How to Create Effective Data Visualizations

Talk from PyCon DE
April 15th 2026

Transcript

1. How to Create Effective Data Visualizations Dominik Haitz Data Scientist @ Otto Group one.O PyCon DE April 15th 2026
2. Agenda 1. Basic principles • Know your message • The Grammar of Graphics • Visual hierarchy 2. Dos and don‘ts • Annotations and highlighting • Colors • Visual comparability • Bonus chart types • Misc. tips and tricks 3. Data visualization landscape in Python
4. Be conscious about your message • What you want to show? What is the message? • Write it down (on the chart) • Design accordingly • Add texts & highlighting Happiness increases with money, but only up to about $ 50 000
5. The Grammar of Graphics 1999 700 pages 2005
6. The Grammar of Graphics data = penguin_data x = bill_length_mm y = bill_depth_mm color = species plot_type = scatter
7. The Grammar of Graphics data = penguins x = bill_length_mm y = bill_depth_mm color = species plot_type = scatter Data visualization = encoding variables into visual properties
8. Visual hierarchy
9. Color, and bold move elements up in the hierarchy Gray moves down size
11. Tailor chart to audience Title: Take-home-message Comment and highlighting support the message Subtitle: Context Sources Explanatory comments help understand
12. Which colors to use for which type of data Categorical data Diverging data (values around neutral value) Sequential data (continuous values) Contrasting hues (use muted colors) Two contrasting colors, light grey in the middle One- or two-color scales from light to dark
13. The rainbow colormap is bad
14. Why the rainbow color map is bad: • No intuitive ordering of colors • Red „alerts“ the eye • Sharp color transitions create artificial edges
15. Use a one- or two-color colormap
16. Higher values → darker color
17. Use intuitive colors Red/green → use other colors (red/blue, pink/green) or additional markers
18. Don’t use more than ~6 colors
19. Don‘t use color unnecessarily
20. Position/length comparison is most accurate More accurate Less accurate
21. Position/length comparison is most accurate More accurate Less accurate
22. Pie charts are usually bad
23. Only good at: Showing sth. is the fraction of a whole, especially at 25 or 75%
24. Consider box or swarm plots instead of just average values
25. Use horizontal bar charts for long labels
26. Avoid legends if possible Put the names directly on the object (using same color)
27. Python Data Viz Landscape
28. Python Data Viz Landscape Matplotlib ecosystem JS-based libraries
29. Matplotlib seaborn yellowbrick mplcyberpunk • Popular and well tested • High configurability → can do all kinds of plots • API can be confusing • (Previously poor stylistic defaults) plotnine
30. Altair & plotly (express) • Interactive out of the box • Grammar-API
31. Which to use • Matplolib for non-standard plots or for high customization • Plotly express or altair for exploration and interactive plotting in notebooks or dashboards
32. great_tables
33. Summary • Tell a story • Be clear about your message • Use comments, highlighting etc. appropriately • Think „Grammar of Graphics“ • Encoding variables into visual properties Sequential Diverging Categorical