#Apply computation within groups in a pandas DataFrame

One useful capability of pandas is to execute computations within groups of a DataFrame. This is achieved using groupby() followed by transform().

# Import libraries
import pandas as pd
import seaborn as sns

# Load sample data in a DataFrame
df = (
    sns.load_dataset('iris')
    .sample(n=12, random_state=24)
    .sort_values('species')
    .reset_index(drop=True)
    [['species', 'sepal_width']]
)
df

##Compute mean at group-level

To get group-level statistics like mean, sum or count, use transform('function').

# Get group mean
df.assign(mean=
    df.groupby('species').transform('mean')
)

##Standardize values

You can center values inside a group by substracting the group mean to each row.

# Standardize values
df.assign(standardized=
    df.groupby('species').transform(lambda x: x - x.mean())
)

##Rank values inside groups

Besides computing group-wise values, you can also rank values within each group.

# Rank values
df.assign(rank=
    df.groupby('species').transform(lambda x: x.rank())
)

##Error when output has multiple columns

If the result of the transform generates more than 1 column, and you try to assign it to a column of an existing DataFrame, you will encounter an error like ValueError: Wrong number of items passed X, placement implies 1. To avoid this, pass only one column in the result:

df.assign(rank=
    df.groupby('species').transform(lambda x: x.rank()).iloc[:, 0]
)