Chapter 5 · Part 2

Grouping & aggregating

Here's the move that turns a table into an answer. "What's the average rating per genre?" "How many films per decade?" Every question shaped like "something, per category" is a groupby — the single most powerful idea in pandas. (We're back to the clean movies table.)

Split, apply, combine

groupby works in three steps, and it's worth picturing them: split the rows into groups by some column, apply a summary to each group, then combine the results into a new table.

Start simple — count how many films are in each genre:

count.py

movies["genre"].value_counts()

output

genre
Sci-Fi       4
Drama        2
Thriller     1
Animation    1
Name: count, dtype: int64

value_counts() is the quick one-column shortcut. The general tool is groupby. Group by genre, pick a column, apply a summary — here, the mean rating of each genre:

groupby.py

movies.groupby("genre")["rating"].mean()

output

genre
Animation    8.400
Drama        8.250
Sci-Fi       8.525
Thriller     8.500
Name: rating, dtype: float64

Read it left to right: groupby("genre") splits the rows, ["rating"] picks the column, .mean() is the summary applied to each group. The genre becomes the new index.

Many summaries at once with .agg

Usually you want several numbers per group. .agg names each output column and pairs it with (source_column, function):

agg.py

movies.groupby("genre").agg(
  count=("title", "size"),
  avg_rating=("rating", "mean"),
  total_votes=("votes", "sum"),
)

output

count  avg_rating  total_votes
genre
Animation      1       8.400          520
Drama          2       8.250         1460
Sci-Fi         4       8.525         6750
Thriller       1       8.500          780

That's a pivot table, in four lines. size counts rows in the group; mean, sum, min, max, std all work too.

Sorting the result

Aggregations come out sorted by the group label. To rank by a value instead, chain sort_values:

sort.py

movies.sort_values("votes", ascending=False)[["title", "votes"]].head(4)

output

title  votes
0     Inception   2400
1    The Matrix   1900
2  Interstellar   1800
4      Whiplash    900

🎯 Your turn

For each genre, compute both the number of films (n) and the average rating (avg), in a single groupby.

You now know every core move. Time to chain them into a real analysis.