Crate ferrolearn_preprocess 

ferrolearn-preprocess

Data preprocessing transformers for the ferrolearn machine learning framework.

This crate provides standard scalers, encoders, imputers, and feature selection utilities that follow the ferrolearn Fit/Transform trait pattern.

Scalers

All scalers are generic over F: Float + Send + Sync + 'static and implement Fit<Array2<F>, ()> (returning a Fitted* type) and FitTransform<Array2<F>>. The fitted types implement Transform<Array2<F>>.

• StandardScaler — zero-mean, unit-variance scaling
• MinMaxScaler — scale features to a given range (default [0, 1])
• RobustScaler — median / IQR-based scaling, robust to outliers
• MaxAbsScaler — scale by maximum absolute value so values are in [-1, 1]
• normalizer::Normalizer — normalize each sample (row) to unit norm
• power_transformer::PowerTransformer — Yeo-Johnson power transform

Encoders

• OneHotEncoder — encode Array2<usize> categorical columns as binary columns
• LabelEncoder — map Array1<String> labels to integer indices
• ordinal_encoder::OrdinalEncoder — map string categories to integers in order of first appearance

Imputers

• imputer::SimpleImputer — fill missing (NaN) values per feature column using Mean, Median, MostFrequent, or Constant strategy.

Feature Selection

• feature_selection::VarianceThreshold — remove features with variance below a configurable threshold.
• feature_selection::SelectKBest — keep the K features with the highest ANOVA F-scores against class labels.
• feature_selection::SelectFromModel — keep features whose importance weight (from a pre-fitted model) meets a configurable threshold.

Feature Engineering

• polynomial_features::PolynomialFeatures — generate polynomial and interaction features
• binarizer::Binarizer — threshold features to binary values
• function_transformer::FunctionTransformer — apply a user-provided function element-wise

Pipeline Integration

StandardScaler<f64>, MinMaxScaler<f64>, RobustScaler<f64>, MaxAbsScaler<f64>, Normalizer<f64>, PowerTransformer<f64>, PolynomialFeatures<f64>, SimpleImputer<f64>, VarianceThreshold<f64>, SelectKBest<f64>, and SelectFromModel<f64> each implement PipelineTransformer so they can be used as steps inside a Pipeline.

Examples

use ferrolearn_preprocess::StandardScaler;
use ferrolearn_core::traits::FitTransform;
use ndarray::array;

let x = array![[1.0_f64, 10.0], [2.0, 20.0], [3.0, 30.0]];
let scaled = StandardScaler::<f64>::new().fit_transform(&x).unwrap();
// scaled columns now have mean ≈ 0 and std ≈ 1

Re-exports

pub use binarizer::Binarizer;

pub use column_transformer::ColumnSelector;

pub use column_transformer::ColumnTransformer;

pub use column_transformer::FittedColumnTransformer;

pub use column_transformer::Remainder;

pub use column_transformer::make_column_transformer;

pub use feature_selection::FittedSelectKBest;

pub use feature_selection::FittedVarianceThreshold;

pub use feature_selection::ScoreFunc;

pub use feature_selection::SelectFromModel;

pub use feature_selection::SelectKBest;

pub use feature_selection::VarianceThreshold;

pub use function_transformer::FunctionTransformer;

pub use imputer::FittedSimpleImputer;

pub use imputer::ImputeStrategy;

pub use imputer::SimpleImputer;

pub use label_encoder::FittedLabelEncoder;

pub use label_encoder::LabelEncoder;

pub use max_abs_scaler::FittedMaxAbsScaler;

pub use max_abs_scaler::MaxAbsScaler;

pub use min_max_scaler::FittedMinMaxScaler;

pub use min_max_scaler::MinMaxScaler;

pub use normalizer::Normalizer;

pub use one_hot_encoder::FittedOneHotEncoder;

pub use one_hot_encoder::OneHotEncoder;

pub use ordinal_encoder::FittedOrdinalEncoder;

pub use ordinal_encoder::OrdinalEncoder;

pub use polynomial_features::PolynomialFeatures;

pub use power_transformer::FittedPowerTransformer;

pub use power_transformer::PowerTransformer;

pub use robust_scaler::FittedRobustScaler;

pub use robust_scaler::RobustScaler;

pub use standard_scaler::FittedStandardScaler;

pub use standard_scaler::StandardScaler;

pub use binary_encoder::BinaryEncoder;

pub use binary_encoder::FittedBinaryEncoder;

pub use iterative_imputer::FittedIterativeImputer;

pub use iterative_imputer::InitialStrategy;

pub use iterative_imputer::IterativeImputer;

pub use kbins_discretizer::BinEncoding;

pub use kbins_discretizer::BinStrategy;

pub use kbins_discretizer::FittedKBinsDiscretizer;

pub use kbins_discretizer::KBinsDiscretizer;

pub use knn_imputer::FittedKNNImputer;

pub use knn_imputer::KNNImputer;

pub use knn_imputer::KNNWeights;

pub use quantile_transformer::FittedQuantileTransformer;

pub use quantile_transformer::OutputDistribution;

pub use quantile_transformer::QuantileTransformer;

pub use rfe::RFE;

pub use rfe::RFECV;

pub use select_percentile::FittedSelectPercentile;

pub use select_percentile::SelectPercentile;

pub use spline_transformer::FittedSplineTransformer;

pub use spline_transformer::KnotStrategy;

pub use spline_transformer::SplineTransformer;

pub use target_encoder::FittedTargetEncoder;

pub use target_encoder::TargetEncoder;

Modules

binarizer

Binarizer: threshold features to binary values.

binary_encoder

Binary encoder: encode categorical integers as binary digits.

column_transformer

Column transformer: apply different transformers to different column subsets.

feature_selection

Feature selection transformers.

function_transformer

Function transformer: apply a user-provided function element-wise.

imputer

Simple imputer: fill missing (NaN) values per feature column.

iterative_imputer

Iterative imputer: fill missing values by modeling each feature as a function of all other features.

kbins_discretizer

K-bins discretizer: bin continuous features into discrete intervals.

knn_imputer

KNN imputer: fill missing (NaN) values using K-nearest neighbors.

label_encoder

Label encoder: maps string labels to integer indices.

max_abs_scaler

Max-absolute scaler: scale each feature by its maximum absolute value.

min_max_scaler

Min-max scaler: scales each feature to a given range.

normalizer

Normalizer: scale each sample (row) to unit norm.

one_hot_encoder

One-hot encoder for categorical integer features.

ordinal_encoder

Ordinal encoder: map string categories to integer indices.

polynomial_features

Polynomial features: generate polynomial and interaction features.

power_transformer

Power transformer: apply a power transform to make data more Gaussian.

quantile_transformer

Quantile transformer: map features to a uniform or normal distribution.

rfe

Recursive Feature Elimination (RFE) and RFE with Cross-Validation (RFECV).

robust_scaler

Robust scaler: median and IQR-based scaling.

select_percentile

Select features by percentile of highest scores.

spline_transformer

Spline transformer: generate B-spline basis functions for each feature.

standard_scaler

Standard scaler: zero-mean, unit-variance scaling.

target_encoder

Target encoder: encode categorical features using target statistics.