Skip to main content

min_max_scale

dataset_ml::preprocessing

Function min_max_scale 

Source 
pub fn min_max_scale(
    features: &Array2<f64>,
) -> Result<(Array2<f64>, Scaler), DatasetError>
Expand description

Rescale each feature column into the [0, 1] range.

This is min-max scaling, (value - min) / (max - min), applied per column. Prefer it over standardize when a bounded range matters more than a comparable spread. Use it for pixel-like features (digits), or as input to a model that expects [0, 1].

As with standardize, the minimum and maximum come from the finite values of each column. Non-finite entries stay untouched. A constant column maps to all zeros rather than dividing by 0.

§Parameters

  • features - The numeric feature matrix, shape (n_samples, n_features).

§Returns

  • (Array2<f64>, Scaler) - The rescaled matrix, and the fitted per-column statistics to replay on later data with apply_scaler.

§Errors

  • DatasetError::ValidationError - Returns this when features has no rows or no columns.

§Example

use dataset_ml::preprocessing::min_max_scale;
use ndarray::array;

let features = array![[1.0, -5.0], [3.0, 5.0]];
let (scaled, _scaler) = min_max_scale(&features).unwrap();

assert_eq!(scaled, array![[0.0, 0.0], [1.0, 1.0]]);