Visualising Data

Jon Reades - j.reades@ucl.ac.uk

1st October 2025

Start with Chart (Part 2)

Building on Week 7, here are some deeper links between models and visualistions:

• What are we trying to model?
• What are we trying to penalise?
• What isn’t fitting into the model?

Inspired by this: https://towardsdatascience.com/when-averages-lie-moving-beyond-single-point-predictions-23201e8c04c8/

Link it to the following kinds of issues:

What are We Modelling?

Issues of skew and leverage. It’s not (so much) about normality per se with OLS.

What are We Penalising?

What isn’t Fitting?

Choices, Choices…

1. matplotlib: the ‘big beast’ of visualisation in Python. Similar to MATLAB. Highly customisable. Very complex.
2. seaborn: a layer that sits over top of matplotlib and makes it easier to produce good-quality graphics.
3. bokeh: web-based visualisation tool that can integrate with Jupyter or output to static HTML files.
4. plotly: another web-based visualisation tool that can integrate with Jupyter.

More emerging all the time: Vega/Altair, HoloViews, etc.

If you’re really wedded to ggplot, plotnine is a clone of ggplot’s interface (Grammer of Graphics) in Python. A brief overview of visualisation libraries could be helpful.

Seaborn

Designed to provide ggplot-like quality output using matplotlib:

• Improve on default colourmaps and colour defaults.
• Integration with pandas data frames (Note: not geopandas!).
• Offers more plot types out of the box.
• Still offers access to matplotlib’s back-end.

Plot Types

Partial Overview of Seaborn Plots

For the fuller overview see Overview of seaborn plotting functions and the full API reference.

In Practice

import seaborn as sns
sns.set_theme(style="darkgrid")
fmri = sns.load_dataset("fmri")
sns.lineplot(x="timepoint", y="signal",
             hue="region", style="event",
             data=fmri)

In Practice 2

sns.set_theme(style="whitegrid", palette="muted")
df = sns.load_dataset("penguins")

ax = sns.swarmplot(data=df, x="body_mass_g", y="sex", hue="species")
ax.set(ylabel="")

Configuring Seaborn

Seaborn ‘themes’ act as shortcuts for setting multiple matplotlib parameters:

Seaborn Command	Accomplishes
set_theme(...)	Set multiple theme parameters in one step.
axes_style(...)	Return a parameter dict for the aesthetic style of the plots.
set_style(...)	Set the aesthetic style of the plots.
plotting_context(...)	Return a parameter dict to scale elements of the figure.
set_context(...)	Set the plotting context parameters.

You can also access:

• Palettes: colormaps can be generated using sns.color_palette(...) and set using sns.set_palette(...).
• Axes Styles: includes darkgrid, whitegrid, dark, white, ticks.

Anatomy of a Figure

Source.

Writing a Figure

There are multiple ways to access/write elements of a plot:

• Figure: high-level features (e.g. title, padding, etc.). Can be accessed via plt.gcf() (get current figure) or upon creation (e.g. f, ax = plt.subplots(1,1) or f = plt.figure()).
• Axes: axis-level features (e.g. labels, tics, spines, limits, etc.). Can be accessed via plt.gca() (get current axes) or upon creation (e.g. f, ax = plt.subplots(1,1) or ax = f.add_subplot(1,1,1)).

Annotations, artists, and other features are typically written into the axes using the coordinate space of the figure (e.g. decimal degrees for lat/long, metres for BNG, etc.).

Adding a 3rd Dimension

This ‘feature’ is less well-developed but does work:

from mpl_toolkits.mplot3d import Axes3D
fig = plt.figure()
ax  = plt.axes(projection='3d')
# OR
fig = plt.figure()
ax  = fig.add_subplot(111, projection='3d')
# THEN
ax.contour3D(X, Y, Z, ...)
ax.plot_surface(x, y, z, ...)
ax.plot3D(xline, yline, zline, ...)
ax.scatter3D(x, y, z, ...)
# ax.plot_surface and ax.plot_wire also give you 3D renderings

You can then set the elevation and azimuth using: ax.view_init(<elevation>, <azimuth>).

Saving Outputs

Straightforward via save figure function, but lots of options!

plt.savefig(fname, dpi=None, facecolor='w', edgecolor='w',
    orientation='portrait', papertype=None, format=None,
    transparent=False, bbox_inches=None, pad_inches=0.1,
    frameon=None, metadata=None)

The format can be largely determined by the file extension in the fname (file name) and the supported formats depends on what you’ve installed! You can find out what’s available to you using: plt.gcf().canvas.get_supported_filetypes().

Jupyter

By default, Jupyter’s output is static matplotlib, but we can extend this in three ways:

1. Make the static plot zoomable and pannable using %matplotlib widget (declare this at the top of your notebook).
2. Make the plot more directly interactive using ipywidgets (import interact and related libs as needed).
3. Use a browser-based visualisation tool such as bokeh, plotly, altair/vega, holoviews, or even d3 (format may be very, very different from what you are ‘used to’ in Python).

Widgets

%matplotlib widget
rs = gpd.sjoin(gdf, hackney, how='left', op='within')
rs.NAME.fillna('None', inplace=True)
ax = hackney.plot(edgecolor='k', facecolor='none')
rs.plot(ax=ax, column='NAME', legend=True)

Interact()

Taking an example from Dani’s work:

from ipywidgets import interact
# Alternatives: interactive, fixed, interact_manual
interact(
    <function>, # Function to make interactive
    <param0>,   # e.g. Data to use
    <param1>,   # e.g. Range start/end/step
    <param2>    # e.g. Fixed value
);

Bokeh

Automation

Plots built on top of matploblib can, to some extent, be automated using functions. For example, to draw circles and place text:

def circle(ax, x, y, radius=0.15):
    from matplotlib.patches import Circle
    from matplotlib.patheffects import withStroke
    circle = Circle((x, y), radius, clip_on=False, zorder=10, 
                    linewidth=1, edgecolor='black', 
                    facecolor=(0, 0, 0, .0125),
                    path_effects=[withStroke(linewidth=5, 
                                  foreground='w')])
    ax.add_artist(circle)

def text(ax, x, y, text):
    ax.text(x, y, text, backgroundcolor="white",
         ha='center', va='top', weight='bold', color='blue')

Don’t Underestimate Text!

Label directly¹:

1. Shamelssly taken from Datawrapper

Don’t Underestimate Text!

Repeat measurement units^{1 2}:

1. Shamelssly taken from Datawrapper

2. Interestingly, this flies in the face of the ‘chart junk minimisation’ of Tufte and Graves-Morris (1983)

Don’t Underestimate Text!

Put the axes where they’re needed¹:

1. Shamelssly taken from Datawrapper

Don’t Underestimate Text!

Emphasise and explain with annotation¹:

1. Shamelssly taken from Datawrapper

Don’t Underestimate Text!

Lead the eye with font sizes, styles, and colors^{1 2}:

1. Shamelssly taken from Datawrapper

2. Just not too many of them in too many different styles!

Don’t Underestimate Text!

See what I said on the previous slide about ‘too much going on’¹:

1. Shamelssly taken from Datawrapper

Don’t Underestimate Text!

For the sake of all that is holy, please don’t center-align text¹:

1. Shamelssly taken from Datawrapper

Don’t Underestimate Text!

Make reading easy and interpretation intuitive^{1 2}:

1. Shamelssly taken from Datawrapper

2. Your readers don’t need to legal names for each country, nor do they like akward hypens.

Don’t Underestimate Text!

Use outlines to allow overlaps^{1 2}:

1. Shamelssly taken from Datawrapper

2. Notice how this supports the focus and interpretation.

Don’t Underestimate Text!

Get to the point in a way that works for the reader^{1 2}:

1. Shamelssly taken from Datawrapper

2. Rephrase for legibility and intelligibility.

Additional Resources

• Matplotlib Cheatsheets
• Introduction to PyPlot (includes lots of parameter information)
• Visualisation with Seaborn
• Seaborn Tutorial
• Elite Data Science Seaborn Tutorial
• Datacamp Seaborn Tutorial
• Three-Dimensional Plotting in Matplotlib
• An easy introduction to 3D plotting with Matplotlib

• DataWrapper Blog
• Using text effectively in data viz
• Choosing fonts for charts and tables
• Bokeh Gallery
• Bokeh User Guide
• Programming Historian: Visualizing Data with Bokeh and Pandas
• Real Python: Data Viz with Bokeh
• Data Viz with Bokeh (Pt. 1)
• Using Interact
• Text in Data Visualizations

Thank You

References

Tufte, Edward R, and Peter R Graves-Morris. 1983. The Visual Display of Quantitative Information. Vol. 2. 9. Graphics press Cheshire, CT.