Annotations guide viewers to insights and add crucial context. Strategic text, arrows, and highlights turn charts into stories, making key findings immediately obvious.

Types of Annotations

1. Text Annotations

import matplotlib.pyplot as plt

plt.plot(months, sales)

# Add text annotation
plt.annotate('Product launch',
             xy=(6, 15000),  # Point to annotate
             xytext=(7, 17000),  # Label position
             arrowprops=dict(arrowstyle='->', color='red'))
plt.show()

2. Highlight Key Areas

# Shade important period
plt.axvspan(6, 8, alpha=0.2, color='yellow', label='Peak Season')

3. Reference Lines

# Add target or threshold line
plt.axhline(y=10000, color='red', linestyle='--', label='Target')

# Add vertical line for event
plt.axvline(x=6, color='green', linestyle=':', label='Campaign Start')

Best Practices

Do:

• Explain "why" something happened
• Highlight the insight, not obvious patterns
• Use consistent style
• Keep text concise
• Position labels to avoid overlaps

Don't:

• Annotate every data point
• State the obvious ("This line goes up")
• Use tiny font sizes
• Clutter the chart
• Cover data with annotations

Effective Annotation Examples

Example 1: Sales Chart

plt.plot(dates, revenue)

# Good annotation - explains WHY
plt.annotate('Black Friday promotion\ndrove 40% revenue spike',
             xy=('2024-11-24', 50000),
             xytext=('2024-12-01', 55000),
             bbox=dict(boxstyle='round', fc='wheat', alpha=0.8),
             arrowprops=dict(arrowstyle='->', lw=1.5))

# Bad annotation - states obvious
# "Revenue increased" ← viewer can see this!

Example 2: Performance Trend

# Highlight anomaly
plt.scatter(dates, values)
plt.annotate('System outage\ncaused data gap',
             xy=('2024-08-15', 0),
             xytext=('2024-08-20', 500),
             arrowprops=dict(arrowstyle='->', color='red', lw=2))
             

When to Annotate

• Anomalies or outliers
• Inflection points
• Key events (product launches, campaigns)
• Targets or thresholds crossed
• Important context not obvious from data