Fit wrappers for parsnip engine dispatch

Bridge functions called by parsnip when fitting psvr model specs. Exported only because parsnip's resolver requires it; not intended for direct use. Call psvr_mape() or psvr_rmspe() instead for direct fitting.

Usage

psvr_mape_rbf_fit(
  x,
  y,
  C,
  eps,
  rbf_sigma = 1,
  sym_type = "none",
  tol = 0.001,
  max_iter = 100000L
)

psvr_mape_poly_fit(
  x,
  y,
  C,
  eps,
  degree = 3L,
  scale_factor = 1,
  sym_type = "none",
  tol = 0.001,
  max_iter = 100000L
)

psvr_mape_linear_fit(
  x,
  y,
  C,
  eps,
  sym_type = "none",
  tol = 0.001,
  max_iter = 100000L
)

psvr_rmspe_rbf_fit(
  x,
  y,
  gamma,
  rbf_sigma = 1,
  sym_type = "none",
  precondition = "auto"
)

psvr_rmspe_poly_fit(
  x,
  y,
  gamma,
  degree = 3L,
  scale_factor = 1,
  sym_type = "none",
  precondition = "auto"
)

psvr_rmspe_linear_fit(x, y, gamma, sym_type = "none", precondition = "auto")

Arguments

x

Numeric predictor matrix (parsnip matrix interface).

y

Numeric outcome vector (strictly positive).

C

Regularization parameter for MAPE models.

eps

Epsilon tube half-width for MAPE models.

rbf_sigma

RBF bandwidth \(\sigma > 0\).

sym_type

Symmetry type. "none" (the default) dispatches to the non-symmetric fitter; "even" and "odd" dispatch to the symmetric fitter with a = 1L and a = -1L respectively.

tol

Solver convergence tolerance for the SMO loop. Default 1e-3.

max_iter

Maximum SMO iterations. Default 100000L. The solver emits a warning() and returns converged = FALSE if it does not converge within max_iter.

degree

Polynomial degree \(\ge 1\).

scale_factor

Polynomial constant term (\(\mathrm{coef}_0\)).

gamma

Regularization parameter for RMSPE models.

precondition

Optional symmetric rescaling preconditioner for the RMSPE LS-SVR fitters. See psvr_rmspe() for accepted values and semantics.

Value

A fitted model object of the S3 class matching the wrapper's model family, returned unmodified from the internal fitter. These are the same classes psvr_mape() and psvr_rmspe() return, so a parsnip fit unwrapped with parsnip::extract_fit_engine() and a directly fitted object are interchangeable. Which class is returned depends on sym_type, since each wrapper dispatches to the symmetric or non-symmetric fitter.

The MAPE wrappers (psvr_mape_rbf_fit(), psvr_mape_poly_fit(), psvr_mape_linear_fit()) with sym_type = "none" return an object of class "psvr_mape": a list with beta (support-vector dual differences), alpha and alpha_star (length-N pre-pruning duals, retained for warm starts), b, X_sv, y_sv, y_train, fitted_values, kernel, C, eps, n_train, p_train, iterations, converged, and block_k4. With sym_type = "even" or "odd" they return class "psvr_mape_sym": the same components plus a (the symmetry type) and spectral (the estimated extreme eigenvalues of the symmetrized kernel matrix, and the diagonal shift applied to it if it was not numerically positive semi-definite).

The RMSPE wrappers (psvr_rmspe_rbf_fit(), psvr_rmspe_poly_fit(), psvr_rmspe_linear_fit()) with sym_type = "none" return class "psvr_rmspe": a list with alpha, b, X_train, y_train, fitted_values, kernel, gamma, n_train, p_train, and precondition_applied. With sym_type = "even" or "odd" they return class "psvr_rmspe_sym": the same components plus a.