ko.ordinal {kosel}

Statistics of the knockoffs procedure for ordinalNet regression models.

Description

Returns the vector of statistics W of the revisited knockoffs procedure for regressions available in the R package ordinalNet. Most of the parameters come from ordinalNet(). See ordinalNet documentation for more details.

Usage

ko.ordinal(x, y, family = "cumulative", reverse = FALSE,
  link = "logit", alpha = 1, parallelTerms = TRUE,
  nonparallelTerms = FALSE, nVal = 100, warn = FALSE,
  random = FALSE)

Arguments

x

Covariate matrix, of dimension nobs x nvars; each row is an observation vector. It is recommended that categorical covariates are converted to a set of indicator variables with a variable for each category (i.e. no baseline category); otherwise the choice of baseline category will affect the model fit.

y

Response variable. Can be a factor, ordered factor, or a matrix where each row is a multinomial vector of counts. A weighted fit can be obtained using the matrix option, since the row sums are essentially observation weights. Non-integer matrix entries are allowed.

family

Specifies the type of model family. Options are "cumulative" for cumulative probability, "sratio" for stopping ratio, "cratio" for continuation ratio, and "acat" for adjacent category.

reverse

Logical. If TRUE, then the "backward" form of the model is fit, i.e. the model is defined with response categories in reverse order. For example, the reverse cumulative model with K+1 response categories applies the link function to the cumulative probabilities P(Y >= 2), …, P(Y >= K+1), rather then P(Y <= 1), …, P(Y <= K).

link

Specifies the link function. The options supported are logit, probit, complementary log-log, and cauchit.

alpha

The elastic net mixing parameter, with 0 <= alpha <= 1. alpha=1 corresponds to the lasso penalty, and alpha=0 corresponds to the ridge penalty.

parallelTerms

Logical. If TRUE, then parallel coefficient terms will be included in the model. parallelTerms and nonparallelTerms cannot both be FALSE.

nonparallelTerms

Logical. if TRUE, then nonparallel coefficient terms will be included in the model. parallelTerms and nonparallelTerms cannot both be FALSE. Default is FALSE. nonparallelTerms = TRUE is highly discouraged.

nVal

Length of lambda sequence - default is 100.

warn

Logical. If TRUE, the following warning message is displayed when fitting a cumulative probability model with nonparallelTerms=TRUE (i.e. nonparallel or semi-parallel model). "Warning message: For out-of-sample data, the cumulative probability model with nonparallelTerms=TRUE may predict cumulative probabilities that are not monotone increasing." The warning is displayed by default, but the user may wish to disable it.

random

If TRUE, the matrix of knockoffs is different for every run. If FALSE, a seed is used so that the knockoffs are the same. The default is FALSE.

Value

A vector of dimension nvars corresponding to the statistics W.

Note

nonparallelTerms = TRUE is highly discouraged because the knockoffs procedure does not suit well to this setting.

See Also

ko.sel

Examples

# see ko.sel