Want more content like this? Subscribe here to be notified of new releases!

Data science tools

Tutorial on how to make beautiful plots with R

Star

R ggplot2 ggrepel gganimate ggspatial sf

Motivation

The Department of Transportation publicly released a dataset that lists flights that occurred in 2015 along with specificities such as delays, flight time and other information. Our previous post detailed the best practices to manipulate data.

This article aims at showing good practices to visualize data using R's most popular libraries. The following are covered:

plots using ggplot2 along with customized visualizations with ggrepel
animated plots using gganimate
map-based plots with the sf and ggspatial libraries with using data coming from maps and rnaturalearth

library(ggplot2)                         # Plots
library(ggrepel)                         # Nice labels
library(gganimate)                       # Animations
library(ggspatial);library(sf)           # Map plots
library(maps);library(rnaturalearth)     # Map data
theme_set(theme_bw())                    # Set theme for all plots

A common strategy to make beautiful plots is to have data ready in a data frame and use the columns and rows of that data frame to plot. This tutorial will focus on the plotting part. As to get the data in the specified format, please refer to the associated tutorial and study guide.

Temporal plots

Evolution of number of flights

We want to plot the temporal evolution of flights of major US airlines:

In order to do that, we note the following:

The geom_line() layer takes in a long-format data frame which plots the different temporal lines for each airline.
The legend text is customized using the theme() function, for which an extensive description of the API can be found here.
The vertical line is drawn using the geom_vline() layer and by indicating the line type with the linetype argument.
The appearance of the date labels on the $x$ axis is customized with the scale_x_date() layer.

The code used to produce the plot is shown below.

# data (date, airline, nb_flights)
ggplot(data = data) +
  geom_line(aes(x = date, y = nb_flights, color = airline)) +
  geom_vline(xintercept = as.Date('2015-07-01'),
             linetype = 'dashed', alpha = 0.5, color = 'black') +
  scale_x_date(date_breaks = '1 month', date_labels = '%b %d') +
  labs(x = 'Time', y = 'Number of daily flights', color = 'Airline',
       title = 'Number of flights in the US in 2015',
       subtitle = 'Top airlines',
       caption = 'Source: publicly available data from DoT') +
  theme(axis.text.x = element_text(angle = 25, vjust = 0.75),
        plot.caption = element_text(vjust = 7)) +
  ylim(0, 4000)

Evolution of traffic per airport

We want to plot temporal evolution of traffic in the top Hawaiian airports:

In order to do that, we note the following:

Time periods have been drawn using the geom_rect() layer and by indicating the start and end date of each time period of interest.
Colors can be customized through the scale_fill_manual() function.

The code used to produce the plot is shown below.

# data (date, airport, nb_flights), time_periods (xmin, xmax, ymin, ymax, name, color)
ggplot(data = data) +
  geom_rect(data = time_periods, alpha = 0.1,
            aes(xmin = xmin, xmax = xmax, ymin = ymin, ymax = ymax, fill = name)) +
  geom_line(aes(x = date, y = nb_flights, color = airport)) +
  scale_x_date(date_breaks = '1 month', date_labels = '%b %d') +
  scale_fill_manual(breaks = time_periods[,'name'], values = time_periods[,'color']) +
  labs(x = 'Time', y = 'Number of daily flights', color = 'Airport', fill = 'Time window',
       title = 'Number of flights departing from Hawaiian airports in 2015',
       subtitle = 'Top 5 Hawaii airports',
       caption = 'Source: publicly available data from DoT') +
  theme(axis.text.x = element_text(angle = 25, vjust = 0.75), plot.caption = element_text(vjust = 7)) +
  ylim(0, 150)

Evolution of delay type

We want to plot the temporal evolution of delays by delay type for the northern part of the US that are more prone to weather-related delays:

In order to do that, we note the following:

The geom_bar() layer along with the position parameter set to 'stack' gives a stacked bar chart that represents various kinds of delays cumulatively for a given time period.
The labels and date_labels parameters within the scale_y_continuous() and scale_x_date() functions enable to customize the appearance of labels.

The code used to produce the plot is shown below.

# data (week, delay_type, perc)
ggplot(data = data) +
  geom_bar(aes(x = week, y = perc, fill = delay_type),
           position = 'stack', stat = 'identity') +
  scale_x_date(date_breaks = '1 month', date_labels = '%b %d') +
  scale_y_continuous(labels = function(x) paste0(x*100, '%')) +
  labs(x = 'Time', y = 'Percentage of weekly flights', fill = 'Delay type',
       title = 'Breakdown of delay type of flights in the northern part of the US in 2015',
       subtitle = 'States that are part of the analysis include AK, IL, IN, MA, ME, MI, MN, NH, NY, VT',
       caption = 'Source: publicly available data from DoT') +
  theme(axis.text.x = element_text(angle = 25, vjust = 0.75), plot.caption = element_text(vjust = 7))

Visualizing across multiple dimensions

Delay by time of the week for top airlines

We want to plot the percentage of delayed flights by hour of the day and day of the week for the top airlines:

In order to do that, we note the following:

The facet_grid() layer has been used to produce a plot that is distinct for each day of the week. In an equivalent way, we could also use the facet_wrap() function and set the number of columns ncol to 7 to have the same looking result.

The code used to produce the plot is shown below.

# data (day_of_week, hour, airline, perc_delay), time_periods (xmin, xmax, ymin, ymax, name, color)
ggplot(data = data) +
  geom_rect(data = time_periods, alpha = 0.2,
            aes(xmin = xmin, xmax = xmax, ymin = ymin, ymax = ymax, fill = name)) +
  geom_line(aes(x = hour, y = perc_delay, color = airline), size = 0.5) +
  geom_point(aes(x = hour, y = perc_delay, color = airline), size = 0.7) +
  facet_grid(~day_of_week) +
  scale_x_continuous(limits = c(0,24), breaks = 8*c(0:3), labels = function(x){
    case_when(x == 0 ~ '', x == 8 ~ '8am', x == 16 ~ '4pm', x == 24 ~ '12am')
    }) +
  labs(x = 'Time', y = 'Percentage of delayed flights', color = 'Airline',
       title = 'Percentage of delayed flight by time of the week averaged across 2015',
       subtitle = 'Top US airlines', caption = 'Source: publicly available data from DoT') +
  theme(plot.caption = element_text(vjust = 7)) +
  guides(fill = guide_legend(title = NULL))

Delay by hour of day for top airlines

We can also plot the same data from a different angle with a heat map:

In order to do that, we note the following:

The geom_tile() layer plots the heatmap of percentage of delayed flights by hour of the day for each airline, by taking in the x, y and fill arguments.
The appearance of the plot is customized within the theme() layer. In particular, the background grid is removed with the element_blank() function.

The code used to produce the plot is shown below.

ggplot(data = data) +
  geom_tile(aes(x = hour, y = airline, fill = perc_delay), color = 'black') +
  scale_x_continuous(breaks = c(5,11,17,22), labels = function(x){
    case_when(x == 5 ~ '5am', x == 11 ~ '11am', x == 17 ~ '5pm', x == 22 ~ '10pm')
  }) +
  scale_fill_distiller(palette = 'Spectral', labels = function(x) paste0(x*100, '%')) +
  labs(x = 'Hour of day', y = 'Airline', fill = 'Delayed flights',
       title = 'Percentage of delayed flights by hour of the day',
       subtitle = 'Averaged across 2015 for top US airlines',
       caption = 'Source: publicly available data from DoT') +
  theme(panel.grid.major = element_blank(), axis.ticks = element_blank(),
        plot.caption = element_text(vjust = 7), axis.title.y = element_blank())

Customizing visualization

Scatter plot of airports with delays

We want to plot:

In order to do that, we note the following:

The text labels are put in place using the geom_label_repel(), which comes from the ggrepel library.
The size and color options are used to customize the appearance of the scatter plot.

The code used to produce the plot is shown below.

# data (airport_code, region, nb_flights, perc_delay, weather)
ggplot(data = data) +
  geom_point(aes(x = weather, y = perc_delay, color = region, size = nb_flights)) +
  geom_label_repel(data = data %>% head(5), nudge_x = 0.002, nudge_y = -0.015,
                   aes(x = weather, y = perc_delay, label = airport_code)) +
  scale_x_continuous(labels = function(x) paste0(x*100, '%')) +
  scale_y_continuous(labels = function(y) paste0(y*100, '%')) +
  labs(x = 'Likelihood weather is causing delay', y = 'Percentage of delayed flights',
       color = 'Region', size = 'Volume of traffic',
       title = 'Percentage of delayed flights vs likelihood of weather causing the delay',
       subtitle = 'Top US airports across 2015', caption = 'Source: publicly available data from DoT') +
  theme(plot.caption = element_text(vjust = 7))

Animated graphs

We want ot have an animated graph to understand the temporal evolution:

In order to do that, we note the following:

The transition_states() layer coming from the gganimate library enables to produce the animation along a given dimension. In this example, the animation is produced along the month variable.

The code used to produce the plot is shown below.

# data (airport_code, region, month, nb_flights, perc_delay, weather)
ggplot(data = data) +
  geom_point(aes(x = weather, y = perc_delay, color = region, size = nb_flights)) +
  scale_x_continuous(labels = function(x) paste0(x*100, '%')) +
  scale_y_continuous(labels = function(y) paste0(y*100, '%')) +
  scale_size(labels = scales::comma) +
  labs(x = 'Likelihood weather is causing delay', y = 'Percentage of delayed flights',
       color = 'Region', size = 'Number of outbound flights',
       title = 'Percentage of delayed flights vs likelihood of weather causing the delay',
       subtitle = paste0('Top US airports in ', '{closest_state}', ' 2015'),
       caption = 'Source: publicly available data from DoT') +
  theme(plot.caption = element_text(vjust = 7)) +
  transition_states(month, state_length = 0)

Percentage of delayed flights with volume

We want to plot:

In order to do that, we note the following:

The second $y$ axis is produced by using the sec.axis() function within the scale_y_continuous() layer.
The color of the axes labels are specified within the theme() layer.
Dates on the $x$ axis are displayed in a customized way using the date_labels argument of the scale_x_date() layer.

The code used to produce the plot is shown below.

# data
ggplot(data = data) +
  geom_line(aes(x = date, y = nb_flights), color = 'green4') +
  geom_line(aes(x = date, y = perc_delay * 20000), color = 'red') +
  scale_x_date(date_breaks = '1 month', date_labels = '%b %d') +
  scale_y_continuous(sec.axis = sec_axis(~ . / 20000,
                                         labels = function(y){paste0(y*100, '%')},
                                         name = 'Percentage of delayed flights')) +
  labs(x = 'Time', y = 'Number of daily flights', caption = 'Source: publicly available data from DoT',
       title = 'Number of flights in the US in 2015') +
  theme(axis.text.x = element_text(angle = 25, vjust = 0.75), plot.caption = element_text(vjust = 7),
        axis.title.y.left = element_text(color = 'green4'),
        axis.title.y.right = element_text(color = 'red'))

Map-based plots

Top 10 routes

We want to plot the top routes departing from Boston airports on a map:

In order to do that, we note the following:

The map of the US and its neighboring countries has been drawn using geom_sf() from polygon shapes using the ne_countries() function coming from the rnaturalearth library.
Curves between cities have been drawn using the geom_curve() layer and specifiying each side of the curve using the x, y, xend, yend parameters.
The annotation_north_arrow() and annotation_scale() functions come from the ggspatial library and provide nice add-ons to the plot by indicating the North direction as well as how distances should be interpreted. Other tutorials making use of similar functions can give you an idea of the range of possibilities that such functionalities can offer.
The coord_sf() layer is used to customize the range of the axes that are shown on the plot.

The code used to produce the plot is shown below.

# world, usa, routes
ggplot() +
  geom_sf(data = world) +
  geom_sf(data = usa, fill = 'chartreuse1', alpha = 0.05) +
  geom_curve(data = routes, aes(x = longitude_o, y = latitude_o, xend = longitude_d, yend = latitude_d)) +
  geom_point(data = routes, aes(x = longitude_d, y = latitude_d), size = 3) +
  geom_point(data = routes, aes(x = longitude_o, y = latitude_o), size = 3, color = 'red') +
  geom_label_repel(data = routes, nudge_x = 0, nudge_y = -0,
                   aes(x = longitude_d, y = latitude_d, label = destination_airport)) +
  geom_label_repel(data = routes %>% select(origin_airport, longitude_o, latitude_o) %>% unique(),
                   nudge_x = 0, nudge_y = -0, color = 'red',
                   aes(x = longitude_o, y = latitude_o, label = origin_airport)) +
  annotation_north_arrow(location = "bl", which_north = "true", style = north_arrow_fancy_orienteering,
                         pad_x = unit(0.25, "in"), pad_y = unit(0.25, "in")) +
  annotation_scale(location = 'bl', width_hint = 0.5) +
  coord_sf(xlim = c(-125, -64), ylim = c(24, 50)) +
  labs(title = 'Top 10 routes departing from Boston Logan International Airport in 2015',
       caption = 'Source: publicly available data from DoT') +
  theme(panel.background = element_rect(fill = 'azure'), axis.title = element_blank())

Most popular airline per state

We want to plot a map that displays the most popular airline by state in the continental US:

In order to do that, we note the following:

The maps library has been used to retrieve geographical data that is then converted to an sf format.
It is then fed into the geom_sf() layer from the ggspatial library to draw polygon shapes.
The coord_sf() layer is used to customize the range of the axes that are shown on the plot.

The code used to produce the plot is shown below.

# data (state, airline, long, lat, group, order)
ggplot() +
  geom_sf(data = data, aes(fill = airline), color = 'black') +
  coord_sf(xlim = c(-125, -67), ylim = c(24, 50)) +
  labs(title = 'Most popular airline per state', subtitle = 'Data averaged across 2015',
       caption = 'Source: publicly available data from DoT', fill = 'Airline') +
  theme_void() +
  theme(plot.title = element_text(hjust = 0.5), plot.subtitle = element_text(hjust = 0.5),
        axis.title = element_blank())

Airports with highest traffic

We want to plot a map that shows the top 10 busiest airports in the western US:

In order to do that, we note the following:

The map itself has been drawn using polygon shapes, for which the data was retrieved using the states data from the maps library, that has then been converted to an sf object using the st_as_sf() function, and then fed into the geom_sf() layer which comes from the ggspatial library.
The text labels have been put in place using the geom_label_repel(), which comes from the ggrepel library.

The code used to produce the plot is shown below.

# data
ggplot() +
  geom_sf(data = west, color = 'black', alpha = 0.1, show.legend = FALSE, aes(fill = ID)) +
  geom_point(data = data, aes(x = long, y = lat, size = nb_flights)) +
  geom_label_repel(data = data, nudge_x = 0.2, nudge_y = -0.015,
                   aes(x = long, y = lat, label = airport_code, color = rank)) +
  geom_text(data = west %>% select(longitude, latitude, abbrev) %>% unique(),
            aes(x = longitude, y = latitude, label = abbrev)) +
  labs(title = 'Top 10 busiest airports in western US', subtitle = 'Data averaged across 2015',
       caption = 'Source: publicly available data from DoT', size = 'Number of outbound flights') +
  scale_color_manual(labels = paste0(data$rank, ' - ', data$airport_name, ' (', data$airport_code, ')'),
                     name = 'Top airports', values = rep('black', 10)) +
  scale_size(labels = scales::comma) +
  theme_void() +
  theme(plot.title = element_text(hjust = 0.5), plot.subtitle = element_text(hjust = 0.5)) +
  guides(color = guide_legend(override.aes = list(size = 0)))

Conclusion

This tutorial presented illustrated examples that showed how to produce visually-appealing plots. This skill is useful when wanting to make a plot that is clear and concise, for instance in a business setting or for high stake presentations.

Tutorial on how to make beautiful plots with R

Motivation

Temporal plots

Evolution of number of flights

Evolution of traffic per airport

Evolution of delay type

Visualizing across multiple dimensions

Delay by time of the week for top airlines

Delay by hour of day for top airlines

Customizing visualization

Scatter plot of airports with delays

Animated graphs

Percentage of delayed flights with volume

Map-based plots

Top 10 routes

Most popular airline per state

Airports with highest traffic

Conclusion

You may also like...