predict.relaxnet


The <code>"relaxnet"</code> object from which to get predictions.


object


Matrix of new values for <code>x</code> at which predictions are to be made. Must be a matrix; can be sparse as in <code>Matrix</code> package. This argument is not used for <code>type=c("coefficients","nonzero")</code>


newx


Specifies the submodel from which predictions are required. <code>"main"</code> indicates the main glmnet model, while an integer indicates one of the relaxed models.


which.model


Value(s) of the penalty parameter <code>lambda</code> at which predictions are required. Default is the entire sequence of lambda values for the model specified by <code>which.model</code>.



See <code>link[glmnet]{predict.glmnet}</code>.


type


Only the default, <code>FALSE</code>, is supported. See <code>link[glmnet]{predict.glmnet}</code>.


exact


Further arguments passed to <code>predict.glmnet</code>. In the current version, these are not guaranteed to work correctly (for example, <code>offset</code> has not yet been implemented for relaxnet).



Similar to other predict methods, this functions predicts fitted values, logits,
coefficients and more from a fitted <code>"relaxnet"</code> object.


Extends the glmnet package with "relaxation", done by running glmnet once on the entire predictor matrix, then again on each different subset of variables from along the regularization path. Relaxation may lead to improved prediction accuracy for truly sparse data generating models, as well as fewer false positives (i.e. fewer noncontributing predictors in the final model). Penalty may be lasso (alpha = 1) or elastic net (0 < alpha < 1). For this version, family may be "gaussian" or "binomial" only. Takes advantage of fast FORTRAN code from the glmnet package.

predict.relaxnet: Predict Method for `"relaxnet"` Objects

Description

Usage

Arguments

Value

See Also