A workflow for creating a plot that shows sleep times and wake times as a bar for each day. Reference plot: https://pbs.twimg.com/media/Cl1Bk-JUgAAZrEX.jpg

sleep_time.R

# Setup
options(stringsAsFactors = FALSE)
library(lubridate)
library(ggplot2)



################################################################################
# Simulating fake data - this is just for my benefit, really
################################################################################

# Number of days to simulate
n_days <- 100

# Simulate sleep times first - starting with an arbitrary time and adding some
# noise with rnorm()
sleepytimes <- data.frame(start_sleep = seq(from = as.POSIXct("2016-06-25 22:30"), 
                                            length.out = n_days, 
                                            by = "1 day") +
                                        rnorm(n = n_days, sd = 60*30)
)

# I'm also going to add a row for going to sleep after midnight, to make sure
# the code works for that scenario
sleepytimes <- rbind(
    sleepytimes,
    data.frame(start_sleep = as.POSIXct(paste(max(date(sleepytimes$start_sleep)) + 2,
                                              "02:17:23")))
)



# Add hours and minutes of sleep... in a mildly absurd way. First, generate
# random, decimal hours of sleep:
sleepytimes$sleep_hours <- rnorm(n = n_days + 1, mean = 8)

# Now: subtract the whole hours and convert the decimal part to minutes
sleepytimes$sleep_minutes <- with(sleepytimes, 
                                  round((sleep_hours - floor(sleep_hours)) * 60)
)

# Then go back and round the decimal hours down to whole hours
sleepytimes$sleep_hours <- floor(sleepytimes$sleep_hours)





################################################################################
# The code you'll actually want to test on your data
################################################################################



# To calculate waking times, calculate a duration() from your hours and minutes,
# and add it to your sleep start time:
sleepytimes$end_sleep <- with(sleepytimes,
    start_sleep + duration(hours = sleep_hours, minutes = sleep_minutes)
)




# Now, to plot... you're totally right that crossing zero is going to be a
# problem if we treat it like a datetime. So I think the easiest thing to do
# subtract midnight from the sleep and wake times...

# First, I'm going to try to associate every sleep period with a particular date,
# even if sleep didn't start until after midnight. Rule of thumb: if sleep
# starts before 12pm on Tuesday, it's counted for Monday night
sleepytimes$night_of <- with(sleepytimes,
                             
    # The ifelse() function returns integers instead of Dates for some reason,
    # so gotta convert back to Dates
    as.Date(
        
        # Sleep started before 12pm? If yes, use previous date
        ifelse(hour(start_sleep) > 12,
               yes = date(start_sleep),
               no = date(start_sleep) - 1),
        
        origin = "1970-01-01")

)


# Now I'll use the difftime() function to calculate hours until/hours since
# midnight for all of the sleep start times - these will usually be positive:
sleepytimes$start_sleep_diff <- with(sleepytimes,
    as.numeric(
        difftime(time1 =as.POSIXct(paste((night_of + 1), "00:00:00")),
                 time2 = start_sleep,
                 units = "hours")
    )
)


# And again for all of the wake times - these will usually be negative:
sleepytimes$end_sleep_diff <- with(sleepytimes,
    as.numeric(
        difftime(time1 = as.POSIXct(paste((night_of + 1), "00:00:00")),
                 time2 = end_sleep,
                 units = "hours")
        )
)



# Now to plot it.


# You might want to tweak this to get the bar spacing just right. 0.1 looks
# pretty good as a starting point:
bar_spacing <- 0.1


# You can tweak the smoothness of the two trend lines by fiddling with this
# parameter. 0.1 makes a good start here, too:
smooth_span <- 0.1


# To get actual time labels, we need a function that will take our y values
# and return actual times. 
label_hours <- function(x) {
    
    require(lubridate)
    
    # Start with midnight, then subtract the "time to midnight" variables - 
    # just reversing what we did before, basically, then formatting
    # to show just the hours (%I) and AM/PM indicator (%p)
    format(as.POSIXct(paste(Sys.Date(), "00:00:00")) - hours(x), "%I:00 %p")
    
}




# Plot it!
ggplot(sleepytimes) +
    
    # geom_rect needs all four corners of the rectangle. xmin and xmax use the
    # night_of variable (which is a Date):
    geom_rect(aes(xmin = night_of + bar_spacing, 
                  xmax = night_of + (1 - bar_spacing),
                  
    # ymin and ymax use the "time to midnight" variables:
                  ymin = start_sleep_diff,
                  ymax = end_sleep_diff),
              fill = "#2b8cbe") +
    

    # Adding smoothed regression lines. 
    geom_smooth(aes(x = night_of, y = start_sleep_diff), 
                method = loess, method.args = list(span = smooth_span),
                se = FALSE, color = "#045a8d") +
    
    geom_smooth(aes(x = night_of, y = end_sleep_diff), 
                method = loess, method.args = list(span = smooth_span),
                se = FALSE, color = "#045a8d") +
    
    # Labels
    labs(x = "Date",
         y = "Time",
         main = "Sleep and Wake Times") +
    
    # Use scale_y_continuous to apply the label_hours function we made
    scale_y_continuous(labels = label_hours) +
    
    # A nicer theme than the default
    theme_bw()
    

Lesson learned: when naming datasets, think carefully about how much you'll regret a whimsical choice after typing it 300 times.