Data Visualization

Use this skill for creating effective visualizations: choosing the right library, chart type, and interactivity level for your data and audience.

When to use this skill

• Choosing a visualization library for a project
• Creating exploratory charts during EDA
• Building interactive dashboards
• Producing publication-quality figures
• Understanding tradeoffs between libraries

Library selection guide (2026)

Library	Best For	Interactivity	Learning Curve
Matplotlib	Publication-quality static plots, fine control	Static	Moderate
Seaborn	Statistical visualization, quick EDA	Static	Easy
Plotly	Interactive web charts, dashboards	High	Easy
Altair	Declarative statistical charts, large datasets	Medium	Easy
hvPlot/HoloViz	Large data, linked brushing, geospatial	High	Moderate
Bokeh	Custom interactive web apps	High	Moderate

Quick decision tree

Static publication figure?
  → Matplotlib (full control) or Seaborn (quick statistical)

Interactive web/dashboard?
  → Plotly (easiest), Dash (full apps)
  → Panel/HoloViz (complex linked views)
  → Bokeh (custom web apps)

Large datasets (100k+ points)?
  → hvPlot + Datashader (automatic rasterization)
  → Altair (smart aggregation with Vega-Lite)

Declarative grammar preferred?
  → Altair (Vega-Lite) or Plotly Express

Already using Pandas?
  → df.plot() → Matplotlib
  → df.hvplot() → HoloViz
  → px.scatter(df) → Plotly

Core principles

1) Match chart to data and question

Question	Chart Type
Distribution?	Histogram, KDE, boxplot, violin
Relationship?	Scatter, line, heatmap (correlation)
Composition?	Pie (avoid), stacked bar, treemap
Comparison?	Bar, grouped bar, dot plot
Trend over time?	Line, area, candlestick
Geographic?	Choropleth, scatter map, heatmap

2) Maximize data-ink ratio

• Remove unnecessary gridlines, borders, backgrounds
• Use color purposefully (not decoration)
• Label directly when possible
• One message per visualization

3) Choose interactivity appropriately

Audience	Interactivity Level
Paper/report	Static (Matplotlib/Seaborn)
Presentation	Limited (Plotly static export)
Exploratory analysis	High (zoom, pan, filter, hover)
Stakeholder dashboard	Medium (linked views, drill-down)

Quick examples

Matplotlib (fine control)

import matplotlib.pyplot as plt

fig, ax = plt.subplots(figsize=(10, 6))
ax.scatter(x, y, c=colors, alpha=0.6, edgecolors='none')
ax.set_xlabel('Feature X', fontsize=12)
ax.set_ylabel('Target Y', fontsize=12)
ax.set_title('Relationship Analysis', fontsize=14, fontweight='bold')
ax.spines['top'].set_visible(False)
ax.spines['right'].set_visible(False)
plt.tight_layout()

Seaborn (statistical)

import seaborn as sns

# Distribution with KDE
sns.histplot(data=df, x='value', hue='category', kde=True, bins=30)

# Correlation heatmap
corr = df.corr()
sns.heatmap(corr, annot=True, fmt='.2f', cmap='coolwarm', center=0)

# Categorical comparison
sns.boxplot(data=df, x='category', y='value', palette='viridis')

Plotly (interactive web)

import plotly.express as px

# Scatter with marginal distributions
fig = px.scatter(df, x='x', y='y', color='category', size='size',
                 marginal_x='histogram', marginal_y='rug',
                 hover_data=['label'])
fig.show()

# Faceted small multiples
fig = px.line(df, x='date', y='value', facet_col='category',
              facet_col_wrap=3, height=800)
fig.show()

Altair (declarative, large data)

import altair as alt

# Smart aggregation for large datasets
chart = alt.Chart(df).mark_circle().encode(
    x=alt.X('x:Q', bin=alt.Bin(maxbins=50)),
    y=alt.Y('y:Q', bin=alt.Bin(maxbins=50)),
    size='count()'
).interactive()

chart.save('chart.html')  # Self-contained HTML

hvPlot/HoloViz (large data, linked views)

import hvplot.pandas
import panel as pn

# Linked brushing
scatter = df.hvplot.scatter(x='x', y='y', c='category', 
                            tools=['box_select'], 
                            width=400, height=400)
hist = df.hvplot.hist(y='y', width=400, height=200)

layout = pn.Row(scatter, hist)
layout.servable()

Bokeh (custom web apps)

from bokeh.plotting import figure, show
from bokeh.models import ColumnDataSource, HoverTool

source = ColumnDataSource(df)

p = figure(title="Interactive Plot", tools="pan,wheel_zoom,box_select")
p.circle('x', 'y', source=source, size=10, alpha=0.6)

hover = HoverTool(tooltips=[("X", "@x"), ("Y", "@y"), ("Label", "@label")])
p.add_tools(hover)

show(p)

Anti-patterns

• ❌ Pie charts with many slices (use bar charts)
• ❌ Dual y-axes (hard to read, try normalization or small multiples)
• ❌ 3D charts (distorts perception)
• ❌ Rainbow colormaps (use perceptually uniform: viridis, plasma)
• ❌ Missing labels, titles, or units
• ❌ Overplotting without handling (sampling, alpha, or Datashader)

Common issues and solutions

Problem	Solution
Overplotting (100k+ points)	Use Datashader (rasterization), hexbin, or 2D histogram
Slow interactivity	Reduce data points, use WebGL (Plotly), or pre-aggregate
Large file size	Save as JSON (Plotly/Altair) or use static images
Color blindness	Use colorblind-friendly palettes (viridis, colorbrewer)

Progressive disclosure

• references/matplotlib-advanced.md — Subplots, annotations, custom styles
• references/seaborn-statistical.md — Complex statistical plots
• references/plotly-dash.md — Full dashboards with callbacks
• references/altair-grammar.md — Vega-Lite transformations
• references/holoviz-datashader.md — Large data visualization
• references/bokeh-server.md — Real-time streaming apps

Related skills

• @data-science-eda — Exploration patterns
• @data-science-interactive-apps — Dashboard deployment
• @data-science-notebooks — Notebook-specific visualization

References

• Matplotlib Documentation
• Seaborn Documentation
• Plotly Python
• Altair Documentation
• HoloViz Tutorial
• Bokeh Documentation
• Python Graph Gallery (examples by chart type)