Fit epsilon-SVR with MAPE loss (deprecated)

Soft-deprecated wrapper retained for backwards compatibility. New code should use psvr(loss = "mape", ...). Returns the legacy psvr_mape object shape; the legacy predict.psvr_mape() / print.psvr_mape() / coef.psvr_mape() methods continue to dispatch correctly.

Usage

mape_svr(X, y, kernel, C, eps, solver = c("smo", "osqp"), tol = 1e-05)

Arguments

X, y: Training matrix and strictly-positive target vector.
kernel: Kernel closure from make_kernel().
C: Regularization parameter C > 0.
eps: Insensitivity tube half-width (% units), eps >= 0.
solver: Backend: "smo" (default) or "osqp".
tol: Solver zero-threshold.

Value

A list of class "psvr_mape".

Examples

if (FALSE) { # \dontrun{
# Use psvr() instead:
fit <- psvr(X, y, loss = "mape", kernel = make_kernel("rbf"),
            C = 10, eps = 5)
} # }