PY-L4-37 · Project — Data Story, A 3-Chart Report

Project Goals

3 min

State one specific question in one sentence.
Pick three charts that each answer a sub-question.
Compose them into a single PNG with a story-arc headline.
Write a four-sentence conclusion in a README.

Warm-Up · The Three-Chart Arc

5 min

Chart 1 — what happened?     overall trend / total / size
Chart 2 — who / where / what?  broken down by category
Chart 3 — what's interesting?   surprise, outliers, correlation, comparison

Three charts. One question. One arc. Every good report follows this shape — Vox explainers, FT Visual Capitalist, the data section of any newspaper.

Today's big idea

A data report stands or falls on the headline question. Write it first. The charts come next.

Plan · One Question → Three Charts

14 min

Sample question: "What does our shop's May 2026 look like?"

Chart 1 — Daily revenue line with 3-day rolling average. Story: "the month overall".
Chart 2 — Bar of revenue by product. Story: "who's driving it".
Chart 3 — Pie of share by customer (or scatter of qty × price). Story: "a takeaway".

Skeleton

# data_story.py
import pandas as pd
import matplotlib.pyplot as plt
from matplotlib.gridspec import GridSpec

df = pd.read_csv("clean.csv", parse_dates=["date"])
df["total"] = df["quantity"] * df["price"]

# --- compute ----------------------------------------------------
daily = (df.groupby(df["date"].dt.date)["total"].sum()
           .rename("rev").to_frame())
daily["roll"] = daily["rev"].rolling(3).mean()
by_product   = df.groupby("product")["total"].sum().sort_values(ascending=False)
by_customer  = df.groupby("customer")["total"].sum().sort_values(ascending=False)

# --- draw -------------------------------------------------------
fig = plt.figure(figsize=(12, 7), constrained_layout=True)
gs  = GridSpec(2, 3, figure=fig)

# Chart 1: trend
a1 = fig.add_subplot(gs[0, :2])
a1.plot(daily.index, daily["rev"], alpha=0.4, label="daily")
a1.plot(daily.index, daily["roll"], linewidth=2, label="3-day avg")
a1.set_title("1.  Revenue trend")
a1.set_ylabel("RM")
a1.legend()

# Chart 2: bars
a2 = fig.add_subplot(gs[0, 2])
a2.bar(by_product.index, by_product.values, color="#4a90e2")
a2.set_title("2.  By product")

# Chart 3: pie
a3 = fig.add_subplot(gs[1, :])
a3.pie(by_customer.head(5), labels=by_customer.head(5).index,
       autopct="%1.0f%%", startangle=90, counterclock=False)
a3.set_title("3.  Top-5 customer share")

fig.suptitle("Sales · May 2026 — total RM {:.0f} across {} orders"
             .format(daily["rev"].sum(), len(df)),
             fontsize=15, weight="bold")
fig.savefig("data_story.png", dpi=200)

Build Your Own

12 min

Pick from your homework dataset. Spend the first 5 minutes writing your question on paper. Then sketch the three charts. Then write the code.

The README

# Data story — &lt;your title&gt;

**Question:** What does &lt;dataset&gt; tell us about &lt;topic&gt;?

**Chart 1:** &lt;what we show&gt;
**Chart 2:** &lt;what we show&gt;
**Chart 3:** &lt;what we show&gt;

## Conclusion
Three or four sentences. State the surprise / pattern / decision the data
supports. Mention one caveat (data source, time range, missing values).

Extensions

13 min

01 🟢 Headline numbers row

Add three KPI text-cards at the top of the figure (Lesson 36 trick): total, count, average.

02 🟡 Annotate the peak

On the trend chart, annotate the day with the highest revenue using ax.annotate.

03 🔴 Brand it

Apply a consistent color palette across all three charts (e.g., shades of one colour). Add a small subtle source line at the bottom-right with fig.text(0.99, 0.01, "source: ...", ha="right", alpha=0.5).

Stretch · Make It a PDF

8 min

Use matplotlib.backends.backend_pdf.PdfPages to put your figure + a second figure (e.g., methodology) into a single multi-page PDF. Now your "data story" is a one-click deliverable.

Show the API

from matplotlib.backends.backend_pdf import PdfPages
with PdfPages("data_story.pdf") as pdf:
    pdf.savefig(fig_story)
    pdf.savefig(fig_method)

Recap

3 min

You shipped your first portfolio-worthy data story. The arc — question, three charts, conclusion — works for school reports, blog posts, work decks. Tomorrow we shift to web apps: Flask.

Homework

4 min

Polish your data story to portfolio quality: one PNG + one README, both committed to a folder. Tweet the headline + image (optional!) — feedback from the world is the best teacher.

# data_story.py import pandas as pd import matplotlib.pyplot as plt from matplotlib.gridspec import GridSpec df = pd.read_csv("clean.csv", parse_dates=["date"]) df["total"] = df["quantity"] * df["price"] # --- compute ---------------------------------------------------- daily = (df.groupby(df["date"].dt.date)["total"].sum() .rename("rev").to_frame()) daily["roll"] = daily["rev"].rolling(3).mean() by_product = df.groupby("product")["total"].sum().sort_values(ascending=False) by_customer = df.groupby("customer")["total"].sum().sort_values(ascending=False) # --- draw ------------------------------------------------------- fig = plt.figure(figsize=(12, 7), constrained_layout=True) gs = GridSpec(2, 3, figure=fig) # Chart 1: trend a1 = fig.add_subplot(gs[0, :2]) a1.plot(daily.index, daily["rev"], alpha=0.4, label="daily") a1.plot(daily.index, daily["roll"], linewidth=2, label="3-day avg") a1.set_title("1. Revenue trend") a1.set_ylabel("RM") a1.legend() # Chart 2: bars a2 = fig.add_subplot(gs[0, 2]) a2.bar(by_product.index, by_product.values, color="#4a90e2") a2.set_title("2. By product") # Chart 3: pie a3 = fig.add_subplot(gs[1, :]) a3.pie(by_customer.head(5), labels=by_customer.head(5).index, autopct="%1.0f%%", startangle=90, counterclock=False) a3.set_title("3. Top-5 customer share") fig.suptitle("Sales · May 2026 — total RM {:.0f} across {} orders" .format(daily["rev"].sum(), len(df)), fontsize=15, weight="bold") fig.savefig("data_story.png", dpi=200)

# Data story — <your title> **Question:** What does <dataset> tell us about <topic>? **Chart 1:** <what we show> **Chart 2:** <what we show> **Chart 3:** <what we show> ## Conclusion Three or four sentences. State the surprise / pattern / decision the data supports. Mention one caveat (data source, time range, missing values).