Capstone: a dashboard

You have orders — 120 orders over March, with a timestamp, a temperature, a cup count, a wait time and an item. Build one figure that answers four questions about the month.

1. Decide the questions

Before any code. A dashboard assembled from whichever charts were easiest is a wall of charts; four questions with four answers is a page somebody reads.

1. Is trade growing? → cups per day, as a line.
2. What sells? → total cups per item, as a sorted barh.
3. How long do people wait? → a histogram of wait.
4. Does the weather matter? → temperature against cups, as a scatter.

Note that the chart type falls out of the question. Over time → line. Across categories → bars. One column's shape → histogram. Two columns together → scatter. That's lessons 5–8 in one sentence.

2. Lay it out

fig, axd = plt.subplot_mosaic(
    [["trend", "trend"],
     ["items", "wait"],
     ["temp",  "temp"]],
    figsize=(10, 9),
    layout="constrained",
)

The trend gets the full width because it's the headline. Nested lists rather than a string, because the names are words (lesson 14).

3. Reshape, then draw

daily = orders.set_index("when")["cups"].resample("D").sum()
axd["trend"].plot(daily.index, daily.values)

by_item = orders.groupby("item")["cups"].sum().sort_values()
axd["items"].barh(by_item.index, by_item.values)

axd["wait"].hist(orders["wait"], bins=12)

axd["temp"].scatter(orders["temp"], orders["cups"], s=14, alpha=0.5)

Each is one line of drawing after one line of reshaping. Note sort_values() ascending on the bars — barh counts upwards, so ascending puts the biggest on top (lesson 7).

4. Make each panel say something

A panel with a bare column name as its title makes the reader do the work. Put the finding in each title:

axd["trend"].set_title(f"{int(daily.sum()):,} cups over {len(daily)} days",
                       loc="left", fontsize=11)
axd["items"].set_title("Plain chai is nearly a third of sales", loc="left", fontsize=11)

And add the things that turn a chart into a claim (lesson 12):

# the busiest day, found not typed
best = daily.idxmax()
axd["trend"].annotate(f"best day: {int(daily.max())}",
                      xy=(best, daily.max()),
                      xytext=(10, -18), textcoords="offset points",
                      arrowprops=dict(arrowstyle="->", color="#9a9a9a"))

# median rather than mean, because the wait times are skewed (lesson 8)
axd["wait"].axvline(orders["wait"].median(), linestyle="--", color="#74dfa2")

# the slope, in units a person can repeat
m, b = np.polyfit(orders["temp"], orders["cups"], 1)
xs = np.array([orders["temp"].min(), orders["temp"].max()])
axd["temp"].plot(xs, m * xs + b, linestyle="--", color="#66aaf9")

textcoords="offset points" is worth knowing: it places the label a fixed distance from the target in points, so it stays put whatever the data range does.

5. Fit it, then save it

fig.suptitle("Chai stall — March", fontsize=15)
fig.savefig("dashboard.png", dpi=150, bbox_inches="tight")

layout="constrained" was set at creation, so nothing collides (lesson 15), and bbox_inches="tight" keeps the suptitle from being cropped (lesson 4).

What to check before you send it

• Every axis labelled, with units.
• Every panel title states a finding, not a column name.
• Bars sorted; bars start at zero.
• The dense scatter has alpha.
• Nothing overlaps in the saved file, not just the preview.
• One colour doing the emphasis, not five competing.