Programming for Data Processing

Lecture 9
Data Visualization with Python

Today

• Tools: Matplotlib, Seaborn, pandas.plot, Plotly Express, Altair
• Goal: clear, honest, useful charts
• Demo: Link to Google Colab Notebook (use VNU account to access)

1.
Introduction

Why Visualize?

• See structure; Spot anomalies; Communicate
• Tasks → charts:
  • Compare → bar / point / line
  • Distribution → hist / KDE / ECDF / box / violin
  • Relationship → scatter / line (+ trend)
  • Composition → stacked bars/area (sparingly)
• Start with a question, not a chart

Design Quick Hits

• Position > Length > Angle/Area/Color: Clear, honest comparisons
• Label units; Add reference lines
• Colorblind-safe palettes; minimize colors
• Use log scale for heavy tails
• Avoid truncated axes without context

2.
Matplotlib

Matplotlib Mental Model

Object-oriented (OO) API

• Figure → Axes → Artists
• OO pattern: fig, ax = plt.subplots() → methods on ax
• Explicit; Reusable; Composable
• Save: fig.savefig(...)

Matplotlib Mental Model

Object-oriented (OO) API

import numpy as np
import matplotlib.pyplot as plt

x = np.linspace(0, 2*np.pi, 200)
y = np.sin(x)

fig, ax = plt.subplots(figsize=(6,4))

ax.plot(x, y, linewidth=2)
ax.set(title="Sine wave", xlabel="x (rad)", ylabel="y")

fig.tight_layout()
    

# Renders a line chart (Figure)
    

Matplotlib Techniques

• Subplots: fig, axs = plt.subplots(r, c, sharex=True)
• Legends/ticks/grid/annotations via ax methods
• Avoid misleading dual y-axes → normalize or small multiples
• Vector export: SVG/PDF for publication-ready articles

3.
Seaborn

Seaborn: What & Why

• High-level stats plots on Matplotlib
• Axes-level (e.g., sns.boxplot)
• Figure-level (e.g., displot, relplot)
• Theme once: sns.set_theme(style="whitegrid", palette="colorblind")

Distributions: KDE vs ECDF

• KDE: smoothed shape; risk of over-smoothing
• ECDF: exact cumulative distribution
• Use both for group comparison (e.g., penguins)
• Order categories by statistic (e.g., median)

Distributions: KDE vs ECDF

import seaborn as sns
penguins = sns.load_dataset("penguins").dropna()

sns.kdeplot(
    penguins,
    x="body_mass_g",
    hue="species",
    fill=True
)
sns.ecdfplot(
    penguins, 
    x="body_mass_g",
    hue="species"
)
    

Relationships & Categoricals

• Trend/uncertainty: sns.lmplot, sns.regplot
• Overplotting: use alpha, jitter, size
• Summaries: sns.pointplot(estimator="median", errorbar=("pi", 50))
• Keep OO: pass ax= when using axes-level funcs

Faceting

• FacetGrid(..., col="island", hue="species")
• map_dataframe(sns.scatterplot, x=..., y=...)
• Shared scales → fair comparisons
• Labels & titles: g.set_axis_labels(...).set_titles("{col_name}")

Faceting

g = sns.FacetGrid(
    penguins, 
    col="island", 
    hue="species", 
    height=3
)

g.map_dataframe(
    sns.scatterplot, 
    x="bill_length_mm", 
    y="bill_depth_mm"
).add_legend()
    

# One column per island; colors map species
    

4.
Direct Plotting with pandas

Quick EDA with pandas.plot

• Fast first pass from DataFrame/Series
• Timeseries: rolling mean, resample
• Then refine in Seaborn/Matplotlib
• Example: df.plot(kind="line"), df.hist()

5.
Interactive Visualization

Interactivity: Plotly Express

• Hover / Zoom / Facet in one line
• Trendline: trendline="ols" (needs statsmodels)
• Share: fig.write_html("figure.html")

import plotly.express as px

tips = sns.load_dataset("tips")
fig = px.scatter(
    tips, x="total_bill", y="tip", color="time",
    size="size", facet_col="sex", trendline="ols",
    title="Interactive scatter with facets"
)
fig.show()
    

Altair: Declarative Grammar

• Map fields → encodings (concise)
• Faceting via row/column
• Tooltips by default; HTML export

import altair as alt
from vega_datasets import data
cars = data.cars()
alt.Chart(cars).mark_point().encode(
    x="Horsepower", y="Miles_per_Gallon",
    color="Origin", tooltip=["Name","Year","Cylinders"]
).properties(width=320, height=220)
    

6.
Best Practices & Pitfalls

Choosing the Right Tool

• pandas.plot → quickest EDA
• Matplotlib → full control, publication quality
• Seaborn → statistical defaults + faceting
• Plotly/Altair → interactivity & shareable HTML
• Pattern: explore → refine → share

Reproducibility & Pitfalls

• Seed randomness; Pin data versions
• Consistent styles across plots
• Common pitfalls:
  • Dual y-axes without care
  • Missing units/labels
  • Rainbow palettes
  • Truncated axes without context
  • Over-smoothed KDE