neat/adaptive/acceptance

Acceptance-gating heuristics for adaptive NEAT.

This category focuses on the minimal-criterion threshold that decides when a genome is good enough to stay in play, making it easier to study selection pressure separately from topology growth or mutation schedules.

The acceptance controller answers a narrower question than the rest of the adaptive subtree: not how much structure to allow or how strongly to mutate, but how selective the current generation should be before selection starts rewarding it.

Read this chapter when you want to understand:

• why adaptive acceptance rewrites the current generation instead of only adjusting future controller policy,
• how threshold initialization, acceptance measurement, threshold tuning, and rejection fit into one same-generation flow,
• where the public controller-facing entrypoint stops and the smaller minimal-criterion bookkeeping helpers begin.

The reading order is easiest to retain as one feedback loop:

1. seed the threshold if the run has not used acceptance pressure before,
2. snapshot the current generation's scores,
3. compare observed acceptance against the configured target band,
4. update the threshold and reject genomes that still fall short.

flowchart TD
  Scores[Current generation scores] --> Init[Initialize threshold if needed]
  Init --> Measure[Measure acceptance at current threshold]
  Measure --> Tune[Adjust threshold toward target acceptance]
  Tune --> Reject[Zero scores below final threshold]
  Reject --> Selection[Same-generation selection sees filtered scores]

neat/adaptive/acceptance/adaptive.acceptance.ts

applyMinimalCriterionAdaptive

applyMinimalCriterionAdaptive(): void

Apply adaptive minimal criterion (MC) acceptance.

This method maintains an MC threshold used to decide whether an individual genome is considered acceptable. It adapts the threshold based on the proportion of the population that meets the current threshold, trying to converge to a target acceptance rate.

Use this controller when you want the population to earn the right to stay in play. Unlike the complexity and lineage controllers, this one writes back to the current population immediately by zeroing scores below the accepted bar, so it directly changes the selection landscape for the same generation.

Behavior summary:

• Initializes _mcThreshold from configuration if undefined.
• Computes the proportion of genomes with score >= threshold.
• Adjusts threshold multiplicatively by adjustRate to move the observed proportion towards targetAcceptance.
• Sets g.score = 0 for genomes that fall below the final threshold — effectively rejecting them from selection.

Returns: Updates _mcThreshold over time and may zero out scores for currently rejected genomes.

Example:

// Example config snippet used by the engine // options.minimalCriterionAdaptive = { enabled: true, initialThreshold: 0.1, targetAcceptance: 0.5, adjustRate: 0.1 } engine.applyMinimalCriterionAdaptive();

applyRejection

applyRejection(
  engine: NeatLikeWithAdaptive,
  threshold: number,
): void

Zero scores below the final threshold.

Rejection is the acceptance chapter's most direct intervention. Instead of queuing a future policy change, it rewrites the current generation's scores so the same selection pass immediately treats low-performing genomes as filtered out.

Example:

const scores = collectScores(engine);
const acceptance = computeAcceptance(scores, engine._mcThreshold ?? 0);
updateThreshold(engine, acceptance, resolveTargetSettings(config));
applyRejection(engine, engine._mcThreshold ?? 0);

Parameters:

• engine - - NEAT engine instance.
• threshold - - Final MC threshold.

Returns: Nothing.

collectScores

collectScores(
  engine: NeatLikeWithAdaptive,
): number[]

Collect population scores into a snapshot array.

The acceptance controller works from one stable view of the current generation rather than mixing threshold updates with in-place rejection while it is still counting. Missing scores are treated as zero so unevaluated or explicitly rejected genomes remain part of the acceptance picture.

Parameters:

• engine - - NEAT engine instance.

Returns: Array of scores (missing scores treated as 0).

computeAcceptance

computeAcceptance(
  scores: number[],
  threshold: number,
): number

Compute acceptance metrics for the current threshold.

Acceptance is deliberately reduced to one proportion: how much of the current generation still clears the bar. That single number is enough for the caller to decide whether the threshold is too lenient, too strict, or already close enough to the configured target band.

Parameters:

• scores - - Population score snapshot.
• threshold - - Current MC threshold.

Returns: Acceptance proportion.

initializeThreshold

initializeThreshold(
  engine: NeatLikeWithAdaptive,
  config: { enabled?: boolean | undefined; initialThreshold?: number | undefined; targetAcceptance?: number | undefined; adjustRate?: number | undefined; },
): void

Initialize MC threshold if missing.

Threshold initialization is lazy because many runs never enable adaptive acceptance at all. The first invocation seeds the long-lived threshold, and later generations reuse the updated value rather than restarting from the original configuration each time.

Parameters:

• engine - - NEAT engine instance.
• config - - Minimal-criterion adaptive configuration.

Returns: Nothing.

resolveTargetSettings

resolveTargetSettings(
  config: { enabled?: boolean | undefined; initialThreshold?: number | undefined; targetAcceptance?: number | undefined; adjustRate?: number | undefined; },
): { targetAcceptance: number; adjustRate: number; }

Resolve target acceptance and adjust rate settings.

This helper centralizes the defaulting rules so later threshold-updating code can focus on policy instead of configuration fallback noise.

Parameters:

• config - - Minimal-criterion adaptive configuration.

Returns: Target settings.

updateThreshold

updateThreshold(
  engine: NeatLikeWithAdaptive,
  acceptance: number,
  tuning: { targetAcceptance: number; adjustRate: number; },
): void

Update the MC threshold based on acceptance proportion.

The threshold only moves when observed acceptance drifts outside the target band. Staying inside the band is treated as success, so the current threshold persists and the controller avoids oscillating every generation.

Parameters:

• engine - - NEAT engine instance.
• acceptance - - Observed acceptance proportion.
• tuning - - Target acceptance and adjustment settings.

Returns: Nothing.

neat/adaptive/acceptance/adaptive.minimal-criterion.utils.ts

Minimal-criterion helpers for adaptive acceptance.

This file owns the small evidence-to-threshold loop behind adaptive acceptance. It stays separate from the root adaptive controller entrypoint so the generated chapter can explain acceptance pressure as a readable pipeline instead of burying the mechanics inside one long method.

The helper flow is intentionally compact:

1. ensure there is a starting threshold,
2. snapshot current scores,
3. measure how much of the population clears the bar,
4. retune the threshold and reject genomes that still miss it.

applyRejection

applyRejection(
  engine: NeatLikeWithAdaptive,
  threshold: number,
): void

Zero scores below the final threshold.

Rejection is the acceptance chapter's most direct intervention. Instead of queuing a future policy change, it rewrites the current generation's scores so the same selection pass immediately treats low-performing genomes as filtered out.

Example:

const scores = collectScores(engine);
const acceptance = computeAcceptance(scores, engine._mcThreshold ?? 0);
updateThreshold(engine, acceptance, resolveTargetSettings(config));
applyRejection(engine, engine._mcThreshold ?? 0);

Parameters:

• engine - - NEAT engine instance.
• threshold - - Final MC threshold.

Returns: Nothing.

collectScores

collectScores(
  engine: NeatLikeWithAdaptive,
): number[]

Collect population scores into a snapshot array.

The acceptance controller works from one stable view of the current generation rather than mixing threshold updates with in-place rejection while it is still counting. Missing scores are treated as zero so unevaluated or explicitly rejected genomes remain part of the acceptance picture.

Parameters:

• engine - - NEAT engine instance.

Returns: Array of scores (missing scores treated as 0).

computeAcceptance

computeAcceptance(
  scores: number[],
  threshold: number,
): number

Compute acceptance metrics for the current threshold.

Acceptance is deliberately reduced to one proportion: how much of the current generation still clears the bar. That single number is enough for the caller to decide whether the threshold is too lenient, too strict, or already close enough to the configured target band.

Parameters:

• scores - - Population score snapshot.
• threshold - - Current MC threshold.

Returns: Acceptance proportion.

initializeThreshold

initializeThreshold(
  engine: NeatLikeWithAdaptive,
  config: { enabled?: boolean | undefined; initialThreshold?: number | undefined; targetAcceptance?: number | undefined; adjustRate?: number | undefined; },
): void

Initialize MC threshold if missing.

Threshold initialization is lazy because many runs never enable adaptive acceptance at all. The first invocation seeds the long-lived threshold, and later generations reuse the updated value rather than restarting from the original configuration each time.

Parameters:

• engine - - NEAT engine instance.
• config - - Minimal-criterion adaptive configuration.

Returns: Nothing.

resolveTargetSettings

resolveTargetSettings(
  config: { enabled?: boolean | undefined; initialThreshold?: number | undefined; targetAcceptance?: number | undefined; adjustRate?: number | undefined; },
): { targetAcceptance: number; adjustRate: number; }

Resolve target acceptance and adjust rate settings.

This helper centralizes the defaulting rules so later threshold-updating code can focus on policy instead of configuration fallback noise.

Parameters:

• config - - Minimal-criterion adaptive configuration.

Returns: Target settings.

updateThreshold

updateThreshold(
  engine: NeatLikeWithAdaptive,
  acceptance: number,
  tuning: { targetAcceptance: number; adjustRate: number; },
): void

Update the MC threshold based on acceptance proportion.

The threshold only moves when observed acceptance drifts outside the target band. Staying inside the band is treated as success, so the current threshold persists and the controller avoids oscillating every generation.

Parameters:

• engine - - NEAT engine instance.
• acceptance - - Observed acceptance proportion.
• tuning - - Target acceptance and adjustment settings.

Returns: Nothing.