Selecting multiple columns in a Pandas dataframe

Selective column choices: a touch of finesse

Let's explore some high-demand column grabbing techniques that Pythonistas need in daily coding.

Select non-sequential columns, they're not shy!

To grab non-adjacent columns, list them within brackets:

df_sub = df[['ColumnA', 'ColumnX']] # Column hoppin', why not?


Boolean masks: Scooby-doo, where are you?

For selecting columns based on some criteria, a Boolean mask is your best friend:

mask = df.columns.isin(['ColumnB', 'ColumnC', 'ColumnD'])
filtered_df = df.loc[:, mask] # Yes, indeed, elementwise magic!


With this Boolean array indexing, you can keep your code flexible and readable.

Views or copies: Identity crisis 101

Grasping when you're dealing with a view or a copy is as nuanced as coffee tasting.

Avoid nasty SettingWithCopyWarning

To minimize stress with SettingWithCopyWarning, make a copy. No more accidental overwrites of your original dataframe:

df_new_copy = df[['Column1', 'Column2']].copy() # New, shiny, mine!


Column selection by index position: Counting matters

To select by index position, embrace iloc. Remember, column counting starts at zero:

df_sub = df.iloc[:, 0:2].copy() # Copies first two columns. Magic, right? 


Remember: In Python, the last is always left behind. So 0:2 nets columns at index 0 and 1.

Power tricks: Show off with style

The .columns property and get_loc function are your secret weapons for column selection.

Dispatch a dictionary of column positions

For frequent column position needs:

column_positions = {df.columns.get_loc(c): c for c in df.columns} # Because names are too mainstream!


This dictionary, besides proving you're cool, doubles down on position-name reference.

Don't mix slicing syntax with column selection

# Incorrect - Selects rows, not columns. Ouch!
df_rows = df['Column1':'Column2']


Save yourself! Leverage loc or list syntax for the column selection.

Master curated DataFrame construction

Creating a brand-new DataFrame from existing columns is like curating an art exhibit.

A condensed DataFrame - less is more

columns = ['Column1', 'Column2']
concise_df = pd.DataFrame(df, columns=columns) # Nicely trimmed DataFrame!


The constructor shrinks the DataFrame to only the specified columns. So, it's like an art exhibit with your best works.

Avoid name conflicts: Narrow escape

Keep column names unique and steer clear of DataFrame methods such as index to avoid unexpected behaivors.