Selecting Columns

Why column selection matters

Datasets often have dozens of columns. Most analysis focuses on a subset. Selecting the right columns reduces memory usage, speeds up computation, and makes your code clearer.

We will use the Titanic dataset throughout this lesson:

import pandas as pd

# titanic.csv ships with the course — load it from the browser file system.
df = pd.read_csv("titanic.csv")

Selecting a single column

Bracket notation, the most common approach:

ages = df["Age"]
print(type(ages))   # <class 'pandas.core.series.Series'>

Dot notation, shorter but only works when the column name has no spaces or special characters:

print(df.Age.head())   # first 5 ages

Both return a Series (one-dimensional). The column name becomes the Series name, and the DataFrame index is preserved.

Selecting multiple columns

Pass a list of column names inside brackets. This returns a DataFrame, not a Series:

subset = df[["Name", "Age", "Fare"]]
print(type(subset))   # <class 'pandas.core.frame.DataFrame'>
print(subset.head())

Output:

                                                Name   Age     Fare
0                            Braund, Mr. Owen Harris  22.0   7.2500
1  Cumings, Mrs. John Bradley (Florence Briggs Th...  38.0  71.2833
2                             Heikkinen, Miss. Laina  26.0   7.9250
3       Futrelle, Mrs. Jacques Heath (Lily May Peel)  35.0  53.1000
0                           Allen, Mr. William Henry  35.0   8.0500

The order of columns in the list determines the order in the output.

Using loc for column selection

loc selects by label and can handle both rows and columns:

# select all rows, specific columns
subset = df.loc[:, ["Name", "Survived"]]

The : means “all rows.” The list of column names selects specific columns. This is equivalent to df[["Name", "Survived"]] but becomes essential when combining row and column selection in one step.

Practical patterns

Rename after selecting, keep only what you need with clearer names:

demographics = df[["Name", "Age", "Sex"]].copy()
demographics.columns = ["passenger", "age", "gender"]

Select columns by data type, useful when you have many columns:

numeric_cols = df.select_dtypes(include=["number"])
print(numeric_cols.columns.tolist())
# ['PassengerId', 'Survived', 'Pclass', 'Age', 'SibSp', 'Parch', 'Fare']

categorical_cols = df.select_dtypes(include=["object"])
print(categorical_cols.columns.tolist())
# ['Name', 'Ticket', 'Cabin', 'Embarked']

Select columns containing a substring:

# useful for wide datasets with naming conventions
age_cols = [col for col in df.columns if "age" in col.lower()]

When to use what

Try It

From the Titanic dataset, select only the columns Name, Pclass, and Fare. Print the first 5 rows. Then select only the numeric columns and print their column names.

import pandas as pd

# titanic.csv ships with the course — load it from the browser file system.
df = pd.read_csv("titanic.csv")

subset = df[["Name", "Pclass", "Fare"]]
print(subset.head())

numeric = df.select_dtypes(include=["number"])
print(numeric.columns.tolist())

Key Takeaways

• Bracket notation df["col"] is the standard way to select a single column
• df[["col1", "col2"]] returns a DataFrame with multiple columns
• loc becomes essential when combining row and column selection
• select_dtypes() is powerful for selecting columns by data type

Practice Challenge

From the Titanic dataset, create a new DataFrame called passenger_info containing only Name, Age, Sex, and Survived. How many rows have missing Age values in this subset? (Hint: use .isna().sum())