Data 550: Data Visualization I

Lecture 4b: Comparing Distributions in R

Dr. Irene Vrbik

University of British Columbia Okanagan

Introduction

• Up until this point we have provided examples mostly in Altair with the understanding that ggplot has a similar counterpart.

• As Altair is relatively new, and ggplot2 is one of the most widely used and documented packages in R, it does have functionalities that Altair has yet to implement.

• One such example is violin plots.

Learning Outcomes

• Create density, box plots, and violin plots using ggplot

We covered the underlying concepts last lecture so this is meant to be more of a quick demo and walk through.

Data

Below is the reprocessed movies data frame (to see how it was processed see the accompanying ipynb)

# the above is the cleaned version
library(rjson)
library(tidyverse)
movies <- fromJSON(file = 'data/lec-movies.json') %>%
    as_tibble() %>%
    unnest(-c(countries, genres))
    
head(movies)

preprocessing unnested/exploded the data set such that each row only contains one countries and one genres. (if movies fell in two genres for example, it appears in the table twice: once under US, once under the UK)

Histogram

Let’s recall how to make a histogram.

ggplot(movies, aes(x = runtime)) +
    geom_histogram(color = 'white')

Density plot

Unlike Altair, ggplot has it’s own density mark, …

ggplot(movies, aes(x = runtime)) +
    geom_density(fill = 'grey', alpha = 0.7)

• this is much less work than in Altair where we had to transform
• These are called composite marks (in Altair) or stat marks in ggplot since then do a calculation before plotting.
• in this case it’s a density transform + a lined geom

Unnesting the data

We need to unnest/explode on genres and countires.

free_genres <- movies %>% unnest(genres)
free_countries <- movies  %>%  unnest(countries)
free_both <- movies %>% unnest(genres) %>%  unnest(countries)

free_genres %>% 
  filter(, title ==  "All Dogs Go to Heaven") %>% 
  select(genres, countries)

free_both %>% 
  filter(title ==  "All Dogs Go to Heaven") %>% 
  select(genres, countries)

Layered Density Plot

ggplot(free_genres, aes(x = runtime,
        fill = genres,
        color = genres)) +
    geom_density(alpha = 0.6)

Notice how you can add the aesthetic rather than including it as an argument within ggplot():

ggplot(free_genres) +
    aes(x = runtime,
        fill = genres,
        color = genres) +
    geom_density(alpha = 0.6)

• talk about why this needs to be done with free_genre instead of free_both
• To color/fill by a variable, we don’t need a separate groupby, but can treat geom_density like any other geom, which is convenient. In dplyr, a list can be unpacked via the unnest function.

Layered Density Plot

Faceting

ggplot(free_both) +
    aes(x = runtime, fill = genres, color = genres) +
    geom_density(alpha = 0.6) +
    facet_wrap(~countries)

Faceting (row and column)

ggplot(free_both, show.legend = FALSE) +
    aes(x = runtime, fill = genres, color = genres) +
    geom_density(alpha = 0.6) +
    facet_grid(countries~genres)

unlike Altair, ggplot extends the KDE so that the dist looks smooth

Boxplots

As in Altair, ggplot unsuprisingly has a boxplot geom, eg.

ggplot(free_both) +
    aes(x = runtime, y = genres,fill = genres) +
    geom_boxplot()

Scaled Boxplots

As in Altair, ggplot unsuprisingly has a boxplot geom, eg.

ggplot(free_both) +
    aes(x = runtime, y = genres,fill = genres) +
    geom_boxplot(varwidth = TRUE)

Violin Plots

The change from boxplot to violin is extremely simple

ggplot(free_both) +
    aes(x = runtime, y = genres,fill = genres) +
    geom_violin()

What are violin plots

• Violin plots are similar to box plots, except that they also show the kernel probability density of the data at different values.

• Typically, violin plots will include a marker for the median of the data and a box indicating the interquartile range, as in standard box plots.

• The function geom_violin() is used to produce a violin plot.

Violin vs Faceted Density Plots

ggplot(free_both) +
    aes(x = runtime, 
        y = genres,
        fill = genres) +
    geom_violin()

ggplot(free_both) +
    aes(x = runtime, fill = genres, color = genres) +
    geom_density(alpha = 0.6) +
    facet_wrap(~genres, ncol = 1)

mirros the KDE and allows us to use the Y-axis for some categorical variable

Faceted Boxplots

As with out density plots, we can also facet by country, eg.

ggplot(free_both) +
    aes(x = runtime, y = genres,fill = genres) +
    geom_boxplot() +
    facet_wrap(~countries)

Violin Plots

To get the violin plots, we simply change the geom:

ggplot(free_both) +
    aes(x = runtime, y = genres,fill = genres) +
    geom_violin() +
    facet_wrap(~countries)

Layering Quanties

We can layer the quantiles shown in the box plots

ggplot(free_both) +
    aes(x = runtime, y = genres, fill = genres) +
    geom_violin(draw_quantiles = c(0.25, 0.5, 0.75)) +
    facet_wrap(~countries)

Comments

• When possible, it is a good idea to have a look at where the individual data points are.

• Of course we could always layer on different marking of our data (using geom_point() for example)

• However when we have a lot of data, this could be impossible to read.

• For this we can use a categorical scatter plot where the dots are spread/jittered¹ randomly on the non-value axis so that they don’t all overlap via geom_jitter().

• similar discussions in KDE plotting, (since violin plots are just mirrored KDEs it makes sense the pitfalls would carry over)

1. jittering is not yet available in Altair

Layering Points

We can layer the points onto the violin plots:

ggplot(free_both) +
    aes(x = runtime, y = genres, fill = genres) +
    geom_violin() + geom_point() +
    facet_wrap(~countries)

• we saw how we can used tickes in Altair but a better option (which is not yet available in Altair) is to jitter your points

Jittering Data

“jittering” adds some noise to the location of each point

ggplot(free_both) +
    aes(x = runtime, y = genres, fill = genres) +
    geom_violin() + geom_jitter() +
    facet_wrap(~countries)

and is a useful way of handling overplotting caused by discreteness in smaller datasets.

Order matters

We can change the default height and order or layers

ggplot(free_both) +
    aes(x = runtime, y = genres, fill = genres) +
    geom_jitter(height = 0.2, alpha = 0.3) + geom_violin() + 
    facet_wrap(~countries)

• argument for this configuration is that violin plots work well when you have many points (so you don’t have to show those points) but not as well where you have nto so many points.
• height makes the rectangle smaller and alpha sets the opacity

Unfaceting

Rather than faceting we could fill by countries

ggplot(free_both) +
    aes(x = runtime, y = genres, fill = countries) + 
    geom_violin(draw_quantiles = c(0.25, 0.5, 0.75))

now we’ve seen multiple channels by which to represent categorical data when comparing distributions:

• colour
• facet
• y-axis