Project: Bicycle Sharing

18.6 How long?

Your Turn: Make a box-and-whisker plot, like Figure 18.2 showing the distribution of the duration of rental events, broken down by the client type. The duration of the rental can be calculated as:

as.numeric(edate - sdate)

The units will be in either hours, minutes, or seconds. It should not be much trouble for you to figure out which one.

Figure 18.2: The distribution of bike-rental durations as a box-and-whisker plot.

When you make your plot, you will likely find that the axis range is being set by a few outliers. These may be bikes that were forgotten. Arrange your scale to ignore these outliers, or filter them out.

18.7 When are bikes used?

The sdate variable in Trips indicates the date and time of day that the bicycle was checked out of the parking station. sdate is stored as a special

Often, you will want discrete components of a date, for instance:

Your Turn: Make histograms or density plots of each of these discrete components. Explain what each plot is showing about how bikes are checked out. For example, Figure 18.3 shows that few bikes are checked out before 5am, and that there are busy times around the rush hour: 8am and 5pm.

Figure 18.3: Distribution of bike trips by hour of the day.

A similar sort of display of events per hour can be accomplished by calculating and displaying each hour’s count, as in Figure 18.4.

Figure 18.4: The number of events in each hour of the day. Compare to the scale in Figure 18.3. Why are the scales different?

The graphic shows a lot of variation of bike use over the course of the day. Now consider two additional variables: the day of the week and the “client” type.

Your Turn: Group the bike rentals by three variables: hour of the day, day of the week, and client type. Find the total number of events in each grouping and plot this count versus hour. Use the group aesthetic to represent one of the other variables and faceting to represent the other. [Click to see note.]Hint: utilize facet_wrap() in the plotting commands.

Your Turn: Make the same sort of display of how bike rental vary of hour, day of the week, and client type, but use geom_density() rather than grouping and counting. Compare the two displays — one of discrete counts and one of density — and describe any major differences.

18.8 How far?

Find the distance between each pair of stations. You know the position from the lat and long variables in Stations. This is enough information to find the distance. The calculation has been implemented in the haversine() function:

source("https://mdbeckman.github.io/dcSupplement/R/haversine.R")

haversine() is a transformation function. To use it, create a data table where a case is a pair of stations and there are variables for the latitude and longitude of the starting station and the ending station. To do this, join the Station data to itself. The following statements show how to create appropriately named variables for joining.

Simple <-
  Stations %>%
  select(name, lat, long) %>%
  rename(sstation = name)

Simple2 <-
  Simple %>%
  rename(estation = sstation, lat2 = lat, long2 = long)

Look at the head() of Simple and Simple2 and make sure you understand how they are related to Stations.

The joining of Simple and Simple2 should match every station to every other station. [Click to see note.]Since a ride can start and end at the same station, it also makes sense to match each station to itself. This sort of matching does not make use of any matching variables; everything is matched to everything else. This is called a full outer join. [Click to see note.]A full outer join matches every case in the left table to each and every case in the right table.

First, try the full outer join of just a few cases from each table, for example 4 from the left and 3 from the right.

merge(head(Simple, 4), head(Simple2, 3), by = NULL)

Make sure you understand what the full outer join does before proceeding. For instance, you should be able to predict how many cases the output will have when the left input has \(n\) cases and the right has \(m\) cases.

• There are 347 cases in the Stations data table. How many cases will there be in a full outer join of Simple to Simple2?

It’s often impractical to carry out a full outer join. For example, joining BabyNames to itself with a full outer join will generate a result with more than three-trillion cases. [Click to see note.]Three trillion cases from the BabyNames data is the equivalent of about 5 million hours of MP3 compressed music. A typical human lifespan is about 0.6 million hours.

Perform the full outer join and then use haversine() to compute the distance between each pair of stations.

StationPairs <- merge(Simple, Simple2, by = NULL)

Check your result for sensibility. Make a histogram of the station-to-station distances and explain where it looks like what you would expect. (Hint: you could use the Internet to look up the distance from one end of Washington, D.C. to the other.)

PairDistances <-
  StationPairs %>%
  mutate(distance = haversine(lat, long, lat2, long2)) %>%
  select(sstation, estation, distance)

Once you have PairDistances, you can join it with Trips to calculate the start-to-end distance of each trip. [Click to see note.]Of course, a rider may not go directly from one station to another.

• Look at the variables in Stations and Trips and explain why Simple and Simple2 were given different variable names for the station.

An inner_join() is appropriate for finding the distance of each ride. [Click to see note.]Watch out! the Trips data and the PairDistances data are large enough that the join is expensive: it takes about a minute.

Your turn: Display the distribution of the ride distances of the rides. Compare it to the distances between pairs of stations. Are they similar? Why or why not?

Figure 18.5: The distribution of trip lengths compared to the distribution of distances between pairs of stations (shaded).

18.9 Mapping the Stations

You can draw detailed maps with the leaflet package. You may need to install it.

devtools::install_github("rstudio/leaflet")

leaflet works much like ggplot() but provides special facilities for maps. Here’s how to make the simple map:

library(leaflet)
stationMap <-
  leaflet(Stations) %>%           # like ggplot()
  addTiles() %>%                  # add the map
  addCircleMarkers(radius = 2, color = "red") %>%
  setView(-77.04, 38.9, zoom = 12)

To display the map, use the object name as a command: stationMap

18.10 Long-distance stations

Your Turn (more challenging, but cool): Around each station on the map, draw a circle whose radius reflects the median distance covered by rentals starting at that station. To draw the circles, use the same leaflet commands as before, but add in a line like this.

addCircles(radius = ~ mid, color = "blue", opacity = 0.0001)

For addCircles() to draw circles at the right scale, the units of the median distance should be presented in meters rather than kilometers. This will create too much overlap, unfortunately. So, set the radius to be half or one-third the median distance in meters. From your map, explain the pattern you see in the relationship between station location and median distance.