Subplot grids

plt.subplots(2, 3) — rows first, then columns. One Figure, six Axes, and axes comes back as a numpy array you index normally:

fig, axes = plt.subplots(2, 3, figsize=(9, 5))

axes.shape        # (2, 3)
axes[0, 2]        # top right
len(fig.axes)     # 6

The shape trap

The shape of axes depends on how you called it, which is a real source of breakage:

plt.subplots()        # a bare Axes — not an array at all
plt.subplots(1, 3)    # shape (3,)   -> axes[1]
plt.subplots(3, 1)    # shape (3,)   -> axes[1]
plt.subplots(2, 3)    # shape (2, 3) -> axes[0, 1]

So a loop written for a 1×3 row raises IndexError the day someone makes it 2×3 — or worse, axes[1] quietly becomes an entire row of Axes rather than one of them. Two ways out:

for ax, name in zip(axes.flat, names):    # works for any shape
    ax.plot(days, week[name])

fig, axes = plt.subplots(1, 3, squeeze=False)   # always 2-D

axes.flat is the one to build the habit around. It iterates every Axes in reading order regardless of the grid, and it's shorter than what it replaces.

Small multiples

This is what grids are for. The same chart repeated once per category, same scale, same style, so the only thing varying between panels is the data:

fig, axes = plt.subplots(2, 3, figsize=(10, 5), sharey=True, layout="constrained")

for ax, item in zip(axes.flat, items):
    sub = orders[orders["item"] == item]
    ax.hist(sub["wait"], bins=12)
    ax.set_title(item, fontsize=10)

Five overlapping histograms on one Axes is a mess; five small ones side by side is readable at a glance. The general rule: when you're tempted to put a fifth series on one chart, make small multiples instead.

Sharing is what makes it work

sharey=True is not a nicety here — it's the thing that makes the comparison valid. Without it every panel autoscales to its own data, so a panel of tiny numbers looks identical to a panel of huge ones and the grid actively misleads.

fig, axes = plt.subplots(2, 3, sharey=True)    # one y scale for all
fig, axes = plt.subplots(3, 1, sharex=True)    # stacked time series

Sharing also hides the interior tick labels, which is why a shared grid looks tidier than one you spaced by hand — no repeated axis running down the middle. And it links the views, so zooming one panel interactively moves the rest.

Leftover panels

Five categories in a 2×3 grid leaves an empty box, and an empty box reads as missing data rather than spare capacity:

for extra in axes.flat[len(items):]:
    fig.delaxes(extra)

Better still, pick a grid that fits — 1×5, or 2×3 with something useful in the sixth cell.

Sizing

figsize is the whole figure, so a 2×3 grid at the default 6.4×4.8 gives each panel about 2×1.6 inches, which is too small for axis labels. Scale it with the grid — roughly 3×2.2 inches per panel is a reasonable starting point.