Struct ModelEnsemble 

pub struct ModelEnsemble {
    pub config: EnsembleConfig,
    pub members: Vec<ModelMember>,
}

Multi-model ensemble aggregator supporting voting, averaging, and stacking.

Fields

config: EnsembleConfig

members: Vec<ModelMember>

Implementations

impl ModelEnsemble

pub fn new(config: EnsembleConfig) -> Self

Create a new ensemble with the given configuration.

pub fn add_member(&mut self, model_id: String, weight: f64) -> &mut Self

Add a model member. Returns &mut Self for builder-style chaining.

pub fn enable_member(&mut self, model_id: &str) -> bool

Enable the member with the given id. Returns false if not found.

pub fn disable_member(&mut self, model_id: &str) -> bool

Disable the member with the given id. Returns false if not found.

pub fn record_call(&mut self, model_id: &str, success: bool)

Record a call result for a member (updates call_count / error_count).

pub fn member_stats(&self) -> Vec<&ModelMember>

Immutable references to all members.

pub fn stats(&self) -> EnsembleStats

Aggregate statistics over all members.

pub fn aggregate( &self, predictions: &[ModelPrediction], ) -> Result<EnsembleResult, EnsembleError>

Aggregate a slice of model predictions according to the configured strategy.

Steps:

1. Filter out predictions whose model_id maps to a disabled member.
2. Optionally check that every enabled member contributed (require_all).
3. Validate the count against min_models.
4. Apply the strategy.

pub fn top_class(outputs: &[f64]) -> usize

Return the index of the maximum element (argmax). Ties: lowest index.

pub fn softmax(logits: &[f64]) -> Vec<f64>

Numerically stable softmax (subtract max before exp).

pub fn normalize_weights(weights: &[f64]) -> Vec<f64>

Divide each weight by the total sum. If the sum is approximately 0, return a uniform distribution.

Trait Implementations

impl Debug for ModelEnsemble

fn fmt(&self, f: &mut Formatter<'_>) -> Result

Formats the value using the given formatter.