Integrations¶

pandas accessor¶

When pandas is installed, bootstrapx registers a .bootstrap accessor.

Series¶

import pandas as pd
import numpy as np
import bootstrapx

s = pd.Series(np.random.default_rng(0).normal(5, 2, 300))
r = s.bootstrap.bca(np.mean, n_resamples=4999, random_state=42)
print(r)

DataFrame summary¶

import pandas as pd
import numpy as np
import bootstrapx

rng = np.random.default_rng(0)
df = pd.DataFrame({
    "control": rng.normal(0, 1, 200),
    "treatment": rng.normal(0.2, 1, 200),
})

summary = df.bootstrap.summary(np.mean, n_resamples=2999, random_state=42)
print(summary)

scikit-learn¶

BootstrapCV implements a scikit-learn compatible cross-validator using bootstrap training samples and out-of-bag evaluation.

from bootstrapx import BootstrapCV
from sklearn.datasets import load_breast_cancer
from sklearn.ensemble import GradientBoostingClassifier
from sklearn.model_selection import cross_val_score

X, y = load_breast_cancer(return_X_y=True)
cv = BootstrapCV(n_splits=200, random_state=42)

scores = cross_val_score(
    GradientBoostingClassifier(n_estimators=100, random_state=0),
    X, y, cv=cv, scoring="roc_auc"
)
print(scores.mean(), scores.std())

Backend guidance¶

bootstrapx is CPU-first by design so it works on most developer and production machines with a normal pip install.

Recommended deployment baseline:

NumPy + SciPy core for correctness and portability.
Optional pandas for tabular workflows.
Optional scikit-learn for model evaluation.
Optional numba only where it helps CPU throughput.