MFKnockoffs.create.fixed: Create fixed-design knockoff variables

In MFKnockoffs: Model-Free Knockoff Filter for Controlled Variable Selection

Description

Creates fixed-design knockoff variables for the original variables.

Usage

MFKnockoffs.create.fixed(X, method = c("sdp", "equi"), sigma = NULL,
  y = NULL, randomize = F)

Arguments

X	normalized n-by-p design matrix (n >= 2p)
method	either 'equi' or 'sdp' (default:'sdp')
sigma	noise level, used to augment the data with extra rows if necessary (default: NULL)
y	vector of observed responses, used to estimate the noise level if 'sigma' is not provided (default: NULL)
randomize	whether the knockoffs are deterministic or randomized (default:False)

Value

An object of class "MFKnockoffs.variables". This object is a list containing at least the following components:

X	The n-by-p matrix of original variables (possibly augmented or transformed)
X_k	The n-by-p matrix of knockoff variables
y	The vector of observed responses (possibly augmented)

References

Barber and Candes, Controlling the false discovery rate via knockoffs. Ann. Statist. 43 (2015), no. 5, 2055–2085. https://projecteuclid.org/euclid.aos/1438606853

Fixed-design knockoff assume a linear regression model for Y|X. Moreover, they only guarantee FDR control with statistics satisfying the "sufficiency" property. In particular, the default statistics with cross-validated lasso does not satisfy this property and should not be used with fixed-design knockoffs.

See Also

Other methods for creating knockoffs: MFKnockoffs.create.approximate_gaussian, MFKnockoffs.create.gaussian

MFKnockoffs documentation built on May 2, 2019, 6:33 a.m.