Skip to contents

Parsnip model specs: LS-SVR with RMSPE loss (Model 3)

psvr_rmspe_specs.Rd

Create parsnip model specifications for psvr_rmspe() with a fixed kernel type. cost maps to the regularization parameter \(\Gamma\).

Usage

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

psvr_rmspe_poly(
  mode = "regression",
  engine = "psvr",
  cost = NULL,
  degree = NULL,
  scale_factor = NULL,
  sym_type = NULL
)

psvr_rmspe_linear(
  mode = "regression",
  engine = "psvr",
  cost = NULL,
  sym_type = NULL
)

Arguments

mode

Only "regression" is supported.

engine

Only "psvr" is available.

cost

Regularization parameter \(\Gamma > 0\). Use hardhat::tune() to optimize. Mapped to cost_psvr(), whose default range [-2, 10] on the log2 scale (\(\Gamma \le 1024\)) is the \(\epsilon\)-SVR range and is too narrow for LS-SVR. \(\Gamma\) enters the LS-SVR system only through the \(y_k^2/\Gamma\) diagonal, so the value that balances that term against the kernel scales with the outcome's variance and with the sample size — the quantity cost_psvr_ls_data() computes. The LS-SVR optimum is therefore routinely orders of magnitude above the static ceiling, and a grid over the default is boundary-trapped whenever it is. Pass cost_psvr_ls_data() built from the training outcome explicitly, via update() on the extracted parameter set or via psvr_option_add_cost_ls() for a workflow set. This cannot be automated: tune finalizes parameters from the molded predictors only and never passes the outcome to dials::finalize(), so no finalize function on cost could compute it.

rbf_sigma

RBF bandwidth \(\sigma > 0\). Use hardhat::tune() to optimize. Mapped to rbf_sigma_psvr(), whose default range [-3, 1] on the log10 scale is a fixed, conservative fallback. The range does not finalize automatically from the training data. rbf_sigma_psvr() sets finalize = NULL, so dials::finalize() leaves it untouched. To centre the range on the data, pass rbf_sigma_psvr_data() computed on the preprocessed predictors explicitly — via update() on the extracted parameter set, or via psvr_option_add() for a workflow set. (RBF specs only.)

sym_type

Symmetry type: "none" (default) fits the non-symmetric LS-SVR of Model 3; "even" (a = 1) and "odd" (a = -1) fit the symmetric LS-SVR of Model 4. Use hardhat::tune() to optimise over the levels during CV; see sym_type_param() to restrict which levels are searched.

degree

Polynomial degree \(\ge 1\). 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.

Engine arguments

The precondition argument of psvr_rmspe() is exposed as a non-tunable engine argument. Pass it via parsnip::set_engine(), e.g. set_engine("psvr", precondition = "always"). Default is "auto". See psvr_rmspe() for accepted values and semantics.

Examples

library(parsnip)
spec <- psvr_rmspe_rbf(cost = 1000, rbf_sigma = 1) |>
  set_engine("psvr")

spec_poly <- psvr_rmspe_poly(cost = 1000, degree = 2, scale_factor = 1) |>
  set_engine("psvr")

spec_lin <- psvr_rmspe_linear(cost = 1000) |>
  set_engine("psvr")

# Symmetric LS-SVR (Model 4) via the sym_type argument:
spec_sym <- psvr_rmspe_rbf(cost = 1000, rbf_sigma = 1,
                           sym_type = "even") |>
  set_engine("psvr")