fwelnet: Fit feature-weighted elastic net
In kjytay/fwelnet: Feature-Weighted Elastic Net

Description Usage Arguments Details Value Examples

View source: R/fwelnet.R

Fit a model with feature-weighted elastic net for a path of lambda values. Fits linear and logistic regression models.

fwelnet(
  x,
  y,
  z,
  lambda = NULL,
  family = c("gaussian", "binomial"),
  alpha = 1,
  standardize = TRUE,
  max_iter = 1,
  ave_mode = 1,
  thresh_mode = 1,
  t = 1,
  a = 0.5,
  thresh = 1e-04,
  verbose = FALSE
)

`x`	Input matrix, of dimension `nobs x nvars`; each row is an observation vector.
`y`	Response variable. Quantitative for `family = "gaussian"`. For `family="binomial"`, should be a numeric vector consisting of 0s and 1s.
`z`	Feature of features matrix, with dimension `nvars x nfeaturevars`.
`lambda`	A user supplied `lambda` sequence. Typical usage is to have the program compute its own `lambda` sequence; supplying a value of lambda overrides this.
`family`	Response type. Either `"gaussian"` (default) for linear regression or `"binomial"` for logistic regression.
`alpha`	The elastic net mixing hyperparameter, a real value number between 0 and 1 (inclusive). Default value is 1.
`standardize`	If `TRUE`, the columns of the input matrix are standardized before the algorithm is run. Default is `TRUE`.
`max_iter`	The number of iterations for the optimization. Default is 1.
`ave_mode`	If equal to 1 (default), the gradient descent direction for `theta` is the mean gradient across the lambda values. If equal to 2, it is the component-wise median gradient across the lambda values.
`thresh_mode`	If equal to 1 (default), backtracking line search for `theta` is done so that the mean objective function (across lambda values) decreases. If equal to 2, it is done so that the median objective function decreases.
`t`	The initial step size for `theta` backtracking line search (default value is 1).
`a`	The factor by which step size is decreased in `theta` backtracking line search (default value is 0.5).
`thresh`	If the mean/median objective function does not decrease by at least this factor, we terminate the optimization early. Default is 1e-4.
`verbose`	If `TRUE`, prints information to console as model is being fit. Default is `FALSE`.

fwelnet always mean centers the columns of the x matrix. If standardize=TRUE, fwelnet will also scale the columns to have standard deviation 1. In all cases, the beta coefficients returned are for the original x values (i.e. uncentered and unscaled).

An object of class "fwelnet".

`beta`	A `p x length(lambda)` matrix of coefficients.
`theta`	Theta value, a `nfeaturevars x 1` matrix.
`a0`	Intercept sequence of length `length(lambda)`.
`lambda`	The actual sequence of `lambda` values used.
`nzero`	The number of non-zero coefficients for each value of `lambda`.
`family`	Response type.
`call`	The call that produced this object.
`obj`	A `max_iter + 1` by `length(lambda)` matrix of objective function values. (Number of rows could be fewer if the optimization stopped early.)

set.seed(1)
n <- 100; p <- 20
x <- matrix(rnorm(n * p), n, p)
beta <- matrix(c(rep(2, 5), rep(0, 15)), ncol = 1)
y <- x %*% beta + rnorm(n)
z <- cbind(1, abs(beta) + rnorm(p))

fwelnet(x, y, z)
fwelnet(x, y, z, ave_mode = 2)
fwelnet(x, y, z, ave_mode = 2, thresh_mode = 2)