neat/multiobjective/objectives

Objective-vector construction helpers for the NEAT multi-objective chapter.

The root multiobjective/ chapter explains the full Pareto-ranking flow. This narrower helper file explains how that flow gets its raw numbers: read one value per objective, preserve schema order, and assemble population-wide vectors and matrices that the dominance and crowding helpers can consume.

Read this chapter when you want to answer questions such as:

• How does the controller turn objective descriptors into numeric vectors?
• Why does value extraction stay separate from dominance and front building?
• What happens if one objective accessor throws during a long run?
• Why is preserving descriptor order so important for the rest of the multi-objective pipeline?

The mental model is a small three-step data-preparation flow:

1. read one safe value per objective descriptor,
2. assemble those values into one vector per genome,
3. assemble genome vectors into a population-wide matrix.

This boundary stays intentionally mechanical. It does not decide dominance, front ordering, or crowding; it only guarantees that later ranking stages receive a stable numeric view of the configured objective schema.

flowchart TD
  Genome[Genome] --> Read[Read one value per objective descriptor]
  Read --> Vector[Assemble per-genome objective vector]
  Vector --> Matrix[Assemble population-wide value matrix]
  Matrix --> Ranking[Dominance and front-building helpers consume matrix]

neat/multiobjective/objectives/multiobjective.objectives.ts

buildGenomeValues

buildGenomeValues(
  genomeItem: default,
  descriptors: ObjectiveDescriptor[],
): number[]

Builds an objective vector for a single genome.

The resulting array order matches the descriptors order exactly. Each component is read via {@link readObjectiveValue} so individual objective accessors are fault-tolerant.

Preserving descriptor order is a hard requirement here. Every later helper in the dominance and crowding pipeline assumes that the same objective occupies the same index in every vector.

Parameters:

• genomeItem - - Genome to evaluate.
• descriptors - - Objective descriptors (vector schema).

Returns: Objective value vector (length equals descriptors.length).

buildValuesMatrix

buildValuesMatrix(
  population: default[],
  descriptors: ObjectiveDescriptor[],
): number[][]

Builds a population-wide objective value matrix.

The resulting matrix is indexed as [genomeIndex][objectiveIndex] where genomeIndex matches the input population order.

This matrix is the handoff format for the rest of the multi-objective stack: rows preserve population order, columns preserve objective-schema order.

Parameters:

• population - - Genomes to evaluate (population order is preserved).
• descriptors - - Objective descriptors (column schema).

Returns: Objective values matrix.

readObjectiveValue

readObjectiveValue(
  genomeItem: default,
  descriptor: ObjectiveDescriptor,
): number

Safely reads a single objective value for a given genome.

This wraps the descriptor accessor in a try/catch so that a buggy objective function cannot crash multi-objective ranking.

Notes:

• If the accessor throws, this returns 0 (a neutral-ish fallback).
• Callers should prefer to surface accessor errors during development; this helper is intentionally defensive for long-running training loops.

Parameters:

• genomeItem - - Genome to evaluate.
• descriptor - - Objective descriptor providing an accessor.

Returns: Numeric objective value; 0 if the accessor throws.

Example:

const score = readObjectiveValue(genome, { accessor: (g) => g.score ?? 0 });