Module preprocessing 

Streaming preprocessing utilities for feature transformation.

These transformers process features incrementally, maintaining running statistics that update with each sample – no batch recomputation needed.

Modules

Module	Purpose
normalizer	Welford-based online standardization (zero-mean, unit-variance)
feature_selector	EWMA importance tracking with dynamic feature masking
ccipca	Candid Covariance-free Incremental PCA – streaming dimensionality reduction
feature_hasher	Feature hashing (hashing trick) for fixed-size dimensionality reduction
min_max	Streaming min-max scaler for feature normalization to a target range
one_hot	Streaming one-hot encoder with online category discovery
target_encoder	Streaming target encoder with Bayesian smoothing
polynomial	Polynomial and interaction feature generation

Example

use irithyll::preprocessing::{IncrementalNormalizer, OnlineFeatureSelector};

let mut norm = IncrementalNormalizer::new();
let standardized = norm.update_and_transform(&[100.0, 0.5, -3.0]);

let mut selector = OnlineFeatureSelector::new(3, 0.5, 0.1, 10);
selector.update_importances(&[0.9, 0.1, 0.8]);
let masked = selector.mask_features(&standardized);

Re-exports

pub use ccipca::CCIPCA;

pub use feature_hasher::FeatureHasher;

pub use feature_selector::OnlineFeatureSelector;

pub use min_max::MinMaxScaler;

pub use normalizer::IncrementalNormalizer;

pub use one_hot::OneHotEncoder;

pub use polynomial::PolynomialFeatures;

pub use target_encoder::TargetEncoder;

pub use crate::pipeline::StreamingPreprocessor;

Modules

ccipca

Candid Covariance-free Incremental PCA (CCIPCA) for streaming dimensionality reduction.

feature_hasher

Feature hashing (hashing trick) for dimensionality reduction.

feature_selector

EWMA-based online feature selector with dynamic importance masking.

min_max

Streaming min-max scaler for feature normalization.

normalizer

Welford online mean/variance normalizer for incremental standardization.

one_hot

Streaming one-hot encoder for categorical features.

polynomial

Degree-2 polynomial feature generation for interaction modeling.

target_encoder

Streaming target encoder for categorical features with Bayesian smoothing.