Struct WeightLearner

pub struct WeightLearner {
    pub weights: ScoringWeights,
    pub learning_rate: f32,
    pub n_updates: usize,
}

Online weight learning from outcome feedback.

Simple gradient update: if the retrieval led to success, decrease the score (make it more likely to be retrieved again). If failure, increase.

Uses a learning rate to control update speed.

Fields

weights: ScoringWeights

Current weights.

learning_rate: f32

Learning rate for gradient updates.

n_updates: usize

Number of updates applied.

Implementations

impl WeightLearner

pub fn new(weights: ScoringWeights, learning_rate: f32) -> Self

Create a new learner with initial weights.

pub fn update(&mut self, candidates: &[CandidateFeatures], outcome: f32)

Update weights based on outcome feedback.

candidates: the features of candidates that were retrieved. outcome: 1.0 for success, 0.0 for failure.

Adjusts weights to make factors that correlated with success stronger, and factors that correlated with failure weaker.