Amazon Neptune ML Data Processing to Model Training

arazzo: 1.0.1
info:
  title: Amazon Neptune ML Data Processing to Model Training
  summary: Run a Neptune ML data processing job to completion, then launch and poll model training.
  description: >-
    Chains the first two stages of the Neptune ML pipeline. The workflow creates
    a data processing job that exports and prepares graph data, polls that job
    until its status is Completed, then launches a model training job that
    consumes the processed output and polls it until training completes. Both poll
    loops use a retry delay to handle the in-progress states. Every step spells
    out its request inline so the flow can be read and executed without opening
    the underlying OpenAPI description.
  version: 1.0.0
  x-realizes-capability-ids:
  - BC-610.60
  x-capability-derivation:
    method: 'deterministic join: sourceDescriptions -> per-tag OpenAPI -> tag/capability edge. No classification at this step.'
    min_confidence: 0.7
    sources:
    - capability_id: BC-610.60
      capability_name: Artificial Intelligence Management
      spec: amazon-neptune-model-training-api-openapi.yml
      confidence: 0.75
    model: Turbo EA Capabilities by Vincent Verdet — Turbo EA, https://github.com/vincentmakes/turbo-ea-capabilities, CC BY 4.0
sourceDescriptions:
- name: dataProcessingApi
  url: ../openapi/amazon-neptune-data-processing-api-openapi.yml
  type: openapi
- name: modelTrainingApi
  url: ../openapi/amazon-neptune-model-training-api-openapi.yml
  type: openapi
workflows:
- workflowId: ml-dataprocessing-to-training
  summary: Process graph data, then train a model on the processed output.
  description: >-
    Creates and polls a data processing job, then creates and polls a model
    training job that consumes the processed data.
  inputs:
    type: object
    required:
    - inputDataS3Location
    - processedDataS3Location
    - trainModelS3Location
    properties:
      inputDataS3Location:
        type: string
        description: S3 URI for the input graph data.
      processedDataS3Location:
        type: string
        description: S3 URI where processed output is written.
      trainModelS3Location:
        type: string
        description: S3 location for trained model artifacts.
      modelName:
        type: string
        description: The model type to train (rgcn, transe, distmult, rotate, custom).
  steps:
  - stepId: startProcessing
    description: >-
      Create a data processing job that exports and prepares graph data into
      the processed-data S3 location.
    operationId: createDataProcessingJob
    requestBody:
      contentType: application/json
      payload:
        inputDataS3Location: $inputs.inputDataS3Location
        processedDataS3Location: $inputs.processedDataS3Location
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      processingJobId: $response.body#/id
  - stepId: pollProcessing
    description: >-
      Poll the data processing job. Retry while it is still in progress and
      continue once its status is Completed.
    operationId: getDataProcessingJobStatus
    parameters:
    - name: id
      in: path
      value: $steps.startProcessing.outputs.processingJobId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      processingStatus: $response.body#/status
    onSuccess:
    - name: processingRunning
      type: retry
      retryAfter: 60
      retryLimit: 120
      criteria:
      - context: $response.body
        condition: $.status != "Completed"
        type: jsonpath
    - name: processingDone
      type: goto
      stepId: startTraining
      criteria:
      - context: $response.body
        condition: $.status == "Completed"
        type: jsonpath
  - stepId: startTraining
    description: >-
      Launch a model training job that consumes the completed data processing
      job's output.
    operationId: createModelTrainingJob
    requestBody:
      contentType: application/json
      payload:
        dataProcessingJobId: $steps.startProcessing.outputs.processingJobId
        trainModelS3Location: $inputs.trainModelS3Location
        modelName: $inputs.modelName
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      trainingJobId: $response.body#/id
  - stepId: pollTraining
    description: >-
      Poll the model training job. Retry while it is still in progress and
      finish once its status is Completed.
    operationId: getModelTrainingJobStatus
    parameters:
    - name: id
      in: path
      value: $steps.startTraining.outputs.trainingJobId
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      trainingStatus: $response.body#/status
    onSuccess:
    - name: trainingRunning
      type: retry
      retryAfter: 60
      retryLimit: 120
      criteria:
      - context: $response.body
        condition: $.status != "Completed"
        type: jsonpath
    - name: trainingDone
      type: end
      criteria:
      - context: $response.body
        condition: $.status == "Completed"
        type: jsonpath
  outputs:
    processingJobId: $steps.startProcessing.outputs.processingJobId
    trainingJobId: $steps.startTraining.outputs.trainingJobId
    trainingStatus: $steps.pollTraining.outputs.trainingStatus