0

I've written code to plot density data for variations of an A/B test. I'd like to improve the visual by shading (with the fill being slightly transparent) the area below each curve. I'm currently using matplot, but understand ggplot might be a better option.

Any ideas? Thanks.

# Setup data frame - these are results from an A/B experiment
conv_data = data.frame(
            VarNames = c("Variation 1", "Variation 2", "Variation 3") # Set variation names
            ,NumSuccess = c(1,90,899) # Set number of successes / conversions
            ,NumTrials = c(10,100,1070) # Set number of trials
            )
conv_data$NumFailures = conv_data$NumTrials - conv_data$NumSuccess # Set number of failures [no conversions]
num_var = NROW(conv_data) # Set total number of variations
plot_col = rainbow(num_var) # Set plot colors

get_density_data <- function(n_var, s, f) {
    x = seq(0,1,length.out=100) # 0.01,0.02,0.03...1
    dens_data = matrix(data = NA, nrow=length(x), ncol=(n_var+1))
    dens_data[,1] = x

    # set density data
    for(j in 1:n_var) {
        # +1 to s[], f[] to ensure uniform prior
        dens_data[,j+1] = dbeta(x, s[j]+1, f[j]+1)
    }
    return(dens_data)
}

density_data = get_density_data(num_var, conv_data$NumSuccess, conv_data$NumFailures)

matplot(density_data[,1]*100, density_data[,-1], type = "l", lty = 1, col = plot_col, ylab = "Probability Density", xlab = "Conversion Rate %", yaxt = "n")
legend("topleft", col=plot_col, legend = conv_data$VarNames, lwd = 1)

This produces the following plot: enter image description here

cometrico
  • 31
  • 1
  • 6
  • I wrote a [blog post about shading regions under a curve](http://www.fromthebottomoftheheap.net/2013/01/11/shading-regions-under-a-curve/) like this. In your case you want to fill the entire region so it simplifies it a bit. Essentially, you want to use `polygon()` to draw the shaded area, then perhaps draw the lines on top. If I get chance and no one has answered this by the time I get home, I'll provide a proper answer. – Gavin Simpson Apr 07 '15 at 19:56
  • Gavin, great blog post. I answered my own question by switching to ggplot, but your post also answered my question through a different means. If you post as answer, I'll mark your response accepted. – cometrico Apr 07 '15 at 20:00

2 Answers2

1
# Setup data frame - these are results from an A/B experiment
conv_data = data.frame(
  VarNames = c("Variation 1", "Variation 2", "Variation 3") # Set variation names
  ,NumSuccess = c(1,90,899) # Set number of successes / conversions
  ,NumTrials = c(10,100,1070) # Set number of trials
)
conv_data$NumFailures = conv_data$NumTrials - conv_data$NumSuccess # Set number of failures [no conversions]
num_var = NROW(conv_data) # Set total number of variations
plot_col = rainbow(num_var) # Set plot colors

get_density_data <- function(n_var, s, f) {
  x = seq(0,1,length.out=100) # 0.01,0.02,0.03...1
  dens_data = matrix(data = NA, nrow=length(x), ncol=(n_var+1))
  dens_data[,1] = x

  # set density data
  for(j in 1:n_var) {
    # +1 to s[], f[] to ensure uniform prior
    dens_data[,j+1] = dbeta(x, s[j]+1, f[j]+1)
  }
  return(dens_data)
}

density_data = get_density_data(num_var, conv_data$NumSuccess, conv_data$NumFailures)

matplot(density_data[,1]*100, density_data[,-1], type = "l",
        lty = 1, col = plot_col, ylab = "Probability Density",
        xlab = "Conversion Rate %", yaxt = "n")
legend("topleft", col=plot_col, legend = conv_data$VarNames, lwd = 1)

## and add this part
for (ii in seq_along(plot_col))
  polygon(c(density_data[, 1] * 100, rev(density_data[, 1] * 100)),
          c(density_data[, ii + 1], rep(0, nrow(density_data))),
          col = adjustcolor(plot_col[ii], alpha.f = .25))

enter image description here

rawr
  • 20,481
  • 4
  • 44
  • 78
0

Was able to answer own question with:

df = as.data.frame(t(conversion_data))
dfs = stack(df)
ggplot(dfs, aes(x=values)) + geom_density(aes(group=ind, colour=ind, fill=ind), alpha=0.3)
cometrico
  • 31
  • 1
  • 6