Data Visualization Intermediate¶

📊 Data & AI

⏱️ ~3 days 📚 Prerequisites: Matplotlib, Seaborn

When you'd use this

Choosing the right chart and telling stories with data.

Communicate findings visually — choosing the right chart and making it readable — for reports, dashboards, and exploratory analysis.

What you'll learn¶

Visualization turns numbers into insight. Where Matplotlib and Seaborn are the tools, this page is about the craft — picking the right chart and communicating clearly.

Concept-focused page

This is about visualization principles, not library syntax (covered in Matplotlib/Seaborn and Scientific Visualization). The tools aren't installed here; the value here is the decision-making.

Why it matters¶

A core question explored in Data Visualization: Why it matters.

Anscombe's quartet is the classic proof: four datasets with identical means, variances, and correlations look utterly different when plotted — one linear, one curved, one with an outlier. Summary statistics hid what a single glance revealed. Always plot your data before trusting summaries.

Visualization serves three jobs: - Explore — find patterns, outliers, and structure while analyzing. - Explain — communicate a finding to others. - Monitor — dashboards showing live system/business state.

Choosing the right chart¶

Match the chart to the question — comparison, trend, distribution, relationship.

The chart should match the question and the data type:

Your question	Chart
How does Y change with X (continuous)?	Line
Is there a relationship between two variables?	Scatter
How is one variable distributed?	Histogram / KDE / box plot
How do categories compare?	Bar chart
What are the parts of a whole?	Stacked bar (rarely pie)
How do two categories interact (counts)?	Heatmap
How does something change over time?	Line / area

Picking wrong obscures the message — a bar chart hiding a distribution, or a pie chart with 12 slices nobody can compare.

Common mistakes¶

Misleading axes, chart junk, and wrong chart types to avoid.

Charts can mislead — often unintentionally

Truncated y-axis — starting a bar chart's axis at a nonzero value exaggerates differences. Bar charts should start at zero.
Pie charts with many slices — humans can't compare angles well; a bar chart is almost always clearer.
Too much on one chart — five overlapping lines become spaghetti. Split or simplify.
Bad color choices — the jet colormap distorts perception; use perceptually-uniform ones like viridis. And ensure colorblind-safe palettes.
3D when 2D would do — 3D bar/pie charts distort proportions for no benefit.
No labels — unlabeled axes/units make a chart useless.

Static vs interactive¶

When a static image suffices vs when interactivity helps exploration.

Static (Matplotlib, Seaborn) — for reports, papers, print, and anywhere a fixed image is right. Precise control.
Interactive (Plotly, Bokeh, Altair) — zoom, hover, filter; great for exploration and web dashboards where users want to dig in.

import plotly.express as px    # pip install plotly
fig = px.scatter(df, x="gdp", y="life_expectancy",
                 size="population", color="continent", hover_name="country")
fig.show()                     # interactive: hover, zoom, pan

Plotly follows documented API

Not installed here. Interactive libraries shine for dashboards and exploration; static libraries for fixed publication-quality figures. Match the medium to the audience.

Telling a story with data¶

Guide the viewer to the insight, not just the numbers.

Beyond correctness, effective visualization communicates:

One chart, one message. Decide the single point the chart should make, and strip everything that doesn't serve it.
Guide the eye. Use color/annotation to highlight the key data point, not decorate everything.
Order meaningfully. Sort bars by value, not alphabetically, when the ranking is the point.
Add context. A reference line, target, or annotation turns a chart from "here's data" into "here's what it means."
Title with the takeaway. "Sales grew 40% after launch" beats "Monthly sales".

Design for your audience

An exploratory chart for yourself can be quick and dense. A chart for stakeholders should make its point in seconds, with the conclusion obvious. Know which you're making — the same data, different presentation.

Practice exercises¶

For five questions (trend over time, category comparison, distribution, correlation, part-to-whole), name the best chart.
Explain how a truncated y-axis can mislead, with a concrete before/after.
Take a "bad" chart idea (12-slice pie) and describe a clearer alternative.
Rewrite a generic chart title into a takeaway-driven one for a made-up finding.
Decide static vs interactive for: a printed report, a live ops dashboard, an exploratory analysis — and justify each.

💬 Discussion

Have a question about this topic? Found an error? Share your thoughts below.