Struct OrderedTargetEncoder 

pub struct OrderedTargetEncoder { /* private fields */ }

Ordered Target Encoder with M-Estimate Smoothing

Encodes categorical features by replacing them with smoothed target means. Uses ordered/streaming approach to prevent target leakage.

Smoothing formula (M-Estimate): encoded = (n × μ_category + m × μ_global) / (n + m)

Where:

• n = count of category
• μ_category = mean target for category
• m = smoothing parameter
• μ_global = global target mean

Implementations

impl OrderedTargetEncoder

pub fn new(smoothing: f64) -> Self

Create a new ordered target encoder

Arguments

• smoothing - M-estimate smoothing parameter (typically 1.0 to 10.0)

pub fn reset(&mut self)

Reset the encoder state

pub fn encode_and_update(&mut self, category: &str, target: f64) -> f64

Encode a single value using current statistics, then update

This is the ordered/streaming approach:

1. Compute encoding using statistics from prior rows only
2. Update statistics with this row’s values

Returns the encoded value.

pub fn encode_column( &mut self, categories: &[String], targets: &[f64], ) -> Vec<f64>

Encode an entire column in streaming order

Arguments

• categories - Category values for each row
• targets - Target values for each row

Returns encoded values with same length as input.

pub fn encode_inference(&self, category: &str) -> f64

Encode using final statistics (for inference)

After training, use this to encode new data using the learned statistics.

pub fn get_encoding_map(&self) -> EncodingMap

Get the learned encoding map for serialization