extendedmosaicperm 0.1.0

Extensions to mosaicperm: sign-flip inference, ridge-regularized residuals, and adaptive tiling for factor models.

extendedmosaicperm

Extended tools for Mosaic Permutation Tests — including:

• ✓ sign-flip inference
• ✓ ridge-regularized residuals
• ✓ adaptive tiling
• ✓ experiment utilities

All built as a thin, clean extension on top of the original mosaicperm package.

The goal is to explore faster, more robust, and more flexible randomization inference procedures for high-dimensional factor models.

---

Installation

Stable version (PyPI)

pip install extendedmosaicperm

Development version (GitHub)

pip install git+https://github.com/skonieczkak/extendedmosaicperm.git

---

Usage Examples

Basic — Sign-flip test

import numpy as np
from extendedmosaicperm.factor import ExtendMosaicFactorTest
import mosaicperm as mp

rng = np.random.default_rng(0)
T, p, k = 200, 50, 3

Y = rng.normal(size=(T, p))
L = rng.normal(size=(p, k))

test = ExtendMosaicFactorTest(
    outcomes=Y,
    exposures=L,
    test_stat=mp.statistics.mean_maxcorr_stat,
    sign_flipping=True,
)

test.fit(nrand=500)
print("p-value:", test.pval)

Adaptive tiling

from extendedmosaicperm.tilings import build_adaptive_tiling

tiling = build_adaptive_tiling(
    outcomes=Y,
    exposures=L,
    batch_size=20,
    seed=0
)

print(len(tiling.tiles))

Monte Carlo experiment

from extendedmosaicperm.experiments.sign_flip import SignFlipExperiment

exp = SignFlipExperiment(
    n_sims=100,
    nrand=200,
    seed=123
)

exp.run()
df = exp.summarize()
print(df.head())

---

Documentation

Full documentation, including the API reference, usage examples, and theoretical background, is available at:

https://extendedmosaicperm.readthedocs.io

---

Testing

Run the unit tests:

pytest extendedmosaicperm/tests

License

MIT License — same as the parent mosaicperm project.

---