Level 1 · Orders per channel with seaborn

The seaborn pattern

seaborn is imported as sns, by a convention as universal as pd and np. Every function has the same shape: you give it the whole table and the names of the columns to use.

import seaborn as sns
sns.set_theme(style="whitegrid")       # once, at the top: a clean default look

ax = sns.countplot(data=o25, x="channel")
ax.set_title("Orders per channel, 2025")

It returns an ordinary matplotlib Axes, so everything from chapter 11 still applies: set_title, set_xlabel, set_ylim, bar_label. seaborn draws the chart; matplotlib finishes it.

countplot counts for you

countplot takes raw rows and counts how many fall in each category, like value_counts() drawn as bars. Its bars come out in the order the categories first appear, so sort them with order=:

counts = o25["channel"].value_counts()      # app 2135, web 1946, marketplace 702
ax = sns.countplot(data=o25, x="channel", order=counts.index)
ax.bar_label(ax.containers[0], padding=3)   # write the count on each bar

ax.containers[0] is the group of bars seaborn drew; bar_label writes each bar’s height above it. For categories with long names, use y="channel" instead of x= to get horizontal bars.

The table behind the chart is the same value_counts, turned into two named columns:

counts.rename_axis("channel").reset_index(name="n_orders")

hue adds a second category

hue= splits every bar by a second column and adds a legend automatically, which is seaborn’s biggest convenience over plain matplotlib:

ax = sns.countplot(data=o25, x="channel", hue="status", order=counts.index,
                   hue_order=["completed", "cancelled", "refunded", "pending"])

In 2025 the app had 133 cancelled orders out of 2,135 (6.2%), the website 102 out of 1,946 (5.2%) and the marketplace 40 out of 702 (5.7%). With the completed bars so tall, small groups are hard to read; when the small groups are the point, chart their rates instead.

Common mistakes

• Giving countplot a table of counts: it would count the rows of that table (one per channel). For precomputed numbers use sns.barplot(data=table, x="channel", y="n_orders").
• Forgetting order=, so the bars come out in data order instead of ranked.
• Writing x=o25["channel"] and data=o25 together. Pass column names as strings with data=.
• Calling sns.set_theme() before every chart. Once at the top is enough.
• Expecting countplot to return the counts. It returns the Axes; the numbers come from value_counts(), which is also what you quote in the text next to the chart, so build the table first and draw from the same data.