dict_column_explosion_benchmark.py

import pandas as pd 
import numpy as np
import perfplot
from string import ascii_lowercase as LOWER, ascii_uppercase as UPPER
import random

# Note: The copy() calls are needed here because `pop()` mutates the dataframe inplace
# so it is essential to make a copy() we don't want to mutate the output across runs
def apply_drop(df):
    return df.join(df['val'].apply(pd.Series),).drop('val', axis=1)  

def json_normalise_drop(df):
    return df.join(pd.json_normalize(df['val'])).drop('val', axis=1)  

def tolist_drop(df):
    return df.join(pd.DataFrame(df['val'].tolist())).drop('val', axis=1)  

random.seed(0)
letters = (LOWER + UPPER)[::2]

M, N = 10, 10
df = pd.DataFrame({'idx': np.arange(N), 'val': [{v: k for k, v in enumerate(random.sample(letters, M), 1)} for _ in range(N)]})

kernels = [apply_drop, json_normalise_drop, tolist_drop]

perfplot.show(
    setup=lambda n: pd.concat([df] * n, ignore_index=True),
    kernels=kernels,
    labels=[str(k.__name__) for k in kernels],
    n_range=[2**k for k in range(12)],
    xlabel='N',
    logx=True,
    logy=True,
    equality_check=lambda df1, df2: df1.equals(df2))