reg_fs: Regression-based feature selection
In prpatil/tdsm: Templated Deterministic Statistical Machines

Description Usage Arguments Details Value

Leverages the fact that the F-statistic for y~x and x~y is the same to quickly choose pairs that add conditional value for predicting an outcome. Also runs a separate pair-on-pair regression to eliminate collinear pairs.

1	reg_fs(pairmat, outcome, covar = NULL, npair = 5)

`pairmat`	An m x n matrix of pairwise features with rows=features, columns=samples (generated by empirical control feature selection)
`outcome`	A vector of outcomes of length n
`covar`	An optional n x p matrix of additional covariates to adjust for
`npair`	The number of paris we wish to select

This function takes the output from empirical_controls.R and reduces it to a set of pairs of size "npairs" that are predictive of the outcome of interest. The paris are chosen conditionally so that the next pair chosen adds predictive ability additional to the pairs that have already been chosen. At each step, we consider a regression of

pairs A vector with the index for each chosen pair

prpatil/tdsm documentation built on May 26, 2019, 10:32 a.m.