data-science-visualization（数据科学可视化）

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)

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

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

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

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)