Interface: LtnPredicateTraining
@kortexya/reasoninglayer / LtnPredicateTraining
Interface: LtnPredicateTraining
Section titled “Interface: LtnPredicateTraining”Defined in: src/types/ltn.ts:342
Optional neural-predicate training block on a train request.
Remarks
Section titled “Remarks”Fits a small MLP p_predicate(x) ∈ [0, 1] on the supplied examples and persists
its weights per tenant (when an LTN_WEIGHTS_DIR is configured server-side).
Example
Section titled “Example”const predicate: LtnPredicateTraining = { predicate: 'is_loyal', examples: [{ features: [0.9, 0.1], label: 1.0 }], epochs: 200, batchSize: 16,};Properties
Section titled “Properties”batchSize?
Section titled “batchSize?”
optionalbatchSize?:number|null
Defined in: src/types/ltn.ts:351
Mini-batch size.
Default Value
Section titled “Default Value”16epochs?
Section titled “epochs?”
optionalepochs?:number|null
Defined in: src/types/ltn.ts:356
Training epochs over the examples (capped server-side).
Default Value
Section titled “Default Value”200examples
Section titled “examples”examples:
LtnPredicateExample[]
Defined in: src/types/ltn.ts:346
Labelled examples (explicit input vectors).
predicate
Section titled “predicate”predicate:
string
Defined in: src/types/ltn.ts:344
Predicate name — keys the per-tenant neural predicate and its checkpoint file.