Skip to content
KortexyaKortexyaReasoning Layer SDK

Interface: LtnTrainRequest

@kortexya/reasoninglayer

@kortexya/reasoninglayer / LtnTrainRequest

Interface: LtnTrainRequest

Section titled “Interface: LtnTrainRequest”

Defined in: src/types/ltn.ts:371

Request to train per-rule certainties (and, optionally, a neural predicate).

Example

Section titled “Example”
const request: LtnTrainRequest = {
  rules: [{ individual: 'rex', sort: 'dog', membership: 0.9, certainty: 0.5 }],
  epochs: 100,
  learningRate: 0.05,
};

Properties

Section titled “Properties”

aggregator?

Section titled “aggregator?”

optional aggregator?: LtnAggregator | null

Defined in: src/types/ltn.ts:375

KB-level aggregator. Absent ⇒ pMeanError(p = 2).

epochs?

Section titled “epochs?”

optional epochs?: number | null

Defined in: src/types/ltn.ts:380

Training epochs for the certainty optimiser (capped server-side).

Default Value

Section titled “Default Value”
100

learningRate?

Section titled “learningRate?”

optional learningRate?: number | null

Defined in: src/types/ltn.ts:385

SGD learning rate for the certainty update.

Default Value

Section titled “Default Value”
0.05

predicate?

Section titled “predicate?”

optional predicate?: LtnPredicateTraining | null

Defined in: src/types/ltn.ts:387

Optional neural-predicate fit, persisted per tenant.

rules

Section titled “rules”

rules: LtnRule[]

Defined in: src/types/ltn.ts:373

The rules whose certainties are trained to maximise SatAgg.