Visualization Cookbook

Pick your question, choose a dataset, get production-ready Python code. Every recipe is designed around what story your data tells — not just which chart looks nice.

⚙

Message First

Start with the question you're answering, not the chart type. The visualization follows the insight.

✂

Data-to-Ink

Every pixel should earn its place. Remove gridlines, borders, and colours that don't carry meaning.

⚓

Anchor in Context

Numbers without context are noise. Show benchmarks, thresholds, or comparisons that make the data meaningful.

☷

Sometimes a Table

If your audience needs exact values or your data has <5 items, a clean table often beats any chart.

Chart Type Guide

When to use each chart — and when not to. Inspired by the Chartosaur principles.

▪ Bar Chart

Compare values across categories. The workhorse of data viz.

Use when: Comparing ≤15 categories on one measure. Horizontal bars for long labels.
Avoid: 3D bars, excessive categories, missing zero baseline.

▮ Histogram

Show the shape of a single continuous variable's distribution.

Use when: You want to see skewness, outliers, or modality. Choose bin widths carefully.
Avoid: Too few bins (hides shape) or too many (noise). Don't use for categories.

☐ Box Plot

Compare distributions across groups — shows median, IQR, and outliers.

Use when: Comparing spread and central tendency across 3–12 groups.
Avoid: When your audience doesn't know how to read box plots. Consider violin or jitter instead.

● Scatter Plot

Reveal relationships between two continuous variables.

Use when: Exploring correlation, clusters, or outliers. Add trend line for emphasis.
Avoid: Overplotting (>5k points without transparency). Consider hex bins or density instead.

⎯ Line Chart

Show trends over a continuous, ordered dimension (usually time).

Use when: Data has a natural order. Limit to 4–5 lines max.
Avoid: Connecting categorical data or unordered groups. Don't truncate y-axis for trend charts.

■ Heatmap

Show patterns in matrices — correlation tables, cross-tabs, or time grids.

Use when: You have two categorical/ordinal axes and a continuous fill value.
Avoid: More than ~20 items per axis. Use diverging colour scales for correlation.

▤ Stacked Bar

Show composition — how parts sum to a whole.

Use when: Showing shares across ≤5 categories. Use 100% stacking for share comparison.
Avoid: Too many segments (>5). Hard to compare middle segments — put the key one at the base.

♣ Violin Plot

Like box plots but show the full distribution shape via kernel density.

Use when: Distributions are bimodal or otherwise non-normal. Half-violins + jitter is powerful.
Avoid: When groups are small (n < 30). KDE can be misleading with small samples.

⋯ Lollipop / Dot Plot

Like bar charts but with less ink. Great for rankings.

Use when: Ranking items, showing gaps to a target, or before/after comparisons.
Avoid: When your audience expects bars. Lollipops work best for data-savvy viewers.

☷ Table

Sometimes the most honest, readable format. Don't force a chart.

Use when: ≤5 rows, exact values matter, or your data tells multiple stories at once.
Avoid: Large tables (>20 rows) without highlighting. Use conditional formatting or sparklines.