Prediction method for Orthogonalizing EM fitted objects

# S3 method for oem
predict(
  object,
  newx,
  s = NULL,
  which.model = 1,
  type = c("link", "response", "coefficients", "nonzero", "class"),
  ...
)

Arguments

object	fitted "oem" model object
newx	Matrix of new values for `x` at which predictions are to be made. Must be a matrix; can be sparse as in the `CsparseMatrix` objects of the Matrix package. This argument is not used for `type=c("coefficients","nonzero")`
s	Value(s) of the penalty parameter lambda at which predictions are required. Default is the entire sequence used to create the model.
which.model	If multiple penalties are fit and returned in the same oem object, the `which.model` argument is used to specify which model to make predictions for. For example, if the oem object `oemobj` was fit with argument `penalty = c("lasso", "grp.lasso")`, then which.model = 2 provides predictions for the group lasso model.
type	Type of prediction required. `type = "link"` gives the linear predictors for the `"binomial"` model; for `"gaussian"` models it gives the fitted values. `type = "response"` gives the fitted probabilities for `"binomial"`. `type = "coefficients"` computes the coefficients at the requested values for `s`. `type = "class"` applies only to `"binomial"` and produces the class label corresponding to the maximum probability.
...	not used

Value

An object depending on the type argument

Examples

set.seed(123)
n.obs <- 1e4
n.vars <- 100
n.obs.test <- 1e3

true.beta <- c(runif(15, -0.5, 0.5), rep(0, n.vars - 15))

x <- matrix(rnorm(n.obs * n.vars), n.obs, n.vars)
y <- rnorm(n.obs, sd = 3) + x %*% true.beta
x.test <- matrix(rnorm(n.obs.test * n.vars), n.obs.test, n.vars)
y.test <- rnorm(n.obs.test, sd = 3) + x.test %*% true.beta

fit <- oem(x = x, y = y,
           penalty = c("lasso", "grp.lasso"),
           groups = rep(1:10, each = 10),
           nlambda = 10)

preds.lasso <- predict(fit, newx = x.test, type = "response", which.model = 1)
preds.grp.lasso <- predict(fit, newx = x.test, type = "response", which.model = 2)

apply(preds.lasso,     2, function(x) mean((y.test - x) ^ 2))
#>  [1] 10.206086  9.305604  9.114948  9.099376  9.122277  9.142751  9.150690
#>  [8]  9.153957  9.155163  9.155600
apply(preds.grp.lasso, 2, function(x) mean((y.test - x) ^ 2))
#>  [1] 10.206086  9.594635  9.141101  9.091859  9.119994  9.141259  9.150381
#>  [8]  9.153852  9.155125  9.155586