Skip to contents

Parsnip model specs: symmetric epsilon-SVR with MAPE loss (Model 2)

psvr_mape_sym_specs.Rd

Create parsnip model specifications for mape_sym_svr() with a fixed kernel type. The symmetry type is exposed as the tunable sym_type argument ("even" for a = 1, "odd" for a = -1); pass sym_type = tune() to let CV select it automatically.

Usage

psvr_mape_sym_rbf(
  mode = "regression",
  engine = "psvr",
  cost = NULL,
  svm_margin = NULL,
  rbf_sigma = NULL,
  sym_type = NULL
)

psvr_mape_sym_poly(
  mode = "regression",
  engine = "psvr",
  cost = NULL,
  svm_margin = NULL,
  degree = NULL,
  scale_factor = NULL
)

psvr_mape_sym_linear(
  mode = "regression",
  engine = "psvr",
  cost = NULL,
  svm_margin = NULL
)

Arguments

mode

Only "regression" is supported.

engine

Only "psvr" is available.

cost

Regularization parameter C > 0. Use hardhat::tune() to optimize. Mapped to cost_psvr() with range [-2, 10] on the log2 scale — wider than dials::cost() to cover the larger values needed by LS-SVR models.

svm_margin

Epsilon tube half-width ε ≥ 0 expressed as a percentage of each target value. Use hardhat::tune() to optimize. Mapped to margin_percentage() with default range [1, 20] (percentage units).

rbf_sigma

RBF bandwidth σ > 0. Use hardhat::tune() to optimize. Mapped to rbf_sigma_psvr(); the search range auto-finalizes using the median-distance heuristic when training data are available. (RBF specs only.)

sym_type

Symmetry type: "even" (default, a = 1) or "odd" (a = -1). Use hardhat::tune() to optimise over both values during CV.

degree

Polynomial degree ≥ 1. Use hardhat::tune() to optimize. (Polynomial specs only.)

scale_factor

Polynomial constant term (coef₀). Use hardhat::tune() to optimize. (Polynomial specs only.)

Value

A parsnip model_spec object of the corresponding class.

Examples

if (FALSE) { # \dontrun{
library(parsnip)
spec <- psvr_mape_sym_rbf(cost = 10, svm_margin = 1, rbf_sigma = 1) |>
  set_engine("psvr")

spec_poly <- psvr_mape_sym_poly(cost = 10, svm_margin = 1, degree = 2,
                                scale_factor = 1) |>
  set_engine("psvr")

spec_lin <- psvr_mape_sym_linear(cost = 10, svm_margin = 1) |>
  set_engine("psvr")
} # }