Execution and evaluation

Cross-layer

On the existing Stage-based path, the Initializer creates the initial population and the OptimizationStrategy advances the work for each generation. The Algorithm generates candidates, the SurrogateManager and AcquisitionFunction predict and score them, and the Evaluator performs the expensive true evaluation. The results update the Algorithm’s population and Archive, and generation processing repeats until Termination meets its stopping condition.

Pipeline flow

The diagram shows the data flow from initialization through the termination check. A configuration without a Surrogate can omit prediction and scoring.

        flowchart TD
    INIT["Initializer<br/>Create initial population"] --> STEP
    subgraph STEP["OptimizationStrategy.step()<br/>Process one generation"]
        direction TB
        ASK["Algorithm.ask()<br/>Generate candidates"] --> PREDICT["SurrogateManager.predict()<br/>Predict"]
        PREDICT --> SCORE["AcquisitionFunction<br/>Score candidates"]
        SCORE --> SEL["Select candidates for true evaluation"]
        SEL --> EVAL["Evaluator → Problem<br/>Expensive evaluation"]
        EVAL --> TELL["Algorithm.tell()<br/>Update population"]
    end
    STEP --> TERM{"Termination?"}
    TERM -- "No" --> STEP
    TERM -- "Yes" --> RESULT([Result])
    EVAL -- "Evaluated points" --> ARC[("Archive")]
    ARC -. "Training data" .-> PREDICT
    

Each Stage emits typed events through the CallbackManager. This lets you observe progress or swap behavior at runtime without subclassing a Stage.

Component roles

Component	Role
Problem	Defines the objective function, design variables, search bounds, constraints, and optimization direction.
Initializer	Creates the initial population and Archive before the loop starts.
Algorithm	Generates candidates with ask() and applies evaluation results to the population with tell().
OptimizationStrategy	Builds one generation’s processing and decides which candidates receive true evaluation.
SurrogateManager	Coordinates Surrogate training and prediction.
Surrogate	Predicts from training data; it does not define scoring.
AcquisitionFunction	Converts predictions into scores for candidate selection.
Evaluator	Evaluates the objective function sequentially or in parallel.
Population	Holds evaluated individuals and population data.
Termination	Determines when generation processing ends.
CallbackManager	Observes and records events from the entire pipeline.

Initialization (Initializer)

Choose how to create the initial population and Archive.

Initializer

Evaluator

Run sequential and parallel objective evaluations.

Evaluator

OptimizationStrategy

Assemble candidate generation, prediction, evaluation planning, and state updates.

OptimizationStrategy

Termination

Combine stopping conditions such as evaluation and generation limits.

Termination