🎯 Why Visualize Data?

Data visualization transforms abstract numbers into visual stories. The human brain processes images 60,000× faster than text. Visualization helps us explore, analyze, and communicate data effectively.

Anscombe's Quartet: Four datasets with nearly identical statistical properties (mean, variance, correlation) that look completely different when plotted. This demonstrates why visualization is essential - statistics alone can be misleading!

Three Purposes of Visualization

1. Exploratory: Discover patterns, anomalies, and insights in your data
2. Explanatory: Communicate findings to stakeholders clearly
3. Confirmatory: Verify hypotheses and validate models
💡 "The greatest value of a picture is when it forces us to notice what we never expected to see." — John Tukey
✅ Always start with visualization before building ML models.

👁️ Visual Perception & Pre-attentive Attributes

The human visual system can detect certain visual attributes almost instantly (< 250ms) without conscious effort. These are called pre-attentive attributes.

Pre-attentive Attributes:
  • Position: Most accurate for quantitative data (use X/Y axes)
  • Length: Bar charts leverage this effectively
  • Color Hue: Best for categorical distinctions
  • Color Intensity: Good for gradients/magnitude
  • Size: Bubble charts, but humans underestimate area
  • Shape: Useful for categories, but limit to 5-7 shapes
  • Orientation: Lines, angles

Cleveland & McGill's Accuracy Ranking

Most Accurate → Least Accurate:
1. Position on common scale (bar chart)
2. Position on non-aligned scale (multiple axes)
3. Length (bar)
4. Angle, Slope
5. Area
6. Volume, Curvature
7. Color saturation, Color hue
⚠️ Pie charts use angle (low accuracy). Bar charts are almost always better!
✅ Use position for most important data, color for categories.

📐 The Grammar of Graphics

The Grammar of Graphics (Wilkinson, 1999) is a framework for describing statistical graphics. It's the foundation of ggplot2 (R) and influences Seaborn, Altair, and Plotly.

Components of a Graphic:
  • Data: The dataset being visualized
  • Aesthetics (aes): Mapping data to visual properties (x, y, color, size)
  • Geometries (geom): Visual elements (points, lines, bars, areas)
  • Facets: Subplots by categorical variable
  • Statistics: Transformations (binning, smoothing, aggregation)
  • Coordinates: Cartesian, polar, map projections
  • Themes: Non-data visual elements (fonts, backgrounds)
💡 Understanding Grammar of Graphics makes you a better visualizer in ANY library.

🎨 Choosing the Right Chart

The best visualization depends on your data type and question. Here's a decision guide:

Single Variable (Univariate):
• Continuous: Histogram, KDE, Box plot, Violin plot
• Categorical: Bar chart, Count plot

Two Variables (Bivariate):
• Both Continuous: Scatter plot, Line chart, Hexbin, 2D histogram
• Continuous + Categorical: Box plot, Violin, Strip, Swarm
• Both Categorical: Heatmap, Grouped bar chart

Multiple Variables (Multivariate):
• Pair plot (scatterplot matrix)
• Parallel coordinates
• Heatmap correlation matrix
• Faceted plots (small multiples)

Common Chart Mistakes

⚠️ Pie charts for many categories - Use bar chart instead
⚠️ 3D effects on 2D data - Distorts perception
⚠️ Truncated Y-axis - Exaggerates differences
⚠️ Rainbow color scales - Not perceptually uniform

🔬 Matplotlib Figure Anatomy

Understanding Matplotlib's object hierarchy is key to creating professional visualizations.

Hierarchical Structure:
Figure → Axes → Axis → Tick → Label

• Figure: The overall window/canvas
• Axes: The actual plot area (NOT the X/Y axis!)
• Axis: The X or Y axis with ticks and labels
• Artist: Everything visible (lines, text, patches)

Two Interfaces

1. pyplot (MATLAB-style): Quick, implicit state
plt.plot(x, y)
plt.xlabel('Time')
plt.show()

2. Object-Oriented (OO): Explicit, recommended for complex plots
fig, ax = plt.subplots()
ax.plot(x, y)
ax.set_xlabel('Time')
✅ Always use OO interface for publication-quality plots.

📈 Basic Matplotlib Plots

Master the fundamental plot types that form the foundation of data visualization.

Code Examples

Line Plot:
ax.plot(x, y, color='blue', linestyle='--', marker='o', label='Series A')

Scatter Plot:
ax.scatter(x, y, c=colors, s=sizes, alpha=0.7, cmap='viridis')

Bar Chart:
ax.bar(categories, values, color='steelblue', edgecolor='black')

Histogram:
ax.hist(data, bins=30, edgecolor='white', density=True)

🔲 Subplots & Multi-panel Layouts

Combine multiple visualizations into a single figure for comprehensive analysis.

Methods:
fig, axes = plt.subplots(2, 2, figsize=(12, 10))
fig, axes = plt.subplots(2, 2, sharex=True, sharey=True)
gs = fig.add_gridspec(3, 3); ax = fig.add_subplot(gs[0, :])
✅ Use plt.tight_layout() or fig.set_constrained_layout(True) to prevent overlaps.

🎨 Styling & Professional Themes

Transform basic plots into publication-quality visualizations.

Available Styles:
plt.style.available → Lists all built-in styles
plt.style.use('seaborn-v0_8-whitegrid')
with plt.style.context('dark_background'):

Color Palettes

Perceptually Uniform: viridis, plasma, inferno, magma, cividis
Sequential: Blues, Greens, Oranges (for magnitude)
Diverging: coolwarm, RdBu (for +/- deviations)
Categorical: tab10, Set2, Paired (discrete groups)

🌊 Seaborn: Statistical Visualization

Seaborn is a high-level library built on Matplotlib that makes statistical graphics beautiful and easy.

Why Seaborn?
  • Beautiful default styles and color palettes
  • Works seamlessly with Pandas DataFrames
  • Statistical estimation built-in (confidence intervals, regression)
  • Faceting for multi-panel figures
  • Functions organized by plot purpose

Seaborn Function Categories

Figure-level: Create entire figures (displot, relplot, catplot)
Axes-level: Draw on specific axes (histplot, scatterplot, boxplot)

By Purpose:
• Distribution: histplot, kdeplot, ecdfplot, rugplot
• Relationship: scatterplot, lineplot, regplot
• Categorical: stripplot, swarmplot, boxplot, violinplot, barplot
• Matrix: heatmap, clustermap

📊 Distribution Plots

Visualize the distribution of a single variable or compare distributions across groups.

Histogram vs KDE:
• Histogram: Discrete bins, shows raw counts
• KDE: Smooth curve, estimates probability density
• Use both together: sns.histplot(data, kde=True)
💡 ECDF (Empirical Cumulative Distribution Function) avoids binning issues entirely.

🔗 Relationship Plots

Explore relationships between two or more continuous variables.

Key Functions:
sns.scatterplot(data=df, x='x', y='y', hue='category', size='magnitude')
sns.regplot(data=df, x='x', y='y', scatter_kws={'alpha':0.5})
sns.pairplot(df, hue='species', diag_kind='kde')

📦 Categorical Plots

Visualize distributions and comparisons across categorical groups.

When to Use:
• Strip/Swarm: Show all data points (small datasets)
• Box: Summary statistics (median, quartiles, outliers)
• Violin: Full distribution shape + summary
• Bar: Mean/count with error bars

🔥 Heatmaps & Correlation Matrices

Visualize matrices of values using color intensity. Essential for EDA correlation analysis.

Best Practices:
• Always annotate with values: annot=True
• Use diverging colormap for correlation: cmap='coolwarm', center=0
• Mask upper/lower triangle: mask=np.triu(np.ones_like(corr))
• Square cells: square=True
💡 Clustermap automatically clusters similar rows/columns together.

🚀 Plotly Express: Interactive Visualization

Plotly creates interactive, web-based visualizations with zoom, pan, hover tooltips, and more.

Why Plotly?
  • Interactive out of the box (zoom, pan, select)
  • Hover tooltips with data details
  • Export as HTML, PNG, or embed in dashboards
  • Works in Jupyter, Streamlit, Dash
  • plotly.express is the high-level API (like Seaborn for Matplotlib)
Common Functions:
px.scatter(df, x='x', y='y', color='category', size='value', hover_data=['name'])
px.line(df, x='date', y='price', color='stock')
px.bar(df, x='category', y='count', color='group', barmode='group')
px.histogram(df, x='value', nbins=50, marginal='box')

🎬 Animated Visualizations

Add time dimension to your visualizations with animations.

Plotly Animation:
px.scatter(df, x='gdp', y='life_exp', animation_frame='year', animation_group='country', size='pop', color='continent')

Matplotlib Animation:
from matplotlib.animation import FuncAnimation
ani = FuncAnimation(fig, update_func, frames=100, interval=50)
✅ Hans Rosling's Gapminder is the classic example of animated scatter plots!

📱 Interactive Dashboards with Streamlit

Build interactive web apps for data exploration without web development experience.

Streamlit Basics:
streamlit run app.py

import streamlit as st
st.title("My Dashboard")
st.slider("Select value", 0, 100, 50)
st.selectbox("Choose", ["A", "B", "C"])
st.plotly_chart(fig)
💡 Streamlit auto-reruns when input changes - no callbacks needed!

🗺️ Geospatial Visualization

Visualize geographic data with maps, choropleth, and point plots.

Libraries:
• Plotly: px.choropleth(df, locations='country', color='value')
• Folium: Interactive Leaflet maps
• Geopandas + Matplotlib: Static maps with shapefiles
• Kepler.gl: Large-scale geospatial visualization

🎲 3D Visualization

Visualize three-dimensional relationships with surface plots, scatter plots, and more.

⚠️ 3D plots can obscure data. Often, multiple 2D views are more effective.
✅ Use Plotly for interactive 3D (rotate, zoom) instead of static Matplotlib 3D.

📖 Data Storytelling

Transform visualizations into compelling narratives that drive action.

The Data Storytelling Framework:
  1. Context: Why does this matter? Who is the audience?
  2. Data: What insights did you discover?
  3. Narrative: What's the storyline (beginning, middle, end)?
  4. Visual: Which chart best supports the story?
  5. Call to Action: What should the audience do?

Design Principles

Remove Clutter: Eliminate chartjunk, gridlines, borders
Focus Attention: Use color strategically (grey + accent)
Think Like a Designer: Alignment, white space, hierarchy
Tell a Story: Title = conclusion, not description
Bad: "Sales by Region"
Good: "West Region Sales Dropped 23% in Q4"
💡 "If you can't explain it simply, you don't understand it well enough." — Einstein
✅ Read "Storytelling with Data" by Cole Nussbaumer Knaflic