Level 2 · Business baskets against consumer baskets

histplot draws a distribution

sns.histplot is seaborn’s histogram. Same pattern as every seaborn function: the table, then column names.

ax = sns.histplot(data=per_order, x="value", bins=30)

Compared with ax.hist, the gain is hue=: one call draws a histogram per group, in different colors, with a legend.

ax = sns.histplot(data=per_order, x="value", hue="segment", bins=30,
                  element="step", stat="density", common_norm=False, kde=True)

• element="step" draws outlines instead of filled bars, so overlapping groups stay readable.
• stat="density" with common_norm=False scales each group to its own size. Without it, the 3,162 consumer orders would dwarf the 1,133 business ones and you’d compare group sizes instead of shapes.
• kde=True adds a smooth curve over each histogram, an estimate of the distribution’s shape.

The numbers behind the shapes

A chart shows the difference; numbers measure it. Named aggregation gives one row per segment:

per_order.groupby("segment", as_index=False).agg(
    n_orders=("order_id", "size"),
    median_value=("value", "median"),
    mean_value=("value", "mean"),
)
#     segment  n_orders  median_value  mean_value
#    business      1133        706.90     1059.42
#    consumer      3162        183.77      348.31

A business order is typically about four times a consumer order. Business accounts placed 26% of 2025’s completed orders but brought 52% of the revenue. In both groups the mean is far above the median, because a few very large baskets pull it up: the medians are the fairer “typical order”.

Trimming the view, not the data

A handful of orders above $3,000 stretch the x-axis so far that the bulk of the chart is squeezed into the left edge. It’s fine to draw only orders up to $3,000 and say so in the axis label. The numbers in result still use every order. Trim what you show, never what you compute, and tell the reader where you cut.

Common mistakes

• Plotting order lines instead of orders: a basket of five lines would count five times.
• Merging the segment with a key that isn’t unique on the customers side. validate="many_to_one" stops that early.
• Comparing two groups of very different sizes with raw counts instead of densities.
• Quoting only the means of skewed groups.
• Leaving kde=True on a small group. With a few dozen values the smooth curve invents bumps that aren’t there; trust the bars and the medians.