How to Perform Bootstrapping in Python

Spread the love


Bootstrapping is a powerful statistical technique used to estimate the sampling distribution of a statistic by resampling with replacement from the original data. It can be applied to almost any data analysis without making strong assumptions about the population or the sampling distribution.

This article will provide a comprehensive guide on how to perform bootstrapping in Python. It will cover the basics of bootstrapping, libraries needed, how to write a simple bootstrap function, how to use bootstrapping for estimating confidence intervals and validating models, and the pros and cons of this method.

The Basics of Bootstrapping

Bootstrapping is a resampling technique based on random sampling with replacement. This method allows estimation of the sampling distribution of almost any statistic using random sampling methods.

Generally, the process of bootstrapping involves repeatedly sampling observations from the original data, with replacement, and recalculating the statistic of interest for each sample. The result is an empirical approximation of the sampling distribution of the statistic.

Libraries Needed for Bootstrapping

The Python libraries we’ll use for bootstrapping include:

  • NumPy: A fundamental library for high-performance numerical computation in Python.
  • Pandas: A powerful data manipulation library in Python.
  • SciPy: A library for scientific computation that provides functions to perform statistical analysis.
  • Matplotlib: A plotting library for creating static, animated, and interactive visualizations in Python.
  • Seaborn: A statistical data visualization library based on Matplotlib. It provides a high-level interface for creating attractive graphs.
  • SKLearn: A machine learning library in Python.

You can install these libraries using pip:

pip install numpy pandas scipy matplotlib seaborn sklearn

Simple Bootstrap Function

We’ll start by writing a simple function to perform bootstrapping:

import numpy as np

def bootstrap(data, n_bootstrap_samples=1000):
    return [np.random.choice(data, size=len(data), replace=True) for _ in range(n_bootstrap_samples)]

This function takes a dataset and a number of bootstrap samples as input. It generates the specified number of bootstrap samples from the dataset and returns these samples in a list.

Each bootstrap sample has the same size as the original dataset and is drawn with replacement, meaning the same observation can appear more than once in the sample.

Bootstrapping for Estimating Confidence Intervals

Now let’s see how we can use bootstrapping to estimate a confidence interval for the mean of a dataset. Confidence intervals calculated from bootstrap samples are often more accurate than those calculated under strong distributional assumptions.

Let’s say we have the following data:

import numpy as np

data = np.array([3.1, 2.3, 3.7, 3.2, 2.8, 3.0, 2.5, 2.9, 3.4, 3.6])

We can create bootstrap samples and calculate the mean of each sample:

bootstrap_samples = bootstrap(data)
bootstrap_means = [np.mean(sample) for sample in bootstrap_samples]

Then, we can calculate a 95% confidence interval for the mean:

lower = np.percentile(bootstrap_means, 2.5)
upper = np.percentile(bootstrap_means, 97.5)
confidence_interval = (lower, upper)

Here, we’re using the percentile() function from NumPy to get the 2.5th and 97.5th percentiles of the bootstrap means, which form the lower and upper bounds of the 95% confidence interval.

Bootstrapping for Model Validation

Bootstrapping can also be used to validate statistical models. One common application is in estimating the prediction error of a machine learning model.

Let’s say we have a classification model and a dataset:

from sklearn.datasets import load_breast_cancer
from sklearn.ensemble import RandomForestClassifier
from sklearn.model_selection import train_test_split
from sklearn.metrics import accuracy_score

# Load the data
data = load_breast_cancer()
X =
y =

# Split the data into training and test sets
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.2, random_state=42)

# Fit the model
model = RandomForestClassifier(random_state=42), y_train)

# Predict the labels for the test set
y_pred = model.predict(X_test)

# Compute the accuracy of the model
accuracy = accuracy_score(y_test, y_pred)

We can use bootstrapping to estimate the variability of the accuracy:

# Generate bootstrap samples of the predictions
bootstrap_preds = bootstrap(y_pred)

# Compute the accuracy for each bootstrap sample
bootstrap_accuracies = [accuracy_score(y_test, pred) for pred in bootstrap_preds]

# Calculate a 95% confidence interval for the accuracy
lower = np.percentile(bootstrap_accuracies, 2.5)
upper = np.percentile(bootstrap_accuracies, 97.5)
confidence_interval = (lower, upper)

Pros and Cons of Bootstrapping

Bootstrapping is a versatile and powerful statistical tool. Its major advantages include:

  • Fewer assumptions: Bootstrapping makes no assumptions about the population or the shape of the distribution. This makes it applicable to many statistical problems.
  • Simplicity: The bootstrap method is simple to understand and implement. It requires no complex mathematical calculations.
  • Flexibility: Bootstrapping can be applied to almost any estimator and any sample size.

Despite these advantages, there are some limitations to bootstrapping:

  • Computationally intensive: As the size of the dataset increases, the computational resources and time required for bootstrapping also increase.
  • Accuracy: Although bootstrapping is generally accurate, it can be less accurate when the sample size is small or when the data are not well-behaved (e.g., with outliers).
  • Not a substitute for a good study design: Like any statistical method, bootstrapping is not a substitute for a well-designed study. If the original data are biased, the bootstrap samples will also be biased.


In this article, we’ve covered a comprehensive guide on how to perform bootstrapping in Python. We’ve seen how bootstrapping can be a simple, flexible, and powerful tool for estimating confidence intervals and validating models. However, like all statistical methods, it’s important to be aware of its limitations and to use it appropriately in the context of a well-designed study.

Leave a Reply