Class PipelineAutoTrainer

Production IModelAutoTrainer: reuse-or-train, publish, repoint, archive.

Implements

IModelAutoTrainer

Index

Constructors

Methods

archiveSuperseded describeRefusal EnsureTrainedModel findReusableModel loadBindings repoint trainAndPublish

Constructors

constructor

new PipelineAutoTrainer(
options?: PipelineAutoTrainerOptions,
): PipelineAutoTrainer
Parameters
- options: PipelineAutoTrainerOptions = {}
Returns PipelineAutoTrainer
- Defined in packages/AI/PredictiveStudio/Engine/src/scoring/auto-train.ts:72

Methods

`Protected`archiveSuperseded

archiveSuperseded(
    oldModelId: string,
    newModelId: string,
    contextUser: UserInfo,
    provider: IMetadataProvider,
): Promise<void>
Archive the superseded model when it was Published (best-effort — never fails the run).
Parameters
- oldModelId: string
- newModelId: string
- contextUser: UserInfo
- provider: IMetadataProvider
Returns Promise<void>
- Defined in packages/AI/PredictiveStudio/Engine/src/scoring/auto-train.ts:227

`Protected`describeRefusal

describeRefusal(outcome: PromoteModelOutcome): string
Plain description of a non-promoted outcome.
Parameters
- outcome: PromoteModelOutcome
Returns string
- Defined in packages/AI/PredictiveStudio/Engine/src/scoring/auto-train.ts:138

EnsureTrainedModel

EnsureTrainedModel(
    request: AutoTrainRequest,
    contextUser: UserInfo,
    provider: IMetadataProvider,
): Promise<string>
Ensure a published, artifact-backed model exists for the pipeline and that the Record Process + its bindings point at it. Returns that model's id.
Parameters
- request: AutoTrainRequest
- contextUser: UserInfo
- provider: IMetadataProvider
Returns Promise<string>
Throws
when training or publishing fails (e.g. leakage-flagged model needing sign-off)
Implementation of IModelAutoTrainer.EnsureTrainedModel
- Defined in packages/AI/PredictiveStudio/Engine/src/scoring/auto-train.ts:79

`Protected`findReusableModel

findReusableModel(
    pipelineId: string,
    contextUser: UserInfo,
    provider: IMetadataProvider,
): Promise<string>
The newest Published model of the pipeline whose artifact loads on this server, if any.
Parameters
- pipelineId: string
- contextUser: UserInfo
- provider: IMetadataProvider
Returns Promise<string>
- Defined in packages/AI/PredictiveStudio/Engine/src/scoring/auto-train.ts:97

`Protected`loadBindings

loadBindings(
    request: AutoTrainRequest,
    recordProcessId: string,
    contextUser: UserInfo,
    provider: IMetadataProvider,
): Promise<MJMLModelScoringBindingEntity[]>
Bindings on the Record Process and/or on the superseded model.
Parameters
- request: AutoTrainRequest
- recordProcessId: string
- contextUser: UserInfo
- provider: IMetadataProvider
Returns Promise<MJMLModelScoringBindingEntity[]>
- Defined in packages/AI/PredictiveStudio/Engine/src/scoring/auto-train.ts:207

`Protected`repoint

repoint(
    request: AutoTrainRequest,
    modelId: string,
    contextUser: UserInfo,
    provider: IMetadataProvider,
): Promise<void>
Point the Record Process (its Configuration.modelId) and its scoring bindings at the trained model. Bindings on the process — and bindings still on the superseded model — are repointed; when none exist and a binding was declared, one is created.
Parameters
- request: AutoTrainRequest
- modelId: string
- contextUser: UserInfo
- provider: IMetadataProvider
Returns Promise<void>
- Defined in packages/AI/PredictiveStudio/Engine/src/scoring/auto-train.ts:163

`Protected`trainAndPublish

trainAndPublish(
    pipelineId: string,
    contextUser: UserInfo,
    provider: IMetadataProvider,
): Promise<string>
Train the pipeline (real TrainingEngine path) and publish the new model through the gate.
Parameters
- pipelineId: string
- contextUser: UserInfo
- provider: IMetadataProvider
Returns Promise<string>
- Defined in packages/AI/PredictiveStudio/Engine/src/scoring/auto-train.ts:123