fanok 0.0.3

Knockoffs in Python

FANOK: Knockoffs in linear time

FANOK is a Python implementation of the Gaussian knockoffs framework developed by Barber-Candès [1] [2]. It provides fast algorithms based on coordinate ascent to generate Gaussian knockoffs in high dimensions.

Installation

Requirements

This package requires NumPy, Scipy, Scikit-Learn. Optionally, CVXPY is required in order to solve the SDP with external solvers, or to run the tests. Use pip install requirements.txt to install them.

Installation

pip install fanok

Documentation

(Here)[https://fanok.readthedocs.io/en/latest/]

Usage

Here is a minimal usage example:

from sklearn.datasets import make_regression
from fanok import GaussianKnockoffs, KnockoffSelector
from fanok.statistics import EstimatorStatistics

X, y, coef = make_regression(n_samples=100, n_features=150, n_informative=20, coef=True)

knockoffs = GaussianKnockoffs()
statistics = EstimatorStatistics()
selector = KnockoffSelector(knockoffs, statistics, alpha=0.2, offset=1)
selector.fit(X, y)

fdp, power = selector.score(X, y, coef)
print(f"FDP: {fdp}, Power: {power}")

See the folder examples/ for more illustrations, and in particular with fixed and low-rank knockoffs. We also provide a notebook with an example on text data.

References

[1] Barber, R. F. and Candès, E. J. (2015). Controlling the false discovery rate via knockoffs. Ann. Statist., 43(5):2055–2085.

[2] Candès, Emmanuel & Fan, Yingying & Janson, Lucas & Lv, Jinchi. (2016). Panning for Gold: Model-free Knockoffs for High-dimensional Controlled Variable Selection. Journal of the Royal Statistical Society: Series B (Statistical Methodology). 80. 10.1111/rssb.12265.