Charts with Matplotlib

A figure and its axes

Matplotlib is Python's standard charting library. Two objects do the work:

• The figure is the whole picture, like a sheet of paper.
• An axes is one chart on that sheet: the plotting area, its two axis lines, its labels.

plt.subplots() makes both at once. You then draw on the axes and label it:

import matplotlib.pyplot as plt

months = ["Jan", "Feb", "Mar", "Apr", "May", "Jun"]
orders = [405, 416, 579, 735, 846, 848]

fig, ax = plt.subplots(figsize=(6, 3.5))
ax.plot(months, orders, marker="o")
ax.set_title("Trail-shoe orders, first half of 2024")
ax.set_ylabel("orders")
ax.set_ylim(bottom=0)

In the academy, every open figure appears under the output when the code finishes. On your own computer you'd add plt.show() to open a window, or fig.savefig("orders.png") to save a file.

The habit to build from day one: every chart gets a title and labeled axes. A chart without them makes the reader guess what they're looking at.

Five charts for five questions

Pick the chart from the question you're asking:

Here are three of them on the practice shop's catalog:

import matplotlib.pyplot as plt
import pandas as pd

catalog = pd.read_csv("https://api.gradient.test/v1/datasets/catalog.csv")

fig, axes = plt.subplots(1, 3, figsize=(12, 3.5))

by_category = catalog.groupby("category")["units_q2"].sum().sort_values()
axes[0].barh(by_category.index, by_category.values)
axes[0].set_title("Units sold in Q2")

axes[1].scatter(catalog["price"], catalog["units_q2"], alpha=0.7)
axes[1].set_title("Price against units sold")
axes[1].set_xlabel("price ($)")
axes[1].set_ylabel("units in Q2")

axes[2].hist(catalog["price"], bins=12)
axes[2].set_title("How prices are spread")
axes[2].set_xlabel("price ($)")
fig.tight_layout()

plt.subplots(1, 3) gives one row of three axes, in an array you index like a list. barh draws the bars sideways, which leaves room for long category names. tight_layout() nudges the three charts apart so their labels don't overlap.

import matplotlib.pyplot as plt
fig, axes = plt.subplots(2, 3)
print(axes.shape, len(fig.axes))

The month that isn't over

Here's the trap that catches the most people. The shoe-orders table holds daily orders from 2019 up to the day it was exported: June 15, 2025. Add the days up by month and plot them:

import matplotlib.pyplot as plt
import pandas as pd

daily = pd.read_csv("https://api.gradient.test/v1/datasets/shoe-orders.csv")
daily["month"] = daily["date"].str[:7]          # "2025-06-15" -> "2025-06"

monthly = daily.groupby("month")["orders"].agg(["sum", "count"])
print(monthly.tail(3))

recent = monthly.loc["2024-01":]
fig, ax = plt.subplots(figsize=(8, 3.5))
ax.plot(recent.index, recent["sum"], marker="o")
ax.set_title("Monthly trail-shoe orders")
ax.set_ylabel("orders")
ax.set_ylim(bottom=0)
ax.tick_params(axis="x", labelrotation=90)

The last point falls off a cliff. A manager would ask what went wrong in June. Nothing did: the count column shows June has only 15 days of data, against 30 or 31 for the others.

agg(["sum", "count"]) works out two summaries at once, so each month's total sits next to the number of days behind it. That's the check that catches the trap. Then either drop the unfinished month, or show it differently and say so on the chart.

Scales and baselines

Two settings change what a reader believes, so set them on purpose:

• Where the axis starts. A bar's length is its value, so bar charts must start at zero. A line chart can zoom in to show small wiggles, as long as the axis labels make the scale clear.
• Linear or log. When values span from 5 to 50,000, a normal axis squashes the small ones flat. ax.set_yscale("log") spaces 10, 100, 1,000 and 10,000 evenly, so every value has room. It also means equal steps up are equal percentage growth, which suits sales and populations.

import matplotlib.pyplot as plt

sellers = ["A", "B", "C", "D", "E", "F"]
units = [12, 48, 150, 900, 2400, 31000]

fig, (left, right) = plt.subplots(1, 2, figsize=(10, 3.2))
left.bar(sellers, units)
left.set_title("Linear: most bars vanish")
right.bar(sellers, units)
right.set_yscale("log")
right.set_title("Log scale: every seller visible")

A log axis needs a label or a note, because many readers won't notice it. And it can't show zero or negative values: log(0) doesn't exist.

import matplotlib.pyplot as plt
import pandas as pd

orders = pd.read_csv("https://api.gradient.test/v1/datasets/orders.csv")
orders["revenue"] = orders["qty"] * orders["unit_price"]

fig, ax = plt.subplots()

Small multiples and a legend

When you compare several groups, you can draw them all on one axes with a legend, or give each its own small chart. The second idea is called small multiples. One rule for choosing: up to three or four lines read fine on one chart. Beyond that, a tangle of lines is worse than a grid of small, simple charts.

import matplotlib.pyplot as plt
import pandas as pd

orders = pd.read_csv("https://api.gradient.test/v1/datasets/orders.csv")
orders["month"] = orders["date"].str[:7]
stores = ["portland", "seattle", "denver"]

fig, ax = plt.subplots(figsize=(8, 3.2))
for store in stores:
    lines = orders[orders["store"] == store].groupby("month")["qty"].sum()
    ax.plot(lines.index, lines.values, label=store)
ax.legend()
ax.set_title("Units sold per month, by store")
ax.tick_params(axis="x", labelrotation=90)

fig2, axes = plt.subplots(1, 3, figsize=(11, 3), sharey=True)
for ax2, store in zip(axes, stores):
    lines = orders[orders["store"] == store].groupby("month")["qty"].sum()
    ax2.plot(range(len(lines)), lines.values)
    ax2.set_title(store)
    ax2.set_xlabel("month number")
axes[0].set_ylabel("units")
fig2.tight_layout()

label= names each line, and ax.legend() draws the key. sharey=True gives the small multiples one y-axis scale. Without it, each chart picks its own scale, and a small store can look as busy as a big one.

Box plots: comparing spreads

An average hides how spread out the values are. A box plot shows the spread. The box runs from the 25th to the 75th percentile, so the middle half of the values sit inside it. The line across the box is the median. The whiskers reach out to the more typical extremes, and dots beyond them are unusual values.

import matplotlib.pyplot as plt
import pandas as pd

catalog = pd.read_csv("https://api.gradient.test/v1/datasets/catalog.csv")
names = sorted(catalog["category"].unique())
groups = [catalog[catalog["category"] == c]["price"] for c in names]

fig, ax = plt.subplots(figsize=(7, 3.5))
ax.boxplot(groups, tick_labels=names)
ax.set_title("Price by category")
ax.set_ylabel("price ($)")

Tents cost the most and vary the most. Camp-kitchen gear is cheap and tightly bunched. A bar chart of average prices would have shown the first fact but hidden the second. That's the reason to reach for a box plot whenever the spread matters as much as the middle.

import matplotlib.pyplot as plt
import pandas as pd

daily = pd.read_csv("https://api.gradient.test/v1/datasets/shoe-orders.csv")
daily["month"] = daily["date"].str[:7]
monthly = daily.groupby("month")["orders"].agg(["sum", "count"])
complete = monthly[monthly["count"] >= 28]
recent = complete.loc["2024-01":]

fig, ax = plt.subplots(figsize=(8, 3.5))
ax.plot(recent.index, recent["sum"], marker="o", label="orders")
ax.set_ylim(bottom=0)
ax.set_title("Trail-shoe orders peak every May to August")
ax.set_ylabel("orders per month")
ax.tick_params(axis="x", labelrotation=90)
ax.spines[["top", "right"]].set_visible(False)
fig.text(0.01, -0.02, "Source: practice shop online orders, complete months only.", fontsize=8)
print(recent["sum"].idxmax(), recent["sum"].max())