Fits multiple penalized regression models in which the residuals are randomly permuted, thereby allowing estimation of the marginal false discovery rate.
permres(fit, ...)
# S3 method for ncvreg
permres(fit, lambda, N=10, seed, trace=FALSE, ...)A list with the following components:
The number of variables selected at each value of
lambda, averaged over the permutation fits.
The actual number of selected variables for the non-permuted data.
The estimated marginal false discovery rate (EF/S).
The loss/deviance, averaged over the permutation fits. This is an estimate of the explanatory power of the model under null conditions, and can be used to adjust the loss of the fitted model in a manner akin to the idea of an adjusted R-squared in classical regression.
A fitted ncvreg model, as produced by
ncvreg(). To use with permres, the model must
be fit using the returnX=TRUE option.
The regularization parameter to use for estimating
residuals. Unlike perm.ncvreg, permres
calculates EF and mFDR for a specific lambda value, not an
entire path. As a result, it runs much faster.
The number of permutation replications. Default is 10.
You may set the seed of the random number generator in order to obtain reproducible results.
If set to TRUE, perm.ncvreg will inform the user of its progress by announcing the beginning of each permutation fit. Default is FALSE.
Not used.
Patrick Breheny <patrick-breheny@uiowa.edu>
The function fits a penalized regression model to the actual data,
then repeats the process N times with a permuted version of the
response vector. This allows estimation of the expected number of
variables included by chance for each value of lambda. The
ratio of this expected quantity to the number of selected variables
using the actual (non-permuted) response is called the marginal false
discovery rate (mFDR).
ncvreg, mfdr, perm.ncvreg
data(Prostate)
fit <- ncvreg(Prostate$X, Prostate$y, N=50)
permres(fit, lambda=0.15)
Run the code above in your browser using DataLab