Data-driven cost range for LS-SVR psvr models

Returns a quant_param whose search range scales with var(y) * N, the standard heuristic for the LS-SVR regularisation parameter Γ (Suykens et al. 2002, Least Squares Support Vector Machines, §3.1.3). On the log2 scale, the lower bound is -2 (i.e. Γ ≥ 0.25) and the upper bound is log2(var(y) * N) + width_log2.

Usage

cost_psvr_ls_data(y, n = length(y), width_log2 = 4)

Arguments

y

Numeric vector of strictly positive training targets.

n

Sample size. Default length(y).

width_log2

Scalar giving the half-width (in log2 units) added above log2(var(y) * n) to set the upper bound. Default 4 (≈16× headroom). Negative values are accepted with a warning, since the resulting upper bound falls below the var(y) * n heuristic and is unlikely to be useful.

Value

A quant_param dials object.

Details

With the default width_log2 = 4, the upper bound covers the typical optimum within ~2 orders of magnitude on benchmark datasets. For Boston Housing (var(medv) ≈ 84.6, N_train = 404 under an 80/20 split), this gives an upper bound of 2^19.06 ≈ 5.4 × 10⁵, two decades above the published optimum Γ ≈ 1.7 × 10⁴ — comfortable headroom for Bayesian optimisation without boundary trapping. The static cost_psvr() range [-2, 10] (i.e. Γ ≤ 1024) underestimates this by more than a decade.

Use this function for m3 (LS-SVR) and m4 (symmetric LS-SVR) workflows. Stick to cost_psvr() for m1/m2 (ε-SVR), where cost maps to C and typical optima lie in [10, 100].

See also

cost_psvr(), psvr_option_add_cost_ls()

Examples

cost_psvr_ls_data(c(10, 20, 30, 40, 50))
#> Cost (quantitative)
#> Transformer: log-2 [1e-100, Inf]
#> Range (transformed scale): [-2, 14.3]