Stop 03 · Data Manipulation & VisualizationTopics · 7/10

Matplotlib

In short

Loss curves, histograms, scatter plots: you'll plot something in every notebook. Matplotlib is the base layer that Seaborn and pandas plotting sit on, so knowing it lets you fix any chart.

Python's foundational plotting library — lower-level than Seaborn/Plotly but the one every other visualization library is built on top of.

The purple blob paints a simple line chart on a canvas at a wooden easel.

Think of it like this

Painting a chart by hand

A painter starts with a blank canvas and adds every stroke, every label, every colour on purpose. Nothing appears unless they paint it. Matplotlib works the same way: more effort than a ready-made template, but you control every detail.

Watch it happen

What a matplotlib chart is made of

Figure

Axes

Line (ax.plot)

Title

X axis + label

Y axis + label

1/4The Figure is the whole canvas. It can hold one chart or a grid of them.

The big ideas

What you need to know

Use fig, ax = plt.subplots()

The object style is clearer than plt.plot everywhere, and it scales to multi-chart layouts.

Pick the chart for the question

Line for change over time, bar for comparing groups, histogram for one distribution, scatter for two numbers.

Always label

A chart without axis labels and units is a guess. Add a title that states the finding.

Save at good quality

fig.savefig('chart.png', dpi=150, bbox_inches='tight') gives a crisp image for reports and READMEs.

See it in code

Walk through the code

python

1import matplotlib.pyplot as plt2 3epochs = [1, 2, 3, 4, 5, 6]4train_loss = [0.92, 0.61, 0.45, 0.37, 0.31, 0.27]5val_loss = [0.95, 0.70, 0.58, 0.55, 0.56, 0.60]6 7fig, ax = plt.subplots(figsize=(6, 4))8ax.plot(epochs, train_loss, marker="o", label="train")9ax.plot(epochs, val_loss, marker="o", label="validation")10ax.set_title("Validation loss rises after epoch 4")11ax.set_xlabel("Epoch")12ax.set_ylabel("Loss")13ax.legend()14ax.grid(alpha=0.3)15fig.savefig("loss.png", dpi=150, bbox_inches="tight")16plt.show()

1/4Made-up loss values for six epochs of training.

You'll draw training curves like this constantly. When validation loss turns up while training loss keeps falling, the model is starting to overfit.

Save yourself the pain

Common mistakes

Oops

Using a pie chart with eight slices to compare categories.

Do this instead

Use a sorted bar chart. Lengths are much easier to compare than angles.

Oops

Calling plt.show() before plt.savefig and getting a blank image.

Do this instead

Save first, then show. Or call fig.savefig on the figure object directly.

Your mission

Three charts, one figure

~20 min

Quick check

You want to show how the marks of 300 students are spread out. Which chart fits best?

Go deeper

1Practice line, bar, scatter and histogram plots from scratch.
2Learn to customize figures: titles, labels, legends, subplots.
3Recreate a chart you've seen in a report using raw matplotlib.