Ez Bootstrap

Ez bootstrap is a python package for performing bootstrap statistics. Read more on the docs.

Installation

It can be installed using pip in a terminal or command prompt.

$ pip install ezbootstrap

Usage

Functions in ezbootstrap generate bootstrap replicates from arbitrary input functions.

Generate bootstrap condidence intervals

from ezbootstrap import bs_1sample
import numpy as np
import scipy.stats

data = scipy.stats.norm(loc=100, scale=10).rvs(50)
reps = bs_1sample(
    x=data,
    func=lambda x: np.mean(x),
    size=10000
)
confint = np.percentile(reps, [0.025, 0.975])
print(f"Confidence interval for the mean of x: {confint}")

Perform permutation tests for difference of mean

from ezbootstrap import permutation_2sample

x = scipy.stats.norm(loc=100, scale=10).rvs(100)
y = scipy.stats.norm(loc=90, scale=10).rvs(100)

observed = np.mean(x) - np.mean(y)
reps = permutation_2sample(
    x=x,
    y=y,
    func=lambda x, y: np.mean(x) - np.mean(y),
    size=5000
)
p = np.mean(np.abs(reps) >= np.abs(observed)) * 2

Perform Hypothesis tests between two samples with arbitrary functions

from ezbootstrap import bs_2sample
from sklearn.metrics import jaccard_score

x = np.random.choice([0, 1], 1000)
y = np.random.choice([0, 1], 1000)

observed = jaccard_score(x, y)
reps = bs_2sample(
    x=x,
    y=y,
    func=lambda x, y: jaccard_score(x, y),
    size=5000
)
p = np.mean(reps >= observed)

Performance

ezbootstrap uses numba's just-in-time compiler to speed up bootstrap replicate generation. If compilation fails, it will revert to python which will be slower.

To increase odds of numba compilation succeeding, pass functions written in pure python or only using numpy functions.

Acknowledgement

Justin Bois's dc_stat_think was the inspiration for this package. ezbootstrap extends dc_stat_think by permitting users to generate bootstrap replicates with arbitrary functions.

Ruairi-osul/ezbootstrap