Level 3 · Pick the columns you need

One column: a Series

Square brackets with a column name give you that column:

customers["email"]

The result is a Series: one column of values plus the index. Series have their own useful methods, like .nunique(), .value_counts() or .str.lower(), which you’ll use all the time.

You may also see customers.email (dot access). It works for simple names, but it breaks for names with spaces or names that clash with DataFrame methods (a column called count or size), and it can’t create new columns. Bracket access always works, so make it your habit.

Several columns: a DataFrame

To select more than one column, pass a list of names:

customers[["customer_id", "email", "state", "segment"]]

The double brackets confuse everyone at first. The outer brackets mean “select from this DataFrame”. The inner brackets are just Python’s list syntax. The same thing written in two steps makes this clearer:

cols = ["customer_id", "email", "state", "segment"]
customers[cols]

The columns come back in the order you list them, so this is also how you reorder columns.

One column as a DataFrame

A list with a single name gives you a one-column DataFrame instead of a Series:

customers["email"]       # Series
customers[["email"]]     # DataFrame with one column

Which one you want depends on what comes next. Series methods (value_counts, .str) want a Series; exporting a table or merging wants a DataFrame.

Selecting columns does not copy the table

The selection is a new object. customers itself still has all nine columns. If you want to keep the narrower table, assign it:

contacts = customers[["customer_id", "email", "state", "segment"]]

If you’re going to modify that new table (add a column, change values), make it an explicit copy: contacts = customers[[...]].copy(). That avoids pandas’ SettingWithCopyWarning and makes it clear the two tables are independent.

Finding columns by pattern

When a table has dozens of columns, filter helps:

orders.filter(like="_id")    # every column whose name contains "_id"

Why bother narrowing?

Selecting only what you need makes outputs readable, makes exports smaller, and is a privacy habit worth having: a marketing audience doesn’t need signup dates, and a shared spreadsheet shouldn’t carry columns nobody asked for.

Common mistakes

• Single brackets around a list of names written without the inner list: customers["customer_id", "email"] raises KeyError, because pandas looks for one column named by that pair.
• Typos and capitalisation: customers["Email"] raises KeyError: 'Email'.
• Expecting customers[["email"]] to be a Series and calling .str on it.

If you know SQL

customers[["customer_id", "email"]] is SELECT customer_id, email FROM customers.