Enum NormalizationTransform 

pub enum NormalizationTransform<T: Value> {
    Domain {
        min: T,
        max: T,
    },
    Clip {
        min: T,
        max: T,
    },
    MeanSubtraction,
    ZScore,
    LogOffset {
        asymptote_epsilon: Option<T>,
    },
}

Available on crate feature transforms only.

Transformations around normalization or otherwise controlling range

Variants

Domain

Normalizes the dataset to a specified range.

Each element is linearly scaled to fit within [min, max]. Useful for mapping values to a standard range before further processing or ML workflows.

Technical Details

xₙ = (x - x_min) / (x_max - x_min) * (max - min) + min
where
  x_min and x_max are the minimum and maximum of the original dataset.

Parameters

• min: Minimum value of the target range.
• max: Maximum value of the target range.

Fields

min: T

Minimum value of the target range

max: T

Maximum value of the target range

Clip

Restricts all values in the dataset to a specified range.

Any element smaller than min is set to min, and any element larger than max is set to max. Useful for bounding outliers or enforcing hard limits.

Technical Details

Element-wise operation:

xₙ = min(max(x, min), max)
where
  min and max are the specified bounds.

Parameters

• min: Lower bound of the allowed range.
• max: Upper bound of the allowed range.

Fields

min: T

Lower bound of the allowed range

max: T

Upper bound of the allowed range

MeanSubtraction

Centers the dataset by subtracting its mean from every element.

After this transformation, the dataset has a mean of zero, but its variance and overall shape remain unchanged. Useful as a preprocessing step in statistics and machine learning.

Technical Details

xₙ = x - mean(x)

ZScore

Normalizes the dataset to zero mean and unit variance.

Each element is centered by subtracting the dataset mean, then scaled by dividing with the standard deviation. This is a common preprocessing step in statistics and machine learning to make features comparable.

Technical Details

Element-wise operation:

xₙ = (x - mean(x)) / std(x)
where
  mean(x) is the average of the dataset
  std(x) is the standard deviation of the dataset

LogOffset

Normalizes the dataset by applying a logarithmic transformation with an offset.

An asymptote - or in other words a value that the data approaches but never reaches - can cause problems for polynomial fitting. This transform tries to approximate that value, then applies a log transform to the data relative to that value

Technical Details

asymptote_epsilon = mean(Y) * 1e-02
asymptote = max(Y) + asymptote_epsilon
yₙ = ln(asymptote - y)

where
  Y is the dataset being transformed
  mean(Y) is the average of the dataset
  max(Y) is the maximum value in the dataset
  asymptote_epsilon is a small buffer added to ensure all points are below the asymptote, which is important for the log transformation

Fields

asymptote_epsilon: Option<T>

A small epsilon added to the estimated asymptote to ensure all points are below it, which is important for the log transformation. If None, a default value of 1% of the standard deviation of the data will be used.

Trait Implementations

impl<T: Value> Transform<T> for NormalizationTransform<T>

fn apply<'a>(&self, data: impl Iterator<Item = &'a mut T>)

Applies the transformation to the given data.

fn apply_to(&self, point: &mut T)

Applies the transformation to a single data point.