lesson 218 Visualization & EDA · advanced

Choosing the right chart

The question decides the chart: change over time is a line, comparing categories is bars, a relationship between two numbers is a scatter plot, and how values are spread is a histogram.

You'll be able to

Match a question to a chart: trend → line, comparison → bar, relationship → scatter, spread → histogram
Draw a scatter plot with ax.scatter and read its points from ax.collections
Explain why a line between unordered categories or a bar chart of a long time series misleads

The idea

A chart answers one question. Line: how did a value change over time? The line connects points that have an order. Bar: which category is bigger? Bars compare separate things; their lengths start at 0. Scatter: do two numbers move together (is a bigger budget a bigger revenue)? Each point is one item. Histogram: how are values spread out, where do most fall, are there extremes?

Each kind leaves different objects on the Axes: lines in ax.get_lines(), bars in ax.patches, scatter points in ax.collections (one collection whose get_offsets() are the (x, y) pairs).

Run each example to see what it prints. Running examples doesn't count toward progress.

example-1.py · Do bigger budgets earn more?

import matplotlib.pyplot as plt
import pandas as pd

films = pd.DataFrame({
    "title": ["The Country of Glass", "The Harbor on Fire", "The Island at Dawn", "Quiet Orchard", "Last Tide"],
    "budget_usd": [41100000, 43100000, 9800000, 51600000, 97500000],
    "revenue_usd": [105800000, 207400000, 10700000, 36600000, 540900000],
})
budget = films["budget_usd"] / 1e6
revenue = films["revenue_usd"] / 1e6
fig, ax = plt.subplots()
ax.scatter(budget, revenue)
ax.set_xlabel("Budget (million USD)")
ax.set_ylabel("Revenue (million USD)")
points = ax.collections[0].get_offsets()
print(len(points), points[4].tolist())
print(round(budget.corr(revenue), 2))
plt.show()

example-2.py · A scatter plot with unequal columns

import matplotlib.pyplot as plt

fig, ax = plt.subplots()
ax.scatter([41.1, 43.1, 9.8], [105.8, 207.4])

example-3.py · A trend over days: line, not bars

import matplotlib.pyplot as plt

days = list(range(1, 11))
tmax = [-0.5, 6.7, 9.4, 7.2, 1.7, 5.0, 7.8, 5.6, -0.5, 6.1]   # Wichita, 1 to 10 January 2024
fig, axes = plt.subplots(1, 2, sharey=True)
axes[0].plot(days, tmax, marker="o")
axes[0].set_title("Trend: a line")
axes[1].hist(tmax, bins=[-5, 0, 5, 10])
axes[1].set_title("Spread: a histogram")
print(len(axes[0].get_lines()), len(axes[0].patches), len(axes[1].patches))
print([int(p.get_height()) for p in axes[1].patches])
plt.show()

Mistakes to avoid: a line through categories (it implies an order and in-between values that don't exist); bar charts of long time series (a line shows the trend with less ink); pie charts for more than a few parts, or for comparing similar sizes (bars compare lengths far more accurately); and a scatter plot without units on both axes.

Review

Trend over an ordered axis → line; comparison of categories → bars; relationship of two numbers → scatter; spread → histogram.
Scatter points are in ax.collections[0].get_offsets().
One chart per question; a grid of panels can answer several.
Next: distributions in detail: bins, quantiles and outliers.

Practise this

How strongly do two numbers move together?