From Tables to Questions

The most common beginner mistake with charts is starting with the chart type:

I need to make a bar chart.

Sometimes that is fine, but it skips the important step. A chart should start from a question:

Which subjects take the most total time?

Now the bar chart has a purpose. It will compare subjects by total minutes.

The chart type should follow the question and the data. In this part, four question families are enough.

This part mostly uses the first three. Relationship charts are useful, but they can quickly lead into statistics. It is enough here to know that relationship questions exist and that not every question is a bar chart question.

Start With One Question

Suppose this is your study-log DataFrame:

import pandas as pd

records = [
    {"subject": "math", "minutes": 45, "date": "2026-05-10"},
    {"subject": "history", "minutes": 30, "date": "2026-05-11"},
    {"subject": "programming", "minutes": 60, "date": "2026-05-11"},
    {"subject": "math", "minutes": 25, "date": "2026-05-12"},
    {"subject": "programming", "minutes": 90, "date": "2026-05-13"},
]

df = pd.DataFrame(records)
df["date"] = pd.to_datetime(df["date"])

Here are three different questions you could ask:

Which subject has the most total minutes?
How many minutes were logged each day?
How long are sessions usually?

They all use the same table, but they do not use the same summary.

For subject totals:

subject_totals = df.groupby("subject")["minutes"].sum()

For daily totals:

daily_totals = df.groupby("date")["minutes"].sum()

For session lengths:

lengths = df["minutes"]

The first two are grouped summaries. The third uses one original column. That difference matters before plotting starts.

A Chart Is Not the First Draft of the Data

The summary helpers from Using Libraries are still useful. If you can print a small summary and predict whether it is right, the chart has a trustworthy source.

For example:

subject_totals = df.groupby("subject")["minutes"].sum()
subject_totals = subject_totals.sort_values(ascending=False)
print(subject_totals)

Output:

subject
programming    150
math            70
history         30
Name: minutes, dtype: int64

Now the chart has a simple job: show programming, math, and history with bars whose heights are 150, 70, and 30. If the chart later looks strange, you can compare it to this printed summary.

This is also why the question should be specific. “Visualize my study log” is too broad. It leaves too many decisions hidden. “Show total minutes by subject, from most to least” is specific enough to check.

Wrong Chart, Right Code

Code can run and still answer the wrong question.

Suppose the question is:

How has my total study time changed from day to day?

A bar chart of subject totals may run correctly:

import matplotlib.pyplot as plt

subject_totals = df.groupby("subject")["minutes"].sum()

fig, ax = plt.subplots()
ax.bar(subject_totals.index, subject_totals.values)

You will meet this plotting code properly in the next chapter. Here the only point is what it draws: one bar per subject. The code runs, but it does not answer the question. It compares subjects. The question asked about time.

The right preparation step points toward a time chart:

daily_totals = df.groupby("date")["minutes"].sum()
daily_totals = daily_totals.sort_index()

Only after that does a line chart make sense.

When you look at a proposed chart, ask:

1. What question does this chart answer?
2. Is that the question I meant to ask?
3. What rows and columns did the code use?
4. Was the data summarized before plotting, if the question needed a summary?

These questions will matter even more in the AI chapter, because AI assistants often produce visually plausible chart code for a slightly different question.

Check Your Understanding

1. Why should a chart start from a question instead of a chart type?
2. Which chart type fits a comparison question, a trend question, and a distribution question?
3. What summary table does “How many minutes were logged each day?” need before plotting?
4. How can plotting code run without errors and still produce the wrong chart?