unscale

Unscale is the operation of reversing a scaling transformation applied to a quantity. In mathematics and computer science, scaling multiplies a value, coordinate, or dataset by a scale factor. Unscaling applies the inverse operation to restore the original size or units.

In data processing and machine learning, features are often scaled to improve numerical stability and convergence. Unscaling refers to applying the inverse transformation to model outputs or predictions to interpret them in the original units. For example, if data were standardized by subtracting a mean and dividing by a standard deviation, then to recover the original values you multiply by the standard deviation and add the mean. If a min–max normalization was used, unscaling is x = x' * (max - min) + min.

In graphics and signal processing, unscale can refer to reversing a prior resize or resample operation, returning an image or signal to its original resolution or scale. This may involve inverse resampling or interpolation. Inaccuracies may occur due to rounding, clipping, or information loss during initial scaling.

Practical considerations include the need to store the parameters used for scaling so they can be correctly inverted later. Unscaling should be applied consistently to inputs and outputs, especially in pipelines where predictions are interpreted or deployed. Some systems use de-normalization to map standardized values to readable units.

See also: scaling, normalization, inverse transformation, de-normalization.