a <- a + geom_point(size = 1.5, color = ColorCode)
    a <- a + geom_smooth(fill = ColorCode, color = ColorCode, size = 0.75,
                         alpha = 0.30, na.rm = TRUE)
    ## plot to file
      WD0 <- "C:/MyDATA/Routput/SONAR2"   # (use your local disk)
      filename=paste("S",j,".png",sep="")
      ggsave(filename, plot = a, path = WD0, width = 16, height = 9, dpi = 120)
      print(a)
}

Signal by return and “outcome” (mine or rock)

I tried first to do the above animations directly from R by using package animation (from the same author as knitr), but got nothing, surely I need more time. Its Github and the package site are sure the starting points. After that, I tried Gimp but the poor quality made me desist. So finally I happily used ezgif.com