Pandas dataframe fillna() only some columns in place

Look ma, no inplace!

inplace=True sounds awesome, but it can cause a mess in the sandpit - yep, irreversible changes. Try this neat trick instead:

df[['col1', 'col2']] = df[['col1', 'col2']].fillna(value=0)


Fills col1 and col2 NaNs with zeros, keeps other toys (columns) intact.

Preserving data: The secret sauce

For the Data-Integrity Chefs out there, back up before changing things. When not using inplace, you can actually preview what's cooking:

df_updated = df.fillna({'col1': val1, 'col2': val2})
# taste the sauce (review df_updated) ...
df = df_updated  # perfect? Serve it hot!


Custom fillna values: NaNs are picky eaters

For NaNs, one size does not fit all. Numeric NaNs don't enjoy being a boring string, nor vice versa. Coddle them with the right diet:

fill_values = {'numeric_col': 0, 'string_col': 'unknown'}
df.fillna(fill_values, inplace=True)


Dodging bullets: Avoid SettingWithCopyWarning

The pesky SettingWithCopyWarning is trigger happy when changing DataFrame copy instead of the original. Dodge this:

df.loc[:, 'col1'] = df['col1'].fillna(val1)
df.loc[:, 'col2'] = df['col2'].fillna(val2)


Remember: Be the Matrix, be the DataFrame.

No Metaphor Zone

Applying .fillna() selectively == custom-made spanners for each missing bolt in the DataFrame-Engine.

dataframe['column'].fillna('value', inplace=True); // Spanner fits bolt. Engine purrs.


MULTIPLE columns, too:

Before: Fast bolt <=> Spanner (Match! 🔨),
Missing bolt 🛠 (Better luck next time)

After: Fast Bolt <=> Spanner (Nice!🔨), Missing Bolt <=> New spanner (Nailed it!🔧)

Assignment: The Picasso of Pandas

Paint your DataFrame with direct assignments instead of inplace splatters. The art here: easy undo, easy viewing.

Text data: NaNs don't enjoy being called '0'

Strings and categoricals squirm at numeric fill values. Serenade them with suitable string-fillers:

df['category_col'] = df['category_col'].fillna('Missing')


Because, every NaN is unique.. 😊