arazzo: 1.0.1
info:
  title: Amazon Forecast Train Predictor
  summary: Create a predictor, poll the listing until it is ACTIVE, and tag it.
  description: >-
    Trains an Amazon Forecast predictor (an ML model) on a dataset group and
    waits for training to complete. The workflow creates the predictor, polls
    the predictor listing until the matching predictor reports an ACTIVE status,
    and then applies metadata tags. Because the Forecast description only exposes
    a describe operation for datasets, predictor status is read from
    listPredictors. 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
sourceDescriptions:
- name: amazonForecastApi
  url: ../openapi/amazon-forecast-openapi.yml
  type: openapi
workflows:
- workflowId: train-predictor
  summary: Create an Amazon Forecast predictor and wait until training is ACTIVE.
  description: >-
    Creates a predictor, polls listPredictors until the matching predictor
    status is ACTIVE, and then associates the supplied tags with the predictor.
  inputs:
    type: object
    required:
    - predictorName
    - forecastHorizon
    - inputDataConfig
    - featurizationConfig
    properties:
      predictorName:
        type: string
        description: A name for the predictor.
      forecastHorizon:
        type: integer
        description: The number of time-steps to predict.
      algorithmArn:
        type: string
        description: The ARN of the algorithm to use for training.
      performAutoML:
        type: boolean
        description: Whether to let Forecast select the optimal algorithm.
      inputDataConfig:
        type: object
        description: The dataset group and supplementary features for training.
      featurizationConfig:
        type: object
        description: The featurization configuration for the predictor.
      tags:
        type: array
        description: The metadata tags to apply once the predictor is ACTIVE.
        items:
          type: object
  steps:
  - stepId: createPredictor
    description: >-
      Create the predictor and start training using the supplied configuration.
    operationId: createPredictor
    requestBody:
      contentType: application/json
      payload:
        PredictorName: $inputs.predictorName
        ForecastHorizon: $inputs.forecastHorizon
        AlgorithmArn: $inputs.algorithmArn
        PerformAutoML: $inputs.performAutoML
        InputDataConfig: $inputs.inputDataConfig
        FeaturizationConfig: $inputs.featurizationConfig
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      predictorArn: $response.body#/PredictorArn
  - stepId: waitForActive
    description: >-
      List predictors and confirm the newly created predictor has reached the
      ACTIVE status. Repeat this step while training is still in progress.
    operationId: listPredictors
    successCriteria:
    - condition: $statusCode == 200
    - context: $response.body
      condition: $.Predictors[?(@.PredictorArn == "$steps.createPredictor.outputs.predictorArn")].Status contains "ACTIVE"
      type: jsonpath
    onSuccess:
    - name: predictorActive
      type: goto
      stepId: tagPredictor
      criteria:
      - context: $response.body
        condition: $.Predictors[?(@.PredictorArn == "$steps.createPredictor.outputs.predictorArn")].Status contains "ACTIVE"
        type: jsonpath
    onFailure:
    - name: stillTraining
      type: retry
      retryAfter: 60
      retryLimit: 30
      criteria:
      - condition: $statusCode == 200
    outputs:
      predictors: $response.body#/Predictors
  - stepId: tagPredictor
    description: >-
      Apply the supplied metadata tags to the now-ACTIVE predictor.
    operationId: tagResource
    parameters:
    - name: resourceArn
      in: path
      value: $steps.createPredictor.outputs.predictorArn
    requestBody:
      contentType: application/json
      payload:
        Tags: $inputs.tags
    successCriteria:
    - condition: $statusCode == 200
    outputs:
      taggedArn: $steps.createPredictor.outputs.predictorArn
  outputs:
    predictorArn: $steps.createPredictor.outputs.predictorArn
    predictors: $steps.waitForActive.outputs.predictors