Split dataframe into two groups

Question

I've simulated this data.frame :

library(plyr); library(ggplot2)
count <- rev(seq(0, 500, 20))
tide <- seq(0, 5, length.out = length(count))
df <- data.frame(count, tide)

count_sim <- unlist(llply(count, function(x) rnorm(20, x, 50)))
count_sim_df <- data.frame(tide=rep(tide,each=20), count_sim)

And it can be plotted like this:

ggplot(df, aes(tide, count)) + geom_jitter(data = count_sim_df, aes(tide, count_sim), position = position_jitter(width = 0.09)) + geom_line(color = "red")

I now want to split count_sim_df into two group: high and low . When I plot the split count_sim_df , it should look like this (everything in green and blue is photoshopped). The bit that I'm finding tricky is getting overlap between high and low around the middle values of tide .

This is how I want to split count_sim_df into high and low:

• assign half of count_sim_df to high and half of count_sim_df to low
• reassign the values of count to create overlap between high and low around the middle values of tide

Answer 1

Here's my revised suggestion. I hope it helps.

middle_tide <- mean(count_sim_df$tide)
hilo_margin <- 0.3
middle_df <- subset(count_sim_df,tide > (middle_tide * (1 - hilo_margin)))
middle_df <- subset(middle_df, tide < (middle_tide * (1 + hilo_margin)))
upper_df <- count_sim_df[count_sim_df$tide > (middle_tide * (1 + hilo_margin)),]
lower_df <- count_sim_df[count_sim_df$tide < (middle_tide * (1 - hilo_margin)),]
idx <- sample(2,nrow(middle_df), replace = T)
count_sim_high <- rbind(middle_df[idx==1,], upper_df)
count_sim_low <- rbind(middle_df[idx==2,], lower_df)
p <- ggplot(df, aes(tide, count)) + 
   geom_jitter(data = count_sim_high, aes(tide, count_sim), position = position_jitter(width = 0.09), alpha=0.4, col=3, size=3) + 
   geom_jitter(data = count_sim_low, aes(tide, count_sim), position = position_jitter(width = 0.09), alpha=0.4, col=4, size=3) + 
   geom_line(color = "red")

I might still not have fully understood your procedure to split into high and low, especially what you mean by "reassigning the value of count". In this case here I have defined an overlap region of 30% around the middle value of tide and assigned randomly half of the points within this transition region to the "high" and the other half to the "low" set.

Answer 2

Here's a way to generate the sample dataset and the groupings using relatively little code and just base R:

library(ggplot2)
count <- rev(seq(0, 500, 20))
tide <- seq(0, 5, length.out = length(count))
df <- data.frame(count, tide)

count_sim_df <- data.frame(tide = rep(tide,each=20),
                           count = rnorm(20 * nrow(df), rep(count, each = 20), 50))
margin <- 0.3
count_sim_df$`tide level` <-
  with(count_sim_df,
    factor((tide >= quantile(tide, 0.5 + margin / 2) |
           (tide >= quantile(tide, 0.5 - margin / 2) & sample(0:1, length(tide), TRUE))),
           labels = c("Low", "High")))
ggplot(df, aes(x = tide, y = count)) +
  geom_line(colour = "red") +
  geom_point(aes(colour = `tide level`), count_sim_df, position = "jitter") +
  scale_colour_manual(values = c(High = "green", Low = "blue"))