Fit normalization parameters on the stable interior of a distribution

Winsorizes x, isolates the stable (non-sentinel, non-NA) interior, selects a transformation type, fits its parameters and centering/scaling constants, and returns a normalize_x_fit object that can be reused via apply_normalization().

Separating fitting from scoring allows the same fitted model to be applied to new data (e.g. a holdout year) without re-estimating parameters, and ensures that sentinel observations do not distort the center and spread estimates used for standardization.

Usage

find_best_normalization(
  x,
  range,
  vtype = NULL,
  winsorize = 0.98,
  offset = 0.001,
  standardize = TRUE,
  robust = TRUE,
  zero_tol = 1e-08,
  one_tol = 1e-08,
  boundary_mass_cutoff = 0.1,
  hurdle_trans = c("rank", "logit"),
  hurdle_range = c(1, 100),
  verbose = TRUE
)

Arguments

x

Numeric vector to fit on.

range

Character range code ("np", "zp", "zo", "nz", or "lo;hi"). Required.

vtype

Optional transformation type override. One of NULL (auto-detect), "asinh", "logit", "rank_normal", "hurdle".

winsorize

Winsorization proportion. Default 0.98.

offset

Sentinel offset for fixed bounds. Default 0.001.

standardize

Logical. If TRUE, fit centering and scaling constants. Default TRUE.

robust

Logical. If TRUE, use median/MAD; otherwise mean/SD. Default TRUE.

zero_tol

Tolerance for zero detection. Default 1e-8.

one_tol

Tolerance for near-one detection. Default 1e-8.

boundary_mass_cutoff

Hurdle/rank_normal trigger threshold. Default 0.10.

hurdle_trans

Positive-part transformation for hurdle variables. "rank" (default) or "logit".

hurdle_range

Rescaled output range for hurdle positive component. Default c(1, 100).

verbose

Logical. If TRUE, print a diagnostic summary. Default TRUE.

Value

An object of class normalize_x_fit — a named list containing transformation type, parameters, centering/scaling constants, and diagnostics. Pass this object to apply_normalization().

See also

apply_normalization(), normalize_x()

Examples

x <- c( rnorm(200), rep(0, 30) )

# fit on training data
fit <- find_best_normalization( x, range = "np", verbose = FALSE )
fit$transform_type
#> [1] "rank_normal"
fit$center
#> [1] -0.03357722
fit$scale
#> [1] 1.000616