Slicing columns

Slicing columns in pandas is one of the most frequent operations — whether you're narrowing down features for modeling, selecting metrics for reporting, or cleaning up wide datasets. There are two main ways to slice columns: label-based with .loc[] (recommended for readability) and position-based with .iloc[] (fast for positional work). Both support slicing with start:stop:step syntax, just like lists.

Here’s a practical guide with real examples you can copy and adapt — including how to avoid common mistakes in 2026.

1. Setup & Sample Data


import pandas as pd

data = {
    'Name': ['John', 'Mary', 'Peter', 'Anna', 'Mike'],
    'Age': [25, 32, 18, 47, 23],
    'Salary': [50000, 80000, 35000, 65000, 45000],
    'Department': ['Sales', 'Engineering', 'Sales', 'HR', 'Engineering'],
    'Bonus': [5000, 8000, 3000, 7000, 4000]
}

df = pd.DataFrame(data)
print(df.columns)  # Index(['Name', 'Age', 'Salary', 'Department', 'Bonus'], dtype='object')

2. Label-Based Column Slicing with .loc[] (Recommended)

Use column names — works even if index is unsorted or columns are reordered.


# Columns from 'Age' to 'Department' (inclusive)
age_to_dept = df.loc[:, 'Age':'Department']
print(age_to_dept.columns)  # ['Age', 'Salary', 'Department']

# Every other column from 'Age' to 'Bonus'
every_other = df.loc[:, 'Age':'Bonus':2]
print(every_other.columns)  # ['Age', 'Department']

# From 'Salary' to end
salary_onward = df.loc[:, 'Salary':]
print(salary_onward.columns)  # ['Salary', 'Department', 'Bonus']

3. Position-Based Column Slicing with .iloc[]

Use integer positions — ignores column names, faster for positional work.


# Columns 1 to 4 (Age, Salary, Department, Bonus) — stop is exclusive
pos_slice = df.iloc[:, 1:4]
print(pos_slice.columns)  # ['Age', 'Salary', 'Department']

# Every second column from position 1 (Age) to end
every_second = df.iloc[:, 1::2]
print(every_second.columns)  # ['Age', 'Department']

# Last 2 columns
last_two = df.iloc[:, -2:]
print(last_two.columns)  # ['Department', 'Bonus']

4. Real-World Use Cases (2026 Examples)

Selecting feature subsets for modeling


# Numeric features only (assuming known positions)
features = df.iloc[:, 1:3]  # Age and Salary

Excluding unwanted columns


# Drop 'Name' and 'Bonus' ? keep everything between
core_df = df.loc[:, 'Age':'Department']

Every other column (e.g., for sampling or testing)


sampled_cols = df.iloc[:, ::2]  # every second column

5. Modern Alternative in 2026: Polars

For large datasets, Polars is often faster and more memory-efficient — slicing syntax is very similar.


import polars as pl

df_pl = pl.DataFrame(data)
# Slice by position (0-based)
subset_pl = df_pl[:, 1:4]
print(subset_pl.columns)  # ['Age', 'Salary', 'Department']

Best Practices & Common Pitfalls

Prefer .loc[] with column names — more readable and safe when columns are reordered
Use .iloc[] only for positional work (e.g., first 5 columns, last 3)
Stop is exclusive — df.loc[:, 'A':'C'] includes C
Negative step reverses order — df.loc[:, ::-1] reverses columns
Always check column order with df.columns before positional slicing
For huge data, prefer Polars slicing — it's faster and more memory-efficient

Conclusion

Slicing columns with .loc[start:end:step] or .iloc[start:end:step] is a core pandas skill — it lets you quickly narrow focus to the data that matters. In 2026, prefer label-based .loc[] for clarity and safety, use .iloc[] for positional tasks, and reach for Polars when performance matters. Master slicing syntax, inclusive/exclusive rules, and negative steps, and you'll manipulate DataFrames with confidence and speed.

Next time you need a subset of columns — slice with .loc or .iloc first.