MFKnockoffs: Model-free controlled variable selection
In MFKnockoffs: Model-Free Knockoff Filter for Controlled Variable Selection

Model-free knockoffs provide a general and powerful tool to perform high-dimensional controlled variable selection.

The method constructs artificial 'knockoff copies' of the variables in a statistical model and then selects those variables that are clearly better than their corresponding fake copies. A wide range of statistics and machine learning tools can be exploited to estimate the importance of each feature, while guaranteeing finite-sample control of the false discovery rate (FDR). This model-free approach makes it possible to use knockoffs for data originating from any conditional model (Y|X), no matter how high-dimensional, provided that the distribution of the covariates (X) is known.

For more information, see the website below and the accompanying paper.

https://statweb.stanford.edu/~candes/MF_Knockoffs/index.html

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