Per-token normalised time, honouring an already-normalised column

Builds the [0, 1] time axis the modelling functions fit on. By default (time_normalised = "auto") the column is inspected with time_already_normalised(): a column that is already proportional is used as-is (clamped to [0, 1]), anything else is min-max rescaled to [0, 1] within each token exactly as before.

Usage

resolve_time_norm(
  data,
  time,
  token,
  time_normalised = c("auto", "no", "yes"),
  quiet = FALSE
)

Arguments

data

A data frame containing the time and token columns.

time

Name of the time column.

token

Name of the token-ID column.

time_normalised

One of "auto" (default; detect and use an already-normalised column as-is), "no" (always rescale per token), or "yes" (declare the column already normalised to [0, 1]; values outside that interval are an error).

quiet

Suppress the message emitted when auto-detection decides the column is already normalised. Default FALSE.

Value

A list with time_norm (numeric vector, one value per row of data) and prenormalised (logical: was the column used as-is?).

Details

In the rescaling path, a token whose time has zero range (a single sample, or all-identical times) gets 0.5 for every row, and NA times inside an otherwise valid token propagate as NA — matching the behaviour the modelling functions have always had.

See also

time_already_normalised() for the detection rule.