Why you must visualize

The Business Analytics section closed by promising a quiet shock: four datasets with identical statistics that look nothing alike. Here it is — and it is the single best argument for plotting before you trust anything.

You can fit a linear regression to a dataset, get a clean R² (the fraction of variance the line explains — 1.0 is a perfect fit) of 0.67, and have no idea that your data is a parabola, an outlier-driven line, or a vertical stripe. Four datasets can share every summary statistic you’d think to check — mean, variance, correlation, regression line — and still look completely different. This is Anscombe’s quartet, and it’s the single best argument for plotting before you trust anything.

Anscombe’s quartet — the reveal

import numpy as np
import matplotlib.pyplot as plt

# The four classic Anscombe datasets (1973)
x1 = np.array([10, 8, 13, 9, 11, 14, 6, 4, 12, 7, 5], dtype=float)
y1 = np.array([8.04, 6.95, 7.58, 8.81, 8.33, 9.96, 7.24, 4.26, 10.84, 4.82, 5.68])
y2 = np.array([9.14, 8.14, 8.74, 8.77, 9.26, 8.10, 6.13, 3.10, 9.13, 7.26, 4.74])
y3 = np.array([7.46, 6.77, 12.74, 7.11, 7.81, 8.84, 6.08, 5.39, 8.15, 6.42, 5.73])
x4 = np.array([8, 8, 8, 8, 8, 8, 8, 19, 8, 8, 8], dtype=float)
y4 = np.array([6.58, 5.76, 7.71, 8.84, 8.47, 7.04, 5.25, 12.50, 5.56, 7.91, 6.89])

datasets = [(x1, y1, "I"), (x1, y2, "II"), (x1, y3, "III"), (x4, y4, "IV")]

# Print the supposedly "identical" stats
for x, y, label in datasets:
    print(f"{label}: mean(x)={x.mean():.2f}  mean(y)={y.mean():.2f}  "
          f"var(x)={x.var(ddof=1):.2f}  corr={np.corrcoef(x,y)[0,1]:.2f}")

# Now plot all four
fig, axes = plt.subplots(2, 2, figsize=(8, 6), sharex=True, sharey=True)
for ax, (x, y, label) in zip(axes.flat, datasets):
    ax.scatter(x, y, color="steelblue")
    # Same regression line fits all four (slope~0.5, intercept~3)
    xs = np.linspace(3, 20, 50)
    ax.plot(xs, 3 + 0.5 * xs, color="crimson", lw=1)
    ax.set_title(f"Dataset {label}")
fig.suptitle("Anscombe's quartet — same stats, different worlds")
fig.tight_layout()
plt.show()

The printed statistics are eerily, exactly identical across all four:

I:   mean(x)=9.00  mean(y)=7.50  var(x)=11.00  corr=0.82
II:  mean(x)=9.00  mean(y)=7.50  var(x)=11.00  corr=0.82
III: mean(x)=9.00  mean(y)=7.50  var(x)=11.00  corr=0.82
IV:  mean(x)=9.00  mean(y)=7.50  var(x)=11.00  corr=0.82

But the plot tells four completely different stories:

Every dataset has the same mean, variance, and correlation. The same regression line goes through all four. But one is roughly linear, one is a clean parabola, one has a single outlier pulling a perfect line off, and one is a vertical strip plus a leverage point. A summary statistic collapses information. A plot preserves it.

The 3 audiences for any plot

When you make a plot, ask first: who’s going to look at this?

• Yourself, during exploration. Ugly is fine. Fast is everything. Default colors, no titles, ten plots in five minutes. The goal is to find something surprising before you commit to a story.
• Your team, during review. Clear axis labels, a title, legible legend. The plot has to defend a claim — “the model overfits after epoch 15” — and a colleague should be able to read it without a verbal explanation.
• Your stakeholders, in a report or dashboard. Now design matters. Consistent palette, callouts on the interesting bits, no jargon in the legend, no chartjunk. Often one or two plots that tell the whole story.

Don’t mix the bars. A messy exploratory plot in a board deck makes you look careless. A heavily-styled, annotated plot in a notebook wastes twenty minutes you should have spent making the next ten plots.

Small multiples

Look back at the Anscombe plot — four small panels, each with the same axes and scales. That’s a “small multiples” layout, and it’s one of the most powerful visualization patterns ever invented. The constant axes do the comparison for you: your eye instantly catches what differs between panels.

Use small multiples whenever you have a categorical dimension you’d otherwise stuff into colors. Twelve countries on one line chart is a spaghetti tangle. Twelve countries as twelve small line charts in a 4×3 grid is a story.

What you’re going to learn

The rest of this section moves from the absolute basics — figures and axes — through the workhorse plots (lines, scatters, subplots), into the ML-specific plots every model card needs, and finally into Seaborn for fast statistical graphics. By the end you’ll have a small kit of plots you can produce in under a minute apiece, and that’s most of the job.

In one breath

Summary statistics collapse information; a plot preserves it — Anscombe’s quartet proves four datasets can share an identical mean, variance, correlation, and regression line while being a line, a parabola, an outlier-bent fit, and a vertical stripe. So plot before you trust any number. Match the plot’s polish to its audience: ugly-and-fast for your own exploration, labelled-and-defensible for your team, designed-and-decluttered for stakeholders. And reach for small multiples — a grid of panels sharing the same axes — whenever you’d otherwise tangle a chart with many coloured lines; the constant scales do the comparison for your eye.

Practice

A question to carry forward

We just leaned on a line of code without explaining it: fig, axes = plt.subplots(2, 2). It returned two things — a fig and a grid of axes — and every ax.scatter(...) afterward drew onto one of them. That pairing is not incidental; it is the entire mental model matplotlib is built on, and almost every confusing matplotlib error traces back to not having it straight.

So the question to carry forward is: what exactly are a Figure and an Axes, and why does every plot you’ll ever make start by getting hold of both? The next lesson, Figure & Axes, draws the one diagram that makes matplotlib finally click — the canvas, the plots that sit on it, and why plt.plot() quietly guessing which axes you meant is the source of so much grief.