Grouping and aggregating

groupby works in three conceptual steps, even though you usually write it as one line:

1. Split — divide the DataFrame into groups based on a column's values (e.g., one group per unique class)
2. Apply — compute something for each group independently (a mean, a count, a max)
3. Combine — collect the per-group results back into a single table

import pandas as pd

df = pd.DataFrame({
    "class": ["A", "A", "B", "B", "B"],
    "score": [88, 92, 79, 95, 61],
})

print(df.groupby("class")["score"].mean())

class
A    90.0
B    78.333333
Name: score, dtype: float64

Beyond .mean(), the same groupby(...)[...] pattern works with several other summaries — and you can request several at once with .agg():

df.groupby("class")["score"].count()   # how many students per class
df.groupby("class")["score"].max()     # top score per class
df.groupby("class")["score"].agg(["mean", "max", "count"])  # all three together

.agg([...]) is worth knowing early — it avoids writing three nearly identical lines when you want three related summaries side by side.

Real questions are often two-dimensional: "average score, by class and by term." Passing a list of column names groups by their combination:

df.groupby(["class", "term"])["score"].mean()

A groupby result sometimes needs to go back into a normal, flat DataFrame — for example, to merge it back with the original data or plot it. .reset_index() does exactly that:

summary = df.groupby("class")["score"].mean().reset_index()
print(summary)

  class      score
0     A  90.000000
1     B  78.333333

Without .reset_index(), class stays as the table's index rather than an ordinary column — usually fine for a quick look, but often inconvenient once you want to combine this result with something else.

Grouping and boolean masking solve related but different problems, and combining them is common: "average score, by class, for students who studied more than 3 hours" is a mask (df[df["hours"] > 3]) followed by a groupby on the filtered result — the two techniques compose naturally rather than competing.

A grouped summary is only as useful as how easily you can read the extremes off it — combining groupby with sort_values from the previous page answers "which group is doing best/worst" in one line:

df.groupby("class")["score"].mean().reset_index().sort_values("score", ascending=False)

Reading this left to right mirrors how you'd describe the question out loud: group by class, take the average score, turn it back into a normal table, then sort so the strongest class is on top. Building comfort with chaining several of these small, well-understood steps together — rather than writing one dense unreadable line from scratch — is a core pandas habit, and it's exactly what the worked example on the next page does with a slightly messier, more realistic dataset.