Applying function with multiple arguments to create a new pandas column

Basic column-wise operations

If you're eyeing for element-wise operations, you don't need a sledgehammer to crack a nut. Simple mathematics works wonders:

df['new_col'] = df['col1'] * df['col2']  # Multiply like there's no tomorrow


The power of numpy vectorization

Why walk when you can fly? Vectorizing your function using numpy can lead to significant performance advantages. Here's your express ticket to efficiency city:

import numpy as np

np_func = np.vectorize(my_func)  # Vectorizing: because for-loops are too mainstream
df['new_col'] = np_func(df['col1'], df['col2'])


And don't forget about numpy's shiny tool multiply for element-wise multiplication:

df['new_col'] = np.multiply(df['col1'], df['col2'])  # Multiplication just got cooler


Managing functions with multiple returns

Does your function return multiple values? No problem, you can tackle all of them at once:

df['new_col1'], df['new_col2'] = zip(*df.apply(lambda x: my_multi_value_func(x['col1'], x['col2']), axis=1))  # Unzipping the knowledge


Row-wise operations with apply

When using apply, remember to use axis=1 so your operation rolls on rows and not columns:

df['new_col'] = df.apply(lambda x: my_func(x["col1"], x["col2"]), axis=1)  # Riding the row roller-coaster


Custom functions for complex logic

Complex logic feels at home in a custom function. Encapsulate your custom logic, and use row-wise apply:

def custom_logic(row):  # This function wears the thinking hat
    # Complex logic goes here
    return result

df['new_col'] = df.apply(lambda x: custom_logic(x), axis=1)


Creating multiple new columns in one shot

If your function outputs more than one value and you want to store them as new columns, split the tuple and conquer:

def get_multiple_metrics(row):  # Hardworking function with multiple outputs
    # Return a tuple
    return metric1, metric2

# Unpack results into new columns
df[['metric1', 'metric2']] = df.apply(lambda x: get_multiple_metrics(x), axis=1, result_type='expand')  # Talk about efficiency!


Apply with care: Handling data diversity

Make sure your function handles data diversity accordingly. It matters when you're dealing with datasets that include different data types or missing values—unless you want "unexpected" to be your middle name.