Making Plots With plotnine

Overview

Teaching: 40 min
Exercises: 50 min
Questions
  • How can I visualize data in Python?

  • What is ‘grammar of graphics’?

Objectives
  • Create a plotnine object.

  • Set universal plot settings.

  • Modify an existing plotnine object.

  • Change the aesthetics of a plot such as color.

  • Edit the axis labels.

  • Build complex plots using a step-by-step approach.

  • Create scatter plots, box plots, and time series plots.

  • Use the facet_wrap and facet_grid commands to create a collection of plots splitting the data by a factor variable.

  • Create customized plot styles to meet their needs.

Disclaimer

Python has powerful built-in plotting capabilities such as matplotlib, but for this episode, we will be using the plotnine package, which facilitates the creation of highly-informative plots of structured data based on the R implementation of ggplot2 and The Grammar of Graphics by Leland Wilkinson. The plotnine package is built on top of Matplotlib and interacts well with Pandas.

Just as with the other packages, plotnine need to be imported. It is good practice to not just load an entire package such as from plotnine import *, but to use an abbreviation as we used pd for Pandas:

%matplotlib inline
import plotnine as p9

From now on, the functions of plotnine are available using p9.. For the exercise, we will use the surveys.csv data set, with the NA values removed

import pandas as pd

surveys_complete = pd.read_csv('data/surveys.csv')
surveys_complete = surveys_complete.dropna()

Plotting with plotnine

The plotnine package (cfr. other packages conform The Grammar of Graphics) supports the creation of complex plots from data in a dataframe. It uses default settings, which help creating publication quality plots with a minimal amount of settings and tweaking.

plotnine graphics are built step by step by adding new elementsadding different elements on top of each other using the + operator. Putting the individual steps together in brackets () provides Python-compatible syntax.

To build a plotnine graphic we need to:

(p9.ggplot(data=surveys_complete))

As we have not defined anything else, just an empty figure is available and presented.

As we have not defined anything else, just an empty figure is available and presented.

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight', y='hindfoot_length')))

The most important aes mappings are: x, y, alpha, color, colour, fill, linetype, shape, size and stroke.

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight', y='hindfoot_length'))
    + p9.geom_point()
)

The + in the plotnine package is particularly useful because it allows you to modify existing plotnine objects. This means you can easily set up plot templates and conveniently explore different types of plots, so the above plot can also be generated with code like this:

# Create
surveys_plot = p9.ggplot(data=surveys_complete,
                         mapping=p9.aes(x='weight', y='hindfoot_length'))

# Draw the plot
surveys_plot + p9.geom_point()

png

Challenge - bar chart

Working on the surveys_complete data set, use the plot-id column to create a bar-plot that counts the number of records for each plot. (Check the documentation of the bar geometry to handle the counts)

Answers

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='plot_id'))
    + p9.geom_bar()
)

png

Notes:

Building your plots iteratively

Building plots with plotnine is typically an iterative process. We start by defining the dataset we’ll use, lay the axes, and choose a geom. Hence, the data, aes and geom-* are the elementary elements of any graph:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight', y='hindfoot_length'))
    + p9.geom_point()
)

Then, we start modifying this plot to extract more information from it. For instance, we can add transparency (alpha) to avoid overplotting:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight', y='hindfoot_length'))
    + p9.geom_point(alpha=0.1)
)

png

We can also add colors for all the points

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight', y='hindfoot_length'))
    + p9.geom_point(alpha=0.1, color='blue')
)

png

Or to color each species in the plot differently, map the species_id column to the color aesthetic:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight',
                          y='hindfoot_length',
                          color='species_id'))
    + p9.geom_point(alpha=0.1)
)

png

Apart from the adaptations of the arguments and settings of the data, aes and geom-* elements, additional elements can be added as well, using the + operator:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight', y='hindfoot_length', color='species_id'))
    + p9.geom_point(alpha=0.1)
    + p9.xlab("Weight (g)")
)

png

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight', y='hindfoot_length', color='species_id'))
    + p9.geom_point(alpha=0.1)
    + p9.xlab("Weight (g)")
    + p9.scale_x_log10()
)

png

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight', y='hindfoot_length', color='species_id'))
    + p9.geom_point(alpha=0.1)
    + p9.xlab("Weight (g)")
    + p9.scale_x_log10()
    + p9.theme_bw()
    + p9.theme(text=p9.element_text(size=16))
)

png

Challenge - Bar plot adaptations

Adapt the bar plot of the previous exercise by mapping the sex variable to the color fill of the bar chart. Change the scale of the color fill by providing the colors blue and orange manually (see API reference to find the appropriate function).

Answers

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='plot_id',
                          fill='sex'))
    + p9.geom_bar()
    + p9.scale_fill_manual(["blue", "orange"])
)

png

Plotting distributions

Visualizing distributions is a common task during data exploration and analysis. To visualize the distribution of weight within each species_id group, a boxplot can be used:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='species_id',
                          y='weight'))
    + p9.geom_boxplot()
)

png

By adding points of the individual observations to the boxplot, we can have a better idea of the number of measurements and of their distribution:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='species_id',
                          y='weight'))
    + p9.geom_jitter(alpha=0.2)
    + p9.geom_boxplot(alpha=0.)
)

png

Challenge - distributions

Boxplots are useful summaries, but hide the shape of the distribution. For example, if there is a bimodal distribution, this would not be observed with a boxplot. An alternative to the boxplot is the violin plot (sometimes known as a beanplot), where the shape (of the density of points) is drawn.

In many types of data, it is important to consider the scale of the observations. For example, it may be worth changing the scale of the axis to better distribute the observations in the space of the plot.

  • Replace the box plot with a violin plot, see geom_violin()
  • Represent weight on the log10 scale, see scale_y_log10()
  • Add color to the datapoints on your boxplot according to the plot from which the sample was taken (plot_id)

Hint: Check the class for plot_id. By using factor() within the aes mapping of a variable, plotnine will handle the values as category values.

Answers

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='species_id',
                          y='weight',
                          color='factor(plot_id)'))
    + p9.geom_jitter(alpha=0.3)
    + p9.geom_violin(alpha=0, color="0.7")
    + p9.scale_y_log10()
)

png

Plotting time series data

Let’s calculate number of counts per year for each species. To do that we need to group data first and count the species (species_id) within each group.

yearly_counts = surveys_complete.groupby(['year', 'species_id'])['species_id'].count()
yearly_counts

When checking the result of the previous calculation, we actually have both the year and the species_id as a row index. We can reset this index to use both as column variable:

yearly_counts = yearly_counts.reset_index(name='counts')
yearly_counts

Timelapse data can be visualised as a line plot (geom_line) with years on x axis and counts on the y axis.

(p9.ggplot(data=yearly_counts,
           mapping=p9.aes(x='year',
                          y='counts'))
    + p9.geom_line()
)

Unfortunately this does not work, because we plot data for all the species together. We need to tell plotnine to draw a line for each species by modifying the aesthetic function and map the species_id to the color:

(p9.ggplot(data=yearly_counts,
           mapping=p9.aes(x='year',
                          y='counts',
                          color='species_id'))
    + p9.geom_line()
)

png

Faceting

As any other library supporting the Grammar of Graphics, plotnine has a special technique called faceting that allows to split one plot into multiple plots based on a factor variable included in the dataset.

Consider our scatter plot of the weight versus the hindfoot_length from the previous sections:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight',
                          y='hindfoot_length',
                          color='species_id'))
    + p9.geom_point(alpha=0.1)
)

We can now keep the same code and at the facet_wrap on a chosen variable to split out the graph and make a separate graph for each of the groups in that variable. As an example, use sex:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight',
                          y='hindfoot_length',
                          color='species_id'))
    + p9.geom_point(alpha=0.1)
    + p9.facet_wrap("sex")
)

png

We can apply the same concept on any of the available categorical variables:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight',
                          y='hindfoot_length',
                          color='species_id'))
    + p9.geom_point(alpha=0.1)
    + p9.facet_wrap("plot_id")
)

png

The facet_wrap geometry extracts plots into an arbitrary number of dimensions to allow them to cleanly fit on one page. On the other hand, the facet_grid geometry allows you to explicitly specify how you want your plots to be arranged via formula notation (rows ~ columns; a . can be used as a placeholder that indicates only one row or column).

# only select the years of interest
survey_2000 = surveys_complete[surveys_complete["year"].isin([2000, 2001])]

(p9.ggplot(data=survey_2000,
           mapping=p9.aes(x='weight',
                          y='hindfoot_length',
                          color='species_id'))
    + p9.geom_point(alpha=0.1)
    + p9.facet_grid("year ~ sex")
)

png

Challenge - facetting

Create a separate plot for each of the species that depicts how the average weight of the species changes through the years.

Answers

yearly_weight = surveys_complete.groupby([‘year’, ‘species_id’])[‘weight’].mean().reset_index()

(p9.ggplot(data=yearly_weight,
           mapping=p9.aes(x='year',
                          y='weight'))
    + p9.geom_line()
    + p9.facet_wrap("species_id")
)

Challenge - facetting

Based on the previous exercise, visually compare how the weights of male and females has changed through time by creating a separate plot for each sex and an individual color assigned to each species_id.

Answers

yearly_weight = surveys_complete.groupby([‘year’, ‘species_id’, ‘sex’])[‘weight’].mean().reset_index()

(p9.ggplot(data=yearly_weight, mapping=p9.aes(x=’year’, y=’weight’, color=’species_id’)) + p9.geom_line() + p9.facet_wrap(“sex”) )

Further customization

As the syntax of plotnine follows the original R package ggplot2, the documentation of ggplot2 can provide information and inspiration to customize graphs. Take a look at the ggplot2 cheat sheet, and think of ways to improve the plot. You can write down some of your ideas as comments in the Etherpad.

The theming options provide a rich set of visual adaptations. Consider the following example of a bar plot with the counts per year.

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='factor(year)'))
    + p9.geom_bar()
)

png

Notice that we use the year here as a categorical variable by using the factor functionality. However, by doing so, we have the individual year labels overlapping with each other. The theme functionality provides a way to rotate the text of the x-axis labels:

(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='factor(year)'))
    + p9.geom_bar()
    + p9.theme_bw()
    + p9.theme(axis_text_x = p9.element_text(angle=90))
)

png

When you like a specific set of theme-customizations you created, you can save them as an object to easily apply them to other plots you may create:

my_custom_theme = p9.theme(axis_text_x = p9.element_text(color="grey", size=10,
                                                         angle=90, hjust=.5),
                           axis_text_y = p9.element_text(color="grey", size=10))
(p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='factor(year)'))
    + p9.geom_bar()
    + my_custom_theme
)

png

Challenge - customization

Please take another five minutes to either improve one of the plots generated in this exercise or create a beautiful graph of your own.

Here are some ideas:

After creating your plot, you can save it to a file in your favourite format. You can easily change the dimension (and its resolution) of your plot by adjusting the appropriate arguments (width, height and dpi):

my_plot = (p9.ggplot(data=surveys_complete,
           mapping=p9.aes(x='weight', y='hindfoot_length'))
    + p9.geom_point()
)
my_plot.save("scatterplot.png", width=10, height=10, dpi=300)

Key Points

  • The data, aes variables and a geometry are the main elements of a plotnine graph

  • With the + operator, additional scale_*, theme_*, xlab/ylab and facet_* elements are added