共用方式為


Microsoft.MachineLearningServices 工作區/作業 2022-02-01-preview

Bicep 資源定義

工作區/作業資源類型可以使用目標作業進行部署:

如需每個 API 版本中已變更屬性的清單,請參閱 變更記錄檔。

資源格式

若要建立 Microsoft.MachineLearningServices/workspaces/jobs 資源,請將下列 Bicep 新增至範本。

resource symbolicname 'Microsoft.MachineLearningServices/workspaces/jobs@2022-02-01-preview' = {
  name: 'string'
  parent: resourceSymbolicName
  properties: {
    computeId: 'string'
    description: 'string'
    displayName: 'string'
    experimentName: 'string'
    identity: {
      identityType: 'string'
      // For remaining properties, see IdentityConfiguration objects
    }
    isArchived: bool
    properties: {
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
    }
    schedule: {
      endTime: 'string'
      scheduleStatus: 'string'
      startTime: 'string'
      timeZone: 'string'
      scheduleType: 'string'
      // For remaining properties, see ScheduleBase objects
    }
    services: {
      {customized property}: {
        endpoint: 'string'
        jobServiceType: 'string'
        port: int
        properties: {
          {customized property}: 'string'
        }
      }
    }
    tags: {
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
      {customized property}: 'string'
    }
    jobType: 'string'
    // For remaining properties, see JobBaseDetails objects
  }
}

JobBaseDetails 物件

設定 jobType 屬性以指定物件的類型。

針對 AutoML,請使用:

  jobType: 'AutoML'
  environmentId: 'string'
  environmentVariables: {
    {customized property}: 'string'
  }
  outputs: {
    {customized property}: {
      description: 'string'
      jobOutputType: 'string'
      // For remaining properties, see JobOutput objects
    }
  }
  resources: {
    instanceCount: int
    instanceType: 'string'
    properties: {
      {customized property}: any()
    }
  }
  taskDetails: {
    logVerbosity: 'string'
    taskType: 'string'
    // For remaining properties, see AutoMLVertical objects
  }

針對 [命令],請使用:

  jobType: 'Command'
  codeId: 'string'
  command: 'string'
  distribution: {
    distributionType: 'string'
    // For remaining properties, see DistributionConfiguration objects
  }
  environmentId: 'string'
  environmentVariables: {
    {customized property}: 'string'
  }
  inputs: {
    {customized property}: {
      description: 'string'
      jobInputType: 'string'
      // For remaining properties, see JobInput objects
    }
  }
  limits: {
    jobLimitsType: 'string'
    timeout: 'string'
  }
  outputs: {
    {customized property}: {
      description: 'string'
      jobOutputType: 'string'
      // For remaining properties, see JobOutput objects
    }
  }
  resources: {
    instanceCount: int
    instanceType: 'string'
    properties: {
      {customized property}: any()
    }
  }

針對 管線,請使用:

  jobType: 'Pipeline'
  inputs: {
    {customized property}: {
      description: 'string'
      jobInputType: 'string'
      // For remaining properties, see JobInput objects
    }
  }
  jobs: {
    {customized property}: any()
  }
  outputs: {
    {customized property}: {
      description: 'string'
      jobOutputType: 'string'
      // For remaining properties, see JobOutput objects
    }
  }
  settings: any()

針對 掃掠,請使用:

  jobType: 'Sweep'
  earlyTermination: {
    delayEvaluation: int
    evaluationInterval: int
    policyType: 'string'
    // For remaining properties, see EarlyTerminationPolicy objects
  }
  inputs: {
    {customized property}: {
      description: 'string'
      jobInputType: 'string'
      // For remaining properties, see JobInput objects
    }
  }
  limits: {
    jobLimitsType: 'string'
    maxConcurrentTrials: int
    maxTotalTrials: int
    timeout: 'string'
    trialTimeout: 'string'
  }
  objective: {
    goal: 'string'
    primaryMetric: 'string'
  }
  outputs: {
    {customized property}: {
      description: 'string'
      jobOutputType: 'string'
      // For remaining properties, see JobOutput objects
    }
  }
  samplingAlgorithm: {
    samplingAlgorithmType: 'string'
    // For remaining properties, see SamplingAlgorithm objects
  }
  searchSpace: any()
  trial: {
    codeId: 'string'
    command: 'string'
    distribution: {
      distributionType: 'string'
      // For remaining properties, see DistributionConfiguration objects
    }
    environmentId: 'string'
    environmentVariables: {
      {customized property}: 'string'
    }
    resources: {
      instanceCount: int
      instanceType: 'string'
      properties: {
        {customized property}: any()
      }
    }
  }

IdentityConfiguration 物件

設定 identityType 屬性以指定物件的類型。

針對 AMLToken,請使用:

  identityType: 'AMLToken'

針對 [受控],請使用:

  identityType: 'Managed'
  clientId: 'string'
  objectId: 'string'
  resourceId: 'string'

針對 UserIdentity,請使用:

  identityType: 'UserIdentity'

ScheduleBase 物件

設定 scheduleType 屬性以指定物件的類型。

針對 Cron,請使用:

  scheduleType: 'Cron'
  expression: 'string'

針對 [週期],請使用:

  scheduleType: 'Recurrence'
  frequency: 'string'
  interval: int
  pattern: {
    hours: [
      int
    ]
    minutes: [
      int
    ]
    weekdays: [
      'string'
    ]
  }

JobOutput 物件

設定 jobOutputType 屬性以指定對象的類型。

針對 CustomModel,請使用:

  jobOutputType: 'CustomModel'
  mode: 'string'
  uri: 'string'

針對 MLFlowModel,請使用:

  jobOutputType: 'MLFlowModel'
  mode: 'string'
  uri: 'string'

針對 MLTable,請使用:

  jobOutputType: 'MLTable'
  mode: 'string'
  uri: 'string'

針對 TritonModel,請使用:

  jobOutputType: 'TritonModel'
  mode: 'string'
  uri: 'string'

針對 UriFile,請使用:

  jobOutputType: 'UriFile'
  mode: 'string'
  uri: 'string'

針對 UriFolder,請使用:

  jobOutputType: 'UriFolder'
  mode: 'string'
  uri: 'string'

AutoMLVertical 物件

設定 taskType 屬性以指定物件的類型。

針對 [分類],請使用:

  taskType: 'Classification'
  allowedModels: [
    'string'
  ]
  blockedModels: [
    'string'
  ]
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      cvSplitColumnNames: [
        'string'
      ]
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      nCrossValidations: {
        mode: 'string'
        // For remaining properties, see NCrossValidations objects
      }
      validationDataSize: int
    }
    weightColumnName: 'string'
  }
  featurizationSettings: {
    blockedTransformers: [
      'string'
    ]
    columnNameAndTypes: {
      {customized property}: 'string'
    }
    datasetLanguage: 'string'
    dropColumns: [
      'string'
    ]
    enableDnnFeaturization: bool
    mode: 'string'
    transformerParams: {
      {customized property}: [
        {
          fields: [
            'string'
          ]
          parameters: any()
        }
      ]
    }
  }
  limitSettings: {
    enableEarlyTermination: bool
    exitScore: int
    maxConcurrentTrials: int
    maxCoresPerTrial: int
    maxTrials: int
    timeout: 'string'
    trialTimeout: 'string'
  }
  primaryMetric: 'string'
  trainingSettings: {
    enableDnnTraining: bool
    enableModelExplainability: bool
    enableOnnxCompatibleModels: bool
    enableStackEnsemble: bool
    enableVoteEnsemble: bool
    ensembleModelDownloadTimeout: 'string'
    stackEnsembleSettings: {
      stackMetaLearnerKWargs: any()
      stackMetaLearnerTrainPercentage: int
      stackMetaLearnerType: 'string'
    }
  }

針對 預測,請使用:

  taskType: 'Forecasting'
  allowedModels: [
    'string'
  ]
  blockedModels: [
    'string'
  ]
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      cvSplitColumnNames: [
        'string'
      ]
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      nCrossValidations: {
        mode: 'string'
        // For remaining properties, see NCrossValidations objects
      }
      validationDataSize: int
    }
    weightColumnName: 'string'
  }
  featurizationSettings: {
    blockedTransformers: [
      'string'
    ]
    columnNameAndTypes: {
      {customized property}: 'string'
    }
    datasetLanguage: 'string'
    dropColumns: [
      'string'
    ]
    enableDnnFeaturization: bool
    mode: 'string'
    transformerParams: {
      {customized property}: [
        {
          fields: [
            'string'
          ]
          parameters: any()
        }
      ]
    }
  }
  forecastingSettings: {
    countryOrRegionForHolidays: 'string'
    cvStepSize: int
    featureLags: 'string'
    forecastHorizon: {
      mode: 'string'
      // For remaining properties, see ForecastHorizon objects
    }
    frequency: 'string'
    seasonality: {
      mode: 'string'
      // For remaining properties, see Seasonality objects
    }
    shortSeriesHandlingConfig: 'string'
    targetAggregateFunction: 'string'
    targetLags: {
      mode: 'string'
      // For remaining properties, see TargetLags objects
    }
    targetRollingWindowSize: {
      mode: 'string'
      // For remaining properties, see TargetRollingWindowSize objects
    }
    timeColumnName: 'string'
    timeSeriesIdColumnNames: [
      'string'
    ]
    useStl: 'string'
  }
  limitSettings: {
    enableEarlyTermination: bool
    exitScore: int
    maxConcurrentTrials: int
    maxCoresPerTrial: int
    maxTrials: int
    timeout: 'string'
    trialTimeout: 'string'
  }
  primaryMetric: 'string'
  trainingSettings: {
    enableDnnTraining: bool
    enableModelExplainability: bool
    enableOnnxCompatibleModels: bool
    enableStackEnsemble: bool
    enableVoteEnsemble: bool
    ensembleModelDownloadTimeout: 'string'
    stackEnsembleSettings: {
      stackMetaLearnerKWargs: any()
      stackMetaLearnerTrainPercentage: int
      stackMetaLearnerType: 'string'
    }
  }

針對 ImageClassification,請使用:

  taskType: 'ImageClassification'
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      validationDataSize: int
    }
  }
  limitSettings: {
    maxConcurrentTrials: int
    maxTrials: int
    timeout: 'string'
  }
  modelSettings: {
    advancedSettings: 'string'
    amsGradient: bool
    augmentations: 'string'
    beta1: int
    beta2: int
    checkpointDatasetId: 'string'
    checkpointFilename: 'string'
    checkpointFrequency: int
    checkpointRunId: 'string'
    distributed: bool
    earlyStopping: bool
    earlyStoppingDelay: int
    earlyStoppingPatience: int
    enableOnnxNormalization: bool
    evaluationFrequency: int
    gradientAccumulationStep: int
    layersToFreeze: int
    learningRate: int
    learningRateScheduler: 'string'
    modelName: 'string'
    momentum: int
    nesterov: bool
    numberOfEpochs: int
    numberOfWorkers: int
    optimizer: 'string'
    randomSeed: int
    splitRatio: int
    stepLRGamma: int
    stepLRStepSize: int
    trainingBatchSize: int
    trainingCropSize: int
    validationBatchSize: int
    validationCropSize: int
    validationResizeSize: int
    warmupCosineLRCycles: int
    warmupCosineLRWarmupEpochs: int
    weightDecay: int
    weightedLoss: int
  }
  primaryMetric: 'string'
  searchSpace: [
    {
      amsGradient: 'string'
      augmentations: 'string'
      beta1: 'string'
      beta2: 'string'
      distributed: 'string'
      earlyStopping: 'string'
      earlyStoppingDelay: 'string'
      earlyStoppingPatience: 'string'
      enableOnnxNormalization: 'string'
      evaluationFrequency: 'string'
      gradientAccumulationStep: 'string'
      layersToFreeze: 'string'
      learningRate: 'string'
      learningRateScheduler: 'string'
      modelName: 'string'
      momentum: 'string'
      nesterov: 'string'
      numberOfEpochs: 'string'
      numberOfWorkers: 'string'
      optimizer: 'string'
      randomSeed: 'string'
      splitRatio: 'string'
      stepLRGamma: 'string'
      stepLRStepSize: 'string'
      trainingBatchSize: 'string'
      trainingCropSize: 'string'
      validationBatchSize: 'string'
      validationCropSize: 'string'
      validationResizeSize: 'string'
      warmupCosineLRCycles: 'string'
      warmupCosineLRWarmupEpochs: 'string'
      weightDecay: 'string'
      weightedLoss: 'string'
    }
  ]
  sweepSettings: {
    earlyTermination: {
      delayEvaluation: int
      evaluationInterval: int
      policyType: 'string'
      // For remaining properties, see EarlyTerminationPolicy objects
    }
    limits: {
      maxConcurrentTrials: int
      maxTrials: int
    }
    samplingAlgorithm: 'string'
  }

針對 ImageClassificationMultilabel,請使用:

  taskType: 'ImageClassificationMultilabel'
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      validationDataSize: int
    }
  }
  limitSettings: {
    maxConcurrentTrials: int
    maxTrials: int
    timeout: 'string'
  }
  modelSettings: {
    advancedSettings: 'string'
    amsGradient: bool
    augmentations: 'string'
    beta1: int
    beta2: int
    checkpointDatasetId: 'string'
    checkpointFilename: 'string'
    checkpointFrequency: int
    checkpointRunId: 'string'
    distributed: bool
    earlyStopping: bool
    earlyStoppingDelay: int
    earlyStoppingPatience: int
    enableOnnxNormalization: bool
    evaluationFrequency: int
    gradientAccumulationStep: int
    layersToFreeze: int
    learningRate: int
    learningRateScheduler: 'string'
    modelName: 'string'
    momentum: int
    nesterov: bool
    numberOfEpochs: int
    numberOfWorkers: int
    optimizer: 'string'
    randomSeed: int
    splitRatio: int
    stepLRGamma: int
    stepLRStepSize: int
    trainingBatchSize: int
    trainingCropSize: int
    validationBatchSize: int
    validationCropSize: int
    validationResizeSize: int
    warmupCosineLRCycles: int
    warmupCosineLRWarmupEpochs: int
    weightDecay: int
    weightedLoss: int
  }
  primaryMetric: 'string'
  searchSpace: [
    {
      amsGradient: 'string'
      augmentations: 'string'
      beta1: 'string'
      beta2: 'string'
      distributed: 'string'
      earlyStopping: 'string'
      earlyStoppingDelay: 'string'
      earlyStoppingPatience: 'string'
      enableOnnxNormalization: 'string'
      evaluationFrequency: 'string'
      gradientAccumulationStep: 'string'
      layersToFreeze: 'string'
      learningRate: 'string'
      learningRateScheduler: 'string'
      modelName: 'string'
      momentum: 'string'
      nesterov: 'string'
      numberOfEpochs: 'string'
      numberOfWorkers: 'string'
      optimizer: 'string'
      randomSeed: 'string'
      splitRatio: 'string'
      stepLRGamma: 'string'
      stepLRStepSize: 'string'
      trainingBatchSize: 'string'
      trainingCropSize: 'string'
      validationBatchSize: 'string'
      validationCropSize: 'string'
      validationResizeSize: 'string'
      warmupCosineLRCycles: 'string'
      warmupCosineLRWarmupEpochs: 'string'
      weightDecay: 'string'
      weightedLoss: 'string'
    }
  ]
  sweepSettings: {
    earlyTermination: {
      delayEvaluation: int
      evaluationInterval: int
      policyType: 'string'
      // For remaining properties, see EarlyTerminationPolicy objects
    }
    limits: {
      maxConcurrentTrials: int
      maxTrials: int
    }
    samplingAlgorithm: 'string'
  }

針對 ImageInstanceSegmentation,請使用:

  taskType: 'ImageInstanceSegmentation'
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      validationDataSize: int
    }
  }
  limitSettings: {
    maxConcurrentTrials: int
    maxTrials: int
    timeout: 'string'
  }
  modelSettings: {
    advancedSettings: 'string'
    amsGradient: bool
    augmentations: 'string'
    beta1: int
    beta2: int
    boxDetectionsPerImage: int
    boxScoreThreshold: int
    checkpointDatasetId: 'string'
    checkpointFilename: 'string'
    checkpointFrequency: int
    checkpointRunId: 'string'
    distributed: bool
    earlyStopping: bool
    earlyStoppingDelay: int
    earlyStoppingPatience: int
    enableOnnxNormalization: bool
    evaluationFrequency: int
    gradientAccumulationStep: int
    imageSize: int
    layersToFreeze: int
    learningRate: int
    learningRateScheduler: 'string'
    maxSize: int
    minSize: int
    modelName: 'string'
    modelSize: 'string'
    momentum: int
    multiScale: bool
    nesterov: bool
    nmsIouThreshold: int
    numberOfEpochs: int
    numberOfWorkers: int
    optimizer: 'string'
    randomSeed: int
    splitRatio: int
    stepLRGamma: int
    stepLRStepSize: int
    tileGridSize: 'string'
    tileOverlapRatio: int
    tilePredictionsNmsThreshold: int
    trainingBatchSize: int
    validationBatchSize: int
    validationIouThreshold: int
    validationMetricType: 'string'
    warmupCosineLRCycles: int
    warmupCosineLRWarmupEpochs: int
    weightDecay: int
  }
  primaryMetric: 'MeanAveragePrecision'
  searchSpace: [
    {
      amsGradient: 'string'
      augmentations: 'string'
      beta1: 'string'
      beta2: 'string'
      boxDetectionsPerImage: 'string'
      boxScoreThreshold: 'string'
      distributed: 'string'
      earlyStopping: 'string'
      earlyStoppingDelay: 'string'
      earlyStoppingPatience: 'string'
      enableOnnxNormalization: 'string'
      evaluationFrequency: 'string'
      gradientAccumulationStep: 'string'
      imageSize: 'string'
      layersToFreeze: 'string'
      learningRate: 'string'
      learningRateScheduler: 'string'
      maxSize: 'string'
      minSize: 'string'
      modelName: 'string'
      modelSize: 'string'
      momentum: 'string'
      multiScale: 'string'
      nesterov: 'string'
      nmsIouThreshold: 'string'
      numberOfEpochs: 'string'
      numberOfWorkers: 'string'
      optimizer: 'string'
      randomSeed: 'string'
      splitRatio: 'string'
      stepLRGamma: 'string'
      stepLRStepSize: 'string'
      tileGridSize: 'string'
      tileOverlapRatio: 'string'
      tilePredictionsNmsThreshold: 'string'
      trainingBatchSize: 'string'
      validationBatchSize: 'string'
      validationIouThreshold: 'string'
      validationMetricType: 'string'
      warmupCosineLRCycles: 'string'
      warmupCosineLRWarmupEpochs: 'string'
      weightDecay: 'string'
    }
  ]
  sweepSettings: {
    earlyTermination: {
      delayEvaluation: int
      evaluationInterval: int
      policyType: 'string'
      // For remaining properties, see EarlyTerminationPolicy objects
    }
    limits: {
      maxConcurrentTrials: int
      maxTrials: int
    }
    samplingAlgorithm: 'string'
  }

針對 ImageObjectDetection,請使用:

  taskType: 'ImageObjectDetection'
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      validationDataSize: int
    }
  }
  limitSettings: {
    maxConcurrentTrials: int
    maxTrials: int
    timeout: 'string'
  }
  modelSettings: {
    advancedSettings: 'string'
    amsGradient: bool
    augmentations: 'string'
    beta1: int
    beta2: int
    boxDetectionsPerImage: int
    boxScoreThreshold: int
    checkpointDatasetId: 'string'
    checkpointFilename: 'string'
    checkpointFrequency: int
    checkpointRunId: 'string'
    distributed: bool
    earlyStopping: bool
    earlyStoppingDelay: int
    earlyStoppingPatience: int
    enableOnnxNormalization: bool
    evaluationFrequency: int
    gradientAccumulationStep: int
    imageSize: int
    layersToFreeze: int
    learningRate: int
    learningRateScheduler: 'string'
    maxSize: int
    minSize: int
    modelName: 'string'
    modelSize: 'string'
    momentum: int
    multiScale: bool
    nesterov: bool
    nmsIouThreshold: int
    numberOfEpochs: int
    numberOfWorkers: int
    optimizer: 'string'
    randomSeed: int
    splitRatio: int
    stepLRGamma: int
    stepLRStepSize: int
    tileGridSize: 'string'
    tileOverlapRatio: int
    tilePredictionsNmsThreshold: int
    trainingBatchSize: int
    validationBatchSize: int
    validationIouThreshold: int
    validationMetricType: 'string'
    warmupCosineLRCycles: int
    warmupCosineLRWarmupEpochs: int
    weightDecay: int
  }
  primaryMetric: 'MeanAveragePrecision'
  searchSpace: [
    {
      amsGradient: 'string'
      augmentations: 'string'
      beta1: 'string'
      beta2: 'string'
      boxDetectionsPerImage: 'string'
      boxScoreThreshold: 'string'
      distributed: 'string'
      earlyStopping: 'string'
      earlyStoppingDelay: 'string'
      earlyStoppingPatience: 'string'
      enableOnnxNormalization: 'string'
      evaluationFrequency: 'string'
      gradientAccumulationStep: 'string'
      imageSize: 'string'
      layersToFreeze: 'string'
      learningRate: 'string'
      learningRateScheduler: 'string'
      maxSize: 'string'
      minSize: 'string'
      modelName: 'string'
      modelSize: 'string'
      momentum: 'string'
      multiScale: 'string'
      nesterov: 'string'
      nmsIouThreshold: 'string'
      numberOfEpochs: 'string'
      numberOfWorkers: 'string'
      optimizer: 'string'
      randomSeed: 'string'
      splitRatio: 'string'
      stepLRGamma: 'string'
      stepLRStepSize: 'string'
      tileGridSize: 'string'
      tileOverlapRatio: 'string'
      tilePredictionsNmsThreshold: 'string'
      trainingBatchSize: 'string'
      validationBatchSize: 'string'
      validationIouThreshold: 'string'
      validationMetricType: 'string'
      warmupCosineLRCycles: 'string'
      warmupCosineLRWarmupEpochs: 'string'
      weightDecay: 'string'
    }
  ]
  sweepSettings: {
    earlyTermination: {
      delayEvaluation: int
      evaluationInterval: int
      policyType: 'string'
      // For remaining properties, see EarlyTerminationPolicy objects
    }
    limits: {
      maxConcurrentTrials: int
      maxTrials: int
    }
    samplingAlgorithm: 'string'
  }

針對 回歸,請使用:

  taskType: 'Regression'
  allowedModels: [
    'string'
  ]
  blockedModels: [
    'string'
  ]
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      cvSplitColumnNames: [
        'string'
      ]
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      nCrossValidations: {
        mode: 'string'
        // For remaining properties, see NCrossValidations objects
      }
      validationDataSize: int
    }
    weightColumnName: 'string'
  }
  featurizationSettings: {
    blockedTransformers: [
      'string'
    ]
    columnNameAndTypes: {
      {customized property}: 'string'
    }
    datasetLanguage: 'string'
    dropColumns: [
      'string'
    ]
    enableDnnFeaturization: bool
    mode: 'string'
    transformerParams: {
      {customized property}: [
        {
          fields: [
            'string'
          ]
          parameters: any()
        }
      ]
    }
  }
  limitSettings: {
    enableEarlyTermination: bool
    exitScore: int
    maxConcurrentTrials: int
    maxCoresPerTrial: int
    maxTrials: int
    timeout: 'string'
    trialTimeout: 'string'
  }
  primaryMetric: 'string'
  trainingSettings: {
    enableDnnTraining: bool
    enableModelExplainability: bool
    enableOnnxCompatibleModels: bool
    enableStackEnsemble: bool
    enableVoteEnsemble: bool
    ensembleModelDownloadTimeout: 'string'
    stackEnsembleSettings: {
      stackMetaLearnerKWargs: any()
      stackMetaLearnerTrainPercentage: int
      stackMetaLearnerType: 'string'
    }
  }

針對 TextClassification,請使用:

  taskType: 'TextClassification'
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      validationDataSize: int
    }
  }
  featurizationSettings: {
    datasetLanguage: 'string'
  }
  limitSettings: {
    maxConcurrentTrials: int
    maxTrials: int
    timeout: 'string'
  }
  primaryMetric: 'string'

針對 TextClassificationMultilabel,請使用:

  taskType: 'TextClassificationMultilabel'
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      validationDataSize: int
    }
  }
  featurizationSettings: {
    datasetLanguage: 'string'
  }
  limitSettings: {
    maxConcurrentTrials: int
    maxTrials: int
    timeout: 'string'
  }

針對 TextNER,請使用:

  taskType: 'TextNER'
  dataSettings: {
    targetColumnName: 'string'
    testData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      testDataSize: int
    }
    trainingData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
    }
    validationData: {
      data: {
        description: 'string'
        jobInputType: 'string'
        mode: 'string'
        uri: 'string'
      }
      validationDataSize: int
    }
  }
  featurizationSettings: {
    datasetLanguage: 'string'
  }
  limitSettings: {
    maxConcurrentTrials: int
    maxTrials: int
    timeout: 'string'
  }

NCrossValidations 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode: 'Auto'

針對 [自定義],請使用:

  mode: 'Custom'
  value: int

ForecastHorizon 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode: 'Auto'

針對 [自定義],請使用:

  mode: 'Custom'
  value: int

季節性物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode: 'Auto'

針對 [自定義],請使用:

  mode: 'Custom'
  value: int

TargetLags 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode: 'Auto'

針對 [自定義],請使用:

  mode: 'Custom'
  values: [
    int
  ]

TargetRollingWindowSize 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode: 'Auto'

針對 [自定義],請使用:

  mode: 'Custom'
  value: int

EarlyTerminationPolicy 物件

設定 policyType 屬性以指定物件的類型。

針對 Bandit,請使用:

  policyType: 'Bandit'
  slackAmount: int
  slackFactor: int

針對 MedianStopping,請使用:

  policyType: 'MedianStopping'

針對 TruncationSelection,請使用:

  policyType: 'TruncationSelection'
  truncationPercentage: int

DistributionConfiguration 物件

設定 distributionType 屬性以指定物件的類型。

針對 Mpi,請使用:

  distributionType: 'Mpi'
  processCountPerInstance: int

針對 PyTorch,請使用:

  distributionType: 'PyTorch'
  processCountPerInstance: int

針對 TensorFlow,請使用:

  distributionType: 'TensorFlow'
  parameterServerCount: int
  workerCount: int

JobInput 物件

設定 jobInputType 屬性以指定物件的類型。

針對 CustomModel,請使用:

  jobInputType: 'CustomModel'
  mode: 'string'
  uri: 'string'

針對 常值,請使用:

  jobInputType: 'Literal'
  value: 'string'

針對 MLFlowModel,請使用:

  jobInputType: 'MLFlowModel'
  mode: 'string'
  uri: 'string'

針對 MLTable,請使用:

  jobInputType: 'MLTable'
  mode: 'string'
  uri: 'string'

針對 TritonModel,請使用:

  jobInputType: 'TritonModel'
  mode: 'string'
  uri: 'string'

針對 UriFile,請使用:

  jobInputType: 'UriFile'
  mode: 'string'
  uri: 'string'

針對 UriFolder,請使用:

  jobInputType: 'UriFolder'
  mode: 'string'
  uri: 'string'

SamplingAlgorithm 物件

設定 samplingAlgorithmType 屬性以指定物件的類型。

針對 貝氏,請使用:

  samplingAlgorithmType: 'Bayesian'

針對 Grid,請使用:

  samplingAlgorithmType: 'Grid'

針對 隨機,請使用:

  samplingAlgorithmType: 'Random'
  rule: 'string'
  seed: int

屬性值

workspaces/jobs

名稱 描述
NAME 資源名稱

瞭解如何在 Bicep 中設定子資源的名稱和類型。
字串 (必要)
父系 (parent) 在 Bicep 中,您可以指定子資源的父資源。 只有在父資源之外宣告子資源時,才需要新增這個屬性。

如需詳細資訊,請參閱 父資源外部的子資源
類型資源的符號名稱: 工作區
properties [必要]實體的其他屬性。 JobBaseDetails (必要)

JobBaseDetails

名稱 描述
computeId 計算資源的 ARM 資源識別碼。 字串
description 資產描述文字。 字串
displayName 工作的顯示名稱。 字串
experimentName 作業所屬的實驗名稱。 如果未設定,作業會放在「預設」實驗中。 字串
身分識別 身分識別設定。 如果設定,這應該是 AmlToken、ManagedIdentity、UserIdentity 或 null 的其中一個。
如果為 null,則預設為 AmlToken。
IdentityConfiguration
isArchived 資產是否已封存? bool
properties 資產屬性字典。 ResourceBaseProperties
schedule 作業的排程定義。
如果未提供排程,作業會在提交后立即執行一次。
ScheduleBase
服務 JobEndpoints 的清單。
針對本機作業,作業端點會有 FileStreamObject 的端點值。
JobBaseServices
tags 標記字典。 標記可以新增、移除和更新。 物件 (object)
jobType 設定物件類型 AutoML
命令
管線
需要掃掠 ()

IdentityConfiguration

名稱 描述
identityType 設定物件類型 AMLToken
受控
必要) 的 UserIdentity (

AmlToken

名稱 描述
identityType [必要]指定身分識別架構的類型。 必要的 『AMLToken』 ()

ManagedIdentity

名稱 描述
identityType [必要]指定身分識別架構的類型。 「受控」 (必要)
clientId 依用戶端識別元指定使用者指派的身分識別。 針對系統指派,請勿設定此欄位。 字串

約束:
最小長度 = 36
最大長度 = 36
模式 = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$
objectId 依物件識別元指定使用者指派的身分識別。 針對系統指派,請勿設定此欄位。 字串

約束:
最小長度 = 36
最大長度 = 36
模式 = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$
resourceId 依 ARM 資源識別碼指定使用者指派的身分識別。 針對系統指派,請勿設定此欄位。 字串

UserIdentity

名稱 描述
identityType [必要]指定身分識別架構的類型。 'UserIdentity' (必要)

ResourceBaseProperties

名稱 描述
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串

ScheduleBase

名稱 描述
EndTime 以 ISO 8601 格式指定排程的結束時間。
如果沒有,排程將會無限期執行
字串
scheduleStatus 指定排程的狀態 'Disabled'
'Enabled'
startTime 指定 ISO 8601 格式的排程開始時間。 字串
timeZone 指定排程執行所在的時區。
TimeZone應遵循 Windows 時區格式。
字串
scheduleType 設定物件類型 Cron
需要週期 性 ()

CronSchedule

名稱 描述
scheduleType [必要]指定排程類型 'Cron' (必要)
expression [必要]指定排程的cron運算式。
表達式應該遵循 NCronTab 格式。
字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

RecurrenceSchedule

名稱 描述
scheduleType [必要]指定排程類型 必要的「週期」 ()
frequency [必要]指定要觸發排程的頻率 'Day'
'Hour'
'Minute'
'Month'
必要) ('Week'
interval [必要]指定排程間隔與頻率 int (必要)
模式 指定週期排程模式 RecurrencePattern

RecurrencePattern

名稱 描述
小時 [必要]週期排程模式的時數清單 int[] (必要)
分鐘 [必要]週期排程模式的分鐘清單 int[] (必要)
平日 週期排程模式的工作日清單 包含任何項目的字串數組:
'Friday'
'Monday'
'Saturday'
'Sunday'
'Thursday'
'Tuesday'
'Wednesday'

JobBaseServices

名稱 描述
{自定義屬性} JobService

JobService

名稱 描述
端點 端點的 URL。 字串
jobServiceType 端點類型。 string
連接埠 端點的埠。 int
properties 在端點上設定的其他屬性。 JobServiceProperties

JobServiceProperties

名稱 描述
{自定義屬性} 字串

AutoMLJob

名稱 描述
jobType [必要]指定作業的類型。 必要的 『AutoML』 ()
environmentId 作業環境規格的 ARM 資源識別碼。
這是選擇性值,如果未提供,AutoML 會在執行作業時,將此預設為生產 AutoML 策劃的環境版本。
字串
environmentVariables 作業中包含的環境變數。 AutoMLJobEnvironmentVariables
outputs 對應作業中使用的輸出數據系結。 AutoMLJobOutputs
resources 作業的計算資源組態。 ResourceConfiguration
taskDetails [必要]這代表可以是其中一個數據表/NLP/Image 的案例 需要 AutoMLVertical ()

AutoMLJobEnvironmentVariables

名稱 描述
{自定義屬性} 字串

AutoMLJobOutputs

名稱 描述
{自定義屬性} JobOutput

JobOutput

名稱 描述
description 輸出的描述。 字串
jobOutputType 設定物件類型 CustomModel
MLFlowModel
MLTable
TritonModel
UriFile
需要 uriFolder ()

CustomModelJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 'CustomModel' (必要)
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

MLFlowModelJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 必要) 的 'MLFlowModel' (
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

MLTableJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 'MLTable' (必要)
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

TritonModelJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 'TritonModel' (必要)
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

UriFileJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 'UriFile' (必要)
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

UriFolderJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 'UriFolder' (必要)
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

ResourceConfiguration

名稱 描述
instanceCount 計算目標所使用的選用執行個體或節點數目。 int
instanceType 計算目標所支持的選擇性 VM 類型。 字串
properties 其他屬性包。 ResourceConfigurationProperties

ResourceConfigurationProperties

名稱 描述
{自定義屬性} 針對 Bicep,您可以使用 any () 函式。

AutoMLVertical

名稱 描述
logVerbosity 作業的記錄詳細資訊。 'Critical'
'Debug'
'Error'
'Info'
'NotSet'
'Warning'
taskType 設定物件類型 分類
預測
ImageClassification
ImageClassificationMultilabel
ImageInstanceSegmentation
ImageObjectDetection
迴歸
TextClassification
TextClassificationMultilabel
需要 TextNER ()

分類

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 「分類」 (必要)
allowedModels 允許分類工作的模型。 包含任何項目的字串數組:
'BernoulliNaiveBaes'
'DecisionTree'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LightGBM'
'LinearSVM'
'LogisticRegression'
'MultinomialNaiveBaes'
'RandomForest'
'SGD'
'SVM'
'XGBoostClassifier'
blockedModels 分類工作的封鎖模型。 包含任何項目的字串數組:
'BernoulliNaiveBaes'
'DecisionTree'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LightGBM'
'LinearSVM'
'LogisticRegression'
'MultinomialNaiveBaes'
'RandomForest'
'SGD'
'SVM'
'XGBoostClassifier'
dataSettings AutoMLJob 的數據輸入。 TableVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 TableVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 TableVerticalLimitSettings
primaryMetric 工作的主要計量。 'AUCWeighted'
'Accuracy'
'AveragePrecisionScoreWeighted'
'NormMacroRecall'
'PrecisionScoreWeighted'
trainingSettings AutoML 作業定型階段的輸入。 TrainingSettings

TableVerticalDataSettings

名稱 描述
targetColumnName [必要]目標數據行名稱:這是預測值數據行。
也稱為分類工作內容中的標籤數據行名稱。
字串 (必要)

約束:
模式 = [a-zA-Z0-9_]
testData 測試數據輸入。 TestDataSettings
trainingData [必要]定型數據輸入。 TrainingDataSettings (必要)
validationData 驗證數據輸入。 TableVerticalValidationDataSettings
weightColumnName 範例加權數據行的名稱。 自動化 ML 支援加權數據行做為輸入,導致數據中的數據列向上或向下加權。 字串

TestDataSettings

名稱 描述
資料 測試數據 MLTable。 MLTableJobInput
testDataSize 測試數據集的分數,必須保留以供驗證之用。
(0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

MLTableJobInput

名稱 描述
description 輸入的描述。 字串
jobInputType [必要]指定作業的類型。 'CustomModel'
'Literal'
'MLFlowModel'
'MLTable'
'TritonModel'
'UriFile'
'UriFolder' (必要)
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]

TrainingDataSettings

名稱 描述
資料 [必要]定型數據 MLTable。 MLTableJobInput (必要)

TableVerticalValidationDataSettings

名稱 描述
cvSplitColumnNames 要用於 CVSplit 資料的數據行。 string[]
資料 驗證數據 MLTable。 MLTableJobInput
nCrossValidations 要套用至定型數據集的交叉驗證折疊數目
未提供驗證數據集時。
NCrossValidations
validationDataSize 需要為驗證目的保留的定型數據集分數。
(0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

NCrossValidations

名稱 描述
mode 設定物件類型 Auto
需要自訂 ()

AutoNCrossValidations

名稱 描述
mode [必要]用於判斷 N 交叉驗證的模式。 需要 『Auto』 ()

CustomNCrossValidations

名稱 描述
mode [必要]用於判斷 N 交叉驗證的模式。 必要) 'Custom' (
[必要]N-Cross 驗證值。 需要 int ()

TableVerticalFeaturizationSettings

名稱 描述
blockedTransformers 這些轉換器不得用於特徵化。 string[]
columnNameAndTypes 數據行名稱的字典及其類型 (int、float、string、datetime 等) 。 TableVerticalFeaturizationSettingsColumnNameAndTypes
datasetLanguage 數據集語言,適用於文字數據。 字串
dropColumns 特徵化期間要從數據卸除的數據行。 string[]
enableDnnFeaturization 判斷是否要使用 Dnn 型特徵化工具進行數據特徵化。 bool
mode 特徵化模式 - 使用者可以保留預設的「自動」模式,而 AutoML 會處理特徵化階段中數據的必要轉換。
如果已選取 [關閉],則不會完成特徵化。
如果選取 [自定義],則使用者可以指定其他輸入,以自定義特徵化的完成方式。
'Auto'
'Custom'
'Off'
transformerParams 用戶可以指定要搭配其套用的數據行和轉換程式建構函式參數使用的其他轉換器。 TableVerticalFeaturizationSettingsTransformerParams

TableVerticalFeaturizationSettingsColumnNameAndTypes

名稱 描述
{自定義屬性} 字串

TableVerticalFeaturizationSettingsTransformerParams

名稱 描述
{自定義屬性} ColumnTransformer[]

ColumnTransformer

名稱 描述
fields 要套用轉換器邏輯的欄位。 string[]
參數 要傳遞至轉換器的不同屬性。
預期的輸入是 JSON 格式的索引鍵、值組字典。
針對 Bicep,您可以使用 any () 函式。

TableVerticalLimitSettings

名稱 描述
enableEarlyTermination 啟用提早終止,判斷如果最後 20 個反覆項目沒有分數改善,AutoMLJob 是否提早終止。 bool
exitScore AutoML 作業的結束分數。 int
maxConcurrentTrials 並行反覆專案上限。 int
maxCoresPerTrial 每次反覆運算的核心數上限。 int
maxTrials 反覆運算次數。 int
timeout AutoML 作業逾時。 字串
trialTimeout 反覆專案逾時。 字串

TrainingSettings

名稱 描述
enableDnnTraining 啟用 DNN 模型的建議。 bool
enableModelExplainability 在最佳模型上開啟可解釋性的旗標。 bool
enableOnnxCompatibleModels 啟用 onnx 相容模型的旗標。 bool
enableStackEnsemble 啟用堆疊Ensemble 執行。 bool
enableVoteEnsemble 啟用投票Ensemble 執行。 bool
ensembleModelDownloadTimeout 在 VotingEnsemble 和 StackEnsemble 模型產生期間,會下載先前子回合中的多個配適模型。
如果需要更多時間,請將此參數設定為高於 300 秒的值。
字串
stackEnsembleSettings 堆疊Ensemble 執行的堆疊Ensemble 設定。 StackEnsembleSettings

StackEnsembleSettings

名稱 描述
stackMetaLearnerKWargs 要傳遞至中繼學習模組初始化表達式的選擇性參數。 針對 Bicep,您可以使用 any () 函式。
stackMetaLearnerTrainPercentage 指定在選擇定型和定型類型的訓練類型時 (定型集的比例,) 保留給訓練中繼學習者。 預設值為 0.2。 int
stackMetaLearnerType 中繼學習模組是在個別異質模型的輸出上定型的模型。 'ElasticNet'
'ElasticNetCV'
'LightGBMClassifier'
'LightGBMRegressor'
'LinearRegression'
'LogisticRegression'
'LogisticRegressionCV'
'None'

預測

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 需要「預測」 ()
allowedModels 預測工作的允許模型。 包含任何項目的字串數組:
'Arimax'
'AutoArima'
'Average'
'DecisionTree'
'ElasticNet'
'ExponentialSmoothing'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LassoLars'
'LightGBM'
'Naive'
'撷拉'
'RandomForest'
'SGD'
'SeasonalAverage'
'SeasonalNaive'
'TCNForecaster'
'XGBoostRegressor'
blockedModels 預測工作的封鎖模型。 包含任何項目的字串數組:
'Arimax'
'AutoArima'
'Average'
'DecisionTree'
'ElasticNet'
'ExponentialSmoothing'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LassoLars'
'LightGBM'
'Naive'
'撷拉'
'RandomForest'
'SGD'
'SeasonalAverage'
'SeasonalNaive'
'TCNForecaster'
'XGBoostRegressor'
dataSettings AutoMLJob 的數據輸入。 TableVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 TableVerticalFeaturizationSettings
forecastingSettings 預測工作特定輸入。 ForecastingSettings
limitSettings AutoMLJob 的執行條件約束。 TableVerticalLimitSettings
primaryMetric 預測工作的主要計量。 'NormalizedMeanAbsoluteError'
'NormalizedRootMeanSquaredError'
'R2Score'
'SpearmanCorrelation'
trainingSettings AutoML 作業定型階段的輸入。 TrainingSettings

ForecastingSettings

名稱 描述
countryOrRegionForHolidays 用於預測工作的假日國家或地區。
這些應該是 ISO 3166 兩個字母的國家/地區代碼,例如 'US' 或 'GB'。
字串
cvStepSize 一個 CV 折疊的原點時間與下一個折疊之間的期間數。 For
例如,如果 CVStepSize = 3 表示每日數據,則每個折疊的原點時間將會是
三天分開。
int
featureLags 針對具有 『auto』 或 null 之數值特徵產生延遲的旗標。 'Auto'
'None'
forecastHorizon 所需的最大預測範圍,以時間序列頻率單位為單位。 ForecastHorizon
frequency 預測時,此參數代表預期預測的期間,例如每日、每周、每年等。預測頻率預設為數據集頻率。 字串
季節性 將時間序列季節性設定為數列頻率的整數倍數。
如果季節性設定為 『auto』,則會推斷它。
季節性
shortSeriesHandlingConfig 參數,定義 AutoML 是否應該處理簡短的時間序列。 'Auto'
'Drop'
'None'
'Pad'
targetAggregateFunction 用來匯總時間序列目標數據行以符合使用者指定頻率的函式。
如果 TargetAggregateFunction 設定為不是 『None』,但未設定 freq 參數,就會引發錯誤。 可能的目標匯總函數包括:「sum」、“ max”
'Max'
'Mean'
'Min'
'None'
'Sum'
targetLags 要從目標數據行延遲的過去期間數。 TargetLags
targetRollingWindowSize 過去期間數,用來建立目標數據行的滾動視窗平均值。 TargetRollingWindowSize
timeColumnName 時間數據行的名稱。 當預測指定用於建置時間序列並推斷其頻率的輸入數據中的 datetime 數據行時,需要此參數。 字串
timeSeriesIdColumnNames 用來分組時間範圍的數據行名稱。 它可以用來建立多個數列。
如果未定義粒紋,則會假設資料集為一個時間序列。 此參數會與工作類型預測搭配使用。
string[]
useStl 設定時間序列目標數據行的 STL 分解。 'None'
'Season'
'SeasonTrend'

ForecastHorizon

名稱 描述
mode 設定物件類型 Auto
自訂 (必要)

AutoForecastHorizon

名稱 描述
mode [必要]設定預測水準值選取模式。 需要 『Auto』 ()

CustomForecastHorizon

名稱 描述
mode [必要]設定預測水準值選取模式。 需要 『Custom』 ()
[必要]預測水準值。 int (必要)

季節性

名稱 描述
mode 設定物件類型 Auto
自訂 (必要)

AutoSeasonality

名稱 描述
mode [必要]季節性模式。 需要 『Auto』 ()

CustomSeasonality

名稱 描述
mode [必要]季節性模式。 需要 『Custom』 ()
[必要]季節性值。 int (必要)

TargetLags

名稱 描述
mode 設定物件類型 Auto
自訂 (必要)

AutoTargetLags

名稱 描述
mode [必要]設定目標延遲模式 - 自動/自定義 需要 『Auto』 ()

CustomTargetLags

名稱 描述
mode [必要]設定目標延遲模式 - 自動/自定義 需要 『Custom』 ()
[必要]設定目標延遲值。 int[] (必要)

TargetRollingWindowSize

名稱 描述
mode 設定物件類型 Auto
自訂 (必要)

AutoTargetRollingWindowSize

名稱 描述
mode [必要]TargetRollingWindowSiz 偵測模式。 需要 『Auto』 ()

CustomTargetRollingWindowSize

名稱 描述
mode [必要]TargetRollingWindowSiz 偵測模式。 需要 『Custom』 ()
[必要]TargetRollingWindowSize 值。 int (必要)

ImageClassification

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'ImageClassification' (必要)
dataSettings [必要]註冊的表格式數據集標識碼和其他定型和驗證模型所需的數據設定集合。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsClassification
primaryMetric 要針對這項工作優化的主要計量。 'AUCWeighted'
'Accuracy'
'AveragePrecisionScoreWeighted'
'NormMacroRecall'
'PrecisionScoreWeighted'
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsClassification[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

ImageVerticalDataSettings

名稱 描述
targetColumnName [必要]目標數據行名稱:這是預測值數據行。
也稱為分類工作內容中的標籤數據行名稱。
字串 (必要)

約束:
模式 = [a-zA-Z0-9_]
testData 測試數據輸入。 TestDataSettings
trainingData [必要]定型數據輸入。 TrainingDataSettings (必要)
validationData 驗證數據集的設定。 ImageVerticalValidationDataSettings

ImageVerticalValidationDataSettings

名稱 描述
資料 驗證數據 MLTable。 MLTableJobInput
validationDataSize 必須針對驗證目的將定型數據集的分數。
介於 (0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

ImageLimitSettings

名稱 描述
maxConcurrentTrials 並行 AutoML 反覆項目的數目上限。 int
maxTrials AutoML 反覆項目的數目上限。 int
timeout AutoML 作業逾時。 字串

ImageModelSettingsClassification

名稱 描述
advancedSettings 進階案例的設定。 字串
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 bool
擴增 使用擴增的設定。 字串
beta1 優化器為 'adam' 或 'adamw' 時的 'beta1' 值。 必須是範圍 [0, 1] 內的浮點數。 int
beta2 優化器為 'adam' 或 'adamw' 時的 'beta2' 值。 必須是範圍 [0, 1] 內的浮點數。 int
checkpointDatasetId 預先定型檢查點的 FileDataset 識別碼, (累加式定型的) 。
請務必傳遞 CheckpointFilename 以及 CheckpointDatasetId。
字串
checkpointFilename FileDataset 中用於累加式定型的預先定型檢查點檔案名稱。
請務必傳遞 CheckpointDatasetId 以及 CheckpointFilename。
字串
checkpointFrequency 儲存模型檢查點的頻率。 必須是正整數。 int
checkpointRunId 先前執行的標識碼,其具有累加定型的預先定型檢查點。 字串
分散式 是否要使用分散式定型。 bool
earlyStopping 在定型期間啟用提早停止邏輯。 bool
earlyStoppingDelay 在主要計量改進之前等候的最小 Epoch 或驗證評估數目
會追蹤早期停止。 必須是正整數。
int
earlyStoppingPatience 之前沒有主要計量改進的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
int
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 bool
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 int
gradientAccumulationStep 漸層累積表示在沒有的情況下執行已設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
int
layersToFreeze 要凍結模型的圖層數目。 必須是正整數。
例如,傳遞 2 做為 'seresnext' 值表示
凍結第0層和第1層。 如需支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
int
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 int
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 'None'
'Step'
'WarmupCosine'
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請流覽官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
動量 優化工具為 『sgd』 時的時刻值。 必須是範圍 [0, 1] 內的浮點數。 int
nesterov 當優化器為 'sgd' 時,請啟用 nesterov。 bool
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 int
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 int
optimizer 最佳化工具的類型。 'Adam'
'Adamw'
'None'
'Sgd'
randomSeed 使用決定性定型時要使用的隨機種子。 int
splitRatio 如果未定義驗證數據,這會指定分割的分割比率
將數據定型成隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
int
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 int
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 int
trainingBatchSize 定型批次大小。 必須是正整數。 int
trainingCropSize 影像裁剪大小,為定型數據集的類神經網路輸入。 必須是正整數。 int
validationBatchSize 驗證批次大小。 必須是正整數。 int
validationCropSize 輸入至驗證數據集之類神經網路的影像裁剪大小。 必須是正整數。 int
validationResizeSize 在針對驗證資料集進行裁切之前,要調整大小的影像大小。 必須是正整數。 int
warmupCosineLRCycles 當學習速率排程器為 『warmup_cosine』 時,餘弦迴圈的值。 必須是範圍 [0, 1] 內的浮點數。 int
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』時,熱身 Epoch 的值。 必須是正整數。 int
weightDecay 優化器為 'sgd'、'adam' 或 'adamw' 時的權數衰減值。 必須是範圍[0, 1] 中的浮點數。 int
weightedLoss 加權損失。 接受的值是 0,不會遺失加權。
1 用於 sqrt. (class_weights) 加權損失。 2 表示具有 class_weights 的加權遺失。 必須是 0 或是 1 或 2。
int

ImageModelDistributionSettingsClassification

名稱 描述
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 字串
擴增 使用擴增的設定。 字串
beta1 當優化器為 'adam' 或 'adamw' 時,'beta1' 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
beta2 當優化器為 'adam' 或 'adamw' 時,'beta2' 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
分散式 是否要使用散發器定型。 字串
earlyStopping 在定型期間啟用提早停止邏輯。 字串
earlyStoppingDelay 在主要計量改進之前等待的最小 Epoch 或驗證評估數目
會追蹤提早停止。 必須是正整數。
字串
earlyStoppingPatience 在之前沒有主要計量改善的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
字串
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 字串
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 字串
gradientAccumulationStep 漸層累積表示執行未設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
字串
layersToFreeze 要凍結模型的層數。 必須是正整數。
例如,傳遞 2 作為 'seresnext' 值表示
凍結 layer0 和 layer1。 如需所支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 字串
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 字串
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請造訪官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
動量 優化工具為 『sgd』 時的動向值。 必須是範圍 [0, 1] 內的浮點數。 字串
nesterov 當優化器為 『sgd』 時啟用 nesterov。 字串
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 字串
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 字串
optimizer 最佳化工具的類型。 必須是 'sgd'、'adam' 或 'adamw'。 字串
randomSeed 使用決定性定型時要使用的隨機種子。 字串
splitRatio 如果未定義驗證數據,這會指定分割的分割比例
將數據定型為隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
字串
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 字串
trainingBatchSize 定型批次大小。 必須是正整數。 字串
trainingCropSize 影像裁剪大小,為定型數據集輸入類神經網路。 必須是正整數。 字串
validationBatchSize 驗證批次大小。 必須是正整數。 字串
validationCropSize 影像裁剪大小,這是驗證數據集之類神經網路的輸入。 必須是正整數。 字串
validationResizeSize 在針對驗證資料集進行裁切之前,要調整大小的影像大小。 必須是正整數。 字串
warmupCosineLRCycles 學習速率排程器為「warmup_cosine」時的餘弦週期值。 必須是範圍 [0, 1] 內的浮點數。 字串
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』 時,暖 Epoch 的值。 必須是正整數。 字串
weightDecay 當優化器為 'sgd'、'adam' 或 'adamw' 時,加權衰減的值。 必須是 range[0, 1] 中的浮點數。 字串
weightedLoss 加權損失。 接受的值為 0,不會遺失加權。
1 用於 sqrt. (class_weights) 的加權損失。 2 表示具有 class_weights 的加權遺失。 必須是 0 或是 1 或 2。
字串

ImageSweepSettings

名稱 描述
earlyTermination 早期終止原則的類型。 EarlyTerminationPolicy
限制 [必要]限制模型掃掠和超參數掃掠的設定。 ImageSweepLimitSettings (必要)
samplingAlgorithm [必要]超參數取樣演算法的類型。 'Bayesian'
'Grid'
需要 'Random' ()

EarlyTerminationPolicy

名稱 描述
delayEvaluation 延遲第一次評估的間隔數目。 int
evaluationInterval 原則評估之間) 執行的間隔 (數目。 int
policyType 設定物件類型 Bandit
MedianStopping
截斷選取 必要 ()

BanditPolicy

名稱 描述
policyType [必要]原則設定的名稱 需要 『Bandit』 ()
slackAmount 最佳執行所允許的絕對距離。 int
slackFactor 允許與最佳執行距離的比例。 int

MedianStoppingPolicy

名稱 描述
policyType [必要]原則設定的名稱 必要 ('MedianStopping')

TruncationSelectionPolicy

名稱 描述
policyType [必要]原則設定的名稱 'TruncationSelection' (必要)
truncationPercentage 要在每個評估間隔取消的執行百分比。 int

ImageSweepLimitSettings

名稱 描述
maxConcurrentTrials 基礎掃掠作業的並行反覆項目數目上限。 int
maxTrials 基礎掃掠作業的反覆項目數目上限。 int

ImageClassificationMultilabel

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'ImageClassificationMultilabel' (必要)
dataSettings [必要]已註冊表格式數據集標識碼的集合,以及定型和驗證模型所需的其他數據設定。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsClassification
primaryMetric 要針對這項工作優化的主要計量。 'AUCWeighted'
「精確度」
'AveragePrecisionScoreWeighted'
'IOU'
'NormMacroRecall'
'PrecisionScoreWeighted'
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsClassification[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

ImageInstanceSegmentation

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'ImageInstanceSegmentation' (必要)
dataSettings [必要]已註冊表格式數據集標識碼的集合,以及定型和驗證模型所需的其他數據設定。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsObjectDetection
primaryMetric 要針對這項工作優化的主要計量。 'MeanAveragePrecision'
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsObjectDetection[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

ImageModelSettingsObjectDetection

名稱 描述
advancedSettings 進階案例的設定。 字串
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 bool
擴增 使用擴增的設定。 字串
beta1 當優化器為 'adam' 或 'adamw' 時,'beta1' 的值。 必須是範圍 [0, 1] 內的浮點數。 int
beta2 當優化器為 'adam' 或 'adamw' 時,'beta2' 的值。 必須是範圍 [0, 1] 內的浮點數。 int
boxDetectionsPerImage 針對所有類別,每個影像的偵測數目上限。 必須是正整數。
注意:'yolov5' 演算法不支援此設定。
int
boxScoreThreshold 在推斷期間,只傳回分類分數大於的建議
BoxScoreThreshold。 必須是 range[0, 1] 中的浮點數。
int
checkpointDatasetId 預先定型檢查點的 FileDataset 識別碼, (累加式定型的) 。
請務必傳遞 CheckpointFilename 以及 CheckpointDatasetId。
字串
checkpointFilename FileDataset 中用於累加式定型的預先定型檢查點檔案名稱。
請務必傳遞 CheckpointDatasetId 以及 CheckpointFilename。
字串
checkpointFrequency 儲存模型檢查點的頻率。 必須是正整數。 int
checkpointRunId 先前執行之標識碼,其具有累加式定型的預先定型檢查點。 字串
分散式 是否要使用分散式定型。 bool
earlyStopping 在定型期間啟用提早停止邏輯。 bool
earlyStoppingDelay 在主要計量改進之前等待的最小 Epoch 或驗證評估數目
會追蹤提早停止。 必須是正整數。
int
earlyStoppingPatience 在之前沒有主要計量改善的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
int
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 bool
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 int
gradientAccumulationStep 漸層累積表示執行未設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
int
imageSize 用於定型和驗證的影像大小。 必須是正整數。
注意:如果大小太大,訓練回合可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
int
layersToFreeze 要凍結模型的層數。 必須是正整數。
例如,傳遞 2 作為 'seresnext' 值表示
凍結 layer0 和 layer1。 如需所支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
int
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 int
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 'None'
'Step'
'WarmupCosine'
maxSize 將影像饋送至骨幹之前要重新調整的影像大小上限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
int
minSize 將影像饋送至骨幹之前要重新調整的影像大小下限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
int
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請造訪官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
modelSize 模型大小。 必須是 'small'、'medium'、'large' 或 'xlarge'。
注意:如果模型大小太大,則定型執行可能會進入 CUDA OOM.
注意:只有 'yolov5' 演算法才支援此設定。
'ExtraLarge'
'Large'
'Medium'
'None'
'Small'
動量 優化工具為 『sgd』 時的動向值。 必須是範圍 [0, 1] 內的浮點數。 int
尺度 以 +/- 50% 的不同影像大小啟用多縮放影像。
注意:如果沒有足夠的 GPU 記憶體,則定型執行可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
bool
nesterov 當優化器為 『sgd』 時啟用 nesterov。 bool
nmsIouThreshold 在NMS後置處理中的推斷期間使用的IOU臨界值。 必須是範圍 [0, 1] 內的浮點數。 int
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 int
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 int
optimizer 最佳化工具的類型。 'Adam'
'Adamw'
'None'
'Sgd'
randomSeed 使用決定性定型時要使用的隨機種子。 int
splitRatio 如果未定義驗證數據,這會指定分割的分割比例
將數據定型為隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
int
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 int
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 int
tileGridSize 要用於每個並排影像的格線大小。 注意:TileGridSize 不得為
無可啟用小型物件偵測邏輯。 字串,包含兩個整數,格式為一個。
注意:'yolov5' 演算法不支援此設定。
字串
tileOverlapRatio 每個維度中相鄰圖格之間的重疊比例。 必須是範圍 [0, 1) 的 float。
注意:'yolov5' 演算法不支援此設定。
int
tilePredictionsNmsThreshold 從圖格和影像合併預測時用來執行 NMS 的 IOU 閾值。
用於驗證/推斷。 必須是範圍 [0, 1] 中的浮點數。
注意:'yolov5' 演算法不支援此設定。
int
trainingBatchSize 定型批次大小。 必須是正整數。 int
validationBatchSize 驗證批次大小。 必須是正整數。 int
validationIouThreshold 計算驗證計量時要使用的IOU臨界值。 必須是範圍 [0, 1] 中的浮點數。 int
validationMetricType 要用於驗證計量的計量計算方法。 'Coco'
'CocoVoc'
'None'
'Voc'
warmupCosineLRCycles 學習速率排程器為「warmup_cosine」時的餘弦週期值。 必須是範圍 [0, 1] 內的浮點數。 int
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』 時,暖 Epoch 的值。 必須是正整數。 int
weightDecay 當優化器為 'sgd'、'adam' 或 'adamw' 時,加權衰減的值。 必須是 range[0, 1] 中的浮點數。 int

ImageModelDistributionSettingsObjectDetection

名稱 描述
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 字串
擴增 使用擴增的設定。 字串
beta1 當優化器為 'adam' 或 'adamw' 時,'beta1' 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
beta2 當優化器為 'adam' 或 'adamw' 時,'beta2' 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
boxDetectionsPerImage 針對所有類別,每個影像的偵測數目上限。 必須是正整數。
注意:'yolov5' 演算法不支援此設定。
字串
boxScoreThreshold 在推斷期間,只傳回分類分數大於的建議
BoxScoreThreshold。 必須是範圍[0, 1] 中的浮點數。
字串
分散式 是否要使用分散式程式定型。 字串
earlyStopping 在定型期間啟用提早停止邏輯。 字串
earlyStoppingDelay 在主要計量改進之前等候的最小 Epoch 或驗證評估數目
會追蹤早期停止。 必須是正整數。
字串
earlyStoppingPatience 之前沒有主要計量改進的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
字串
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 字串
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 字串
gradientAccumulationStep 漸層累積表示在沒有的情況下執行已設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
字串
imageSize 用於定型和驗證的影像大小。 必須是正整數。
注意:如果大小太大,定型回合可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
字串
layersToFreeze 要凍結模型的圖層數目。 必須是正整數。
例如,傳遞 2 做為 'seresnext' 值表示
凍結第0層和第1層。 如需支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 字串
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 字串
maxSize 將影像饋送至骨幹之前要重新調整的影像大小上限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
字串
minSize 將影像饋送至骨幹之前要重新調整的影像大小下限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
字串
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請流覽官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
modelSize 模型大小。 必須是 'small'、'medium'、'large' 或 'xlarge'。
注意:如果模型大小太大,則定型執行可能會進入 CUDA OOM.
注意:只有 'yolov5' 演算法才支援此設定。
字串
動量 優化工具為 『sgd』 時的時刻值。 必須是範圍 [0, 1] 內的浮點數。 字串
尺度 依 +/- 50% 的不同影像大小啟用多縮放映射。
注意:如果沒有足夠的 GPU 記憶體,則定型執行可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
字串
nesterov 當優化器為 'sgd' 時,請啟用 nesterov。 字串
nmsIouThreshold 在NMS後置處理中推斷期間使用的IOU臨界值。 必須是範圍 [0, 1] 中的 float。 字串
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 字串
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 字串
optimizer 最佳化工具的類型。 必須是 『sgd』、『adam』或 『adamw』。 字串
randomSeed 使用決定性定型時要使用的隨機種子。 字串
splitRatio 如果未定義驗證數據,這會指定分割的分割比率
將數據定型成隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
字串
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 字串
tileGridSize 要用於每個並排影像的格線大小。 注意:TileGridSize 不得為
無可啟用小型物件偵測邏輯。 包含兩個整數的字串,其格式為#。
注意:'yolov5' 演算法不支援此設定。
字串
tileOverlapRatio 每個維度中相鄰圖格之間的重疊比例。 範圍 [0, 1) 必須是 float。
注意:'yolov5' 演算法不支援此設定。
字串
tilePredictionsNmsThreshold 從圖格和影像合併預測時用來執行 NMS 的 IOU 閾值。
用於驗證/推斷。 必須是範圍 [0, 1] 中的 float。
注意:'yolov5' 演算法不支援此設定。
NMS:非最大隱藏
字串
trainingBatchSize 定型批次大小。 必須是正整數。 字串
validationBatchSize 驗證批次大小。 必須是正整數。 字串
validationIouThreshold 計算驗證計量時要使用的IOU臨界值。 必須是範圍 [0, 1] 中的 float。 字串
validationMetricType 要用於驗證計量的計量計算方法。 必須是 『none』、'coco'、'voc' 或 'coco_voc'。 字串
warmupCosineLRCycles 當學習速率排程器為 『warmup_cosine』 時,餘弦迴圈的值。 必須是範圍 [0, 1] 內的浮點數。 字串
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』時,熱身 Epoch 的值。 必須是正整數。 字串
weightDecay 優化器為 'sgd'、'adam' 或 'adamw' 時的權數衰減值。 必須是範圍[0, 1] 中的浮點數。 字串

ImageObjectDetection

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'ImageObjectDetection' (必要)
dataSettings [必要]註冊的表格式數據集標識碼和其他定型和驗證模型所需的數據設定集合。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsObjectDetection
primaryMetric 要針對這項工作優化的主要計量。 'MeanAveragePrecision'
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsObjectDetection[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

迴歸

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 需要 「回歸」 ()
allowedModels 回歸工作的允許模型。 包含任何項目的字串數組:
'DecisionTree'
'ElasticNet'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LassoLars'
'LightGBM'
'RandomForest'
'SGD'
'XGBoostRegressor'
blockedModels 回歸工作的封鎖模型。 包含任何項目的字串數組:
'DecisionTree'
'ElasticNet'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LassoLars'
'LightGBM'
'RandomForest'
'SGD'
'XGBoostRegressor'
dataSettings AutoMLJob 的數據輸入。 TableVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 TableVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 TableVerticalLimitSettings
primaryMetric 回歸工作的主要計量。 'NormalizedMeanAbsoluteError'
'NormalizedRootMeanSquaredError'
'R2Score'
'SpearmanCorrelation'
trainingSettings AutoML 作業定型階段的輸入。 TrainingSettings

TextClassification

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'TextClassification' (必要)
dataSettings AutoMLJob 的數據輸入。 NlpVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 NlpVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 NlpVerticalLimitSettings
primaryMetric Text-Classification 工作的主要計量。 'AUCWeighted'
'Accuracy'
'AveragePrecisionScoreWeighted'
'NormMacroRecall'
'PrecisionScoreWeighted'

NlpVerticalDataSettings

名稱 描述
targetColumnName [必要]目標數據行名稱:這是預測值數據行。
也稱為分類工作內容中的標籤數據行名稱。
字串 (必要)

約束:
模式 = [a-zA-Z0-9_]
testData 測試數據輸入。 TestDataSettings
trainingData [必要]定型數據輸入。 TrainingDataSettings (必要)
validationData 驗證數據輸入。 NlpVerticalValidationDataSettings

NlpVerticalValidationDataSettings

名稱 描述
資料 驗證數據 MLTable。 MLTableJobInput
validationDataSize 必須針對驗證目的將定型數據集的分數。
介於 (0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

NlpVerticalFeaturizationSettings

名稱 描述
datasetLanguage 數據集語言,適用於文字數據。 字串

NlpVerticalLimitSettings

名稱 描述
maxConcurrentTrials 並行 AutoML 反覆專案上限。 int
maxTrials AutoML 反覆項目的數目。 int
timeout AutoML 作業逾時。 字串

TextClassificationMultilabel

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'TextClassificationMultilabel' (必要)
dataSettings AutoMLJob 的數據輸入。 NlpVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 NlpVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 NlpVerticalLimitSettings

TextNer

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 需要 『TextNER』 ()
dataSettings AutoMLJob 的數據輸入。 NlpVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 NlpVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 NlpVerticalLimitSettings

CommandJob

名稱 描述
jobType [必要]指定作業的類型。 'Command' (必要)
codeId 程式代碼資產的 ARM 資源識別碼。 字串
命令 [必要]在作業啟動時執行的命令。 例如 “python train.py” 字串 (必要)

約束:
最小長度 = 1
模式 = [a-zA-Z0-9_]
散發 (distribution) 作業的散發組態。 如果設定,這應該是 Mpi、Tensorflow、PyTorch 或 null 的其中一個。 DistributionConfiguration
environmentId [必要]作業環境規格的 ARM 資源識別碼。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]
environmentVariables 作業中包含的環境變數。 CommandJobEnvironmentVariables
輸入 對應作業中使用的輸入數據系結。 CommandJobInputs
限制 命令作業限制。 CommandJobLimits
outputs 對應作業中使用的輸出數據系結。 CommandJobOutputs
resources 作業的計算資源組態。 ResourceConfiguration

DistributionConfiguration

名稱 描述
distributionType 設定物件類型 Mpi
PyTorch
TensorFlow (必要)

Mpi

名稱 描述
distributionType [必要]指定散發架構的類型。 需要 『Mpi』 ()
processCountPerInstance 每個 MPI 節點的進程數目。 int

PyTorch

名稱 描述
distributionType [必要]指定散發架構的類型。 需要 『PyTorch』 ()
processCountPerInstance 每個節點的進程數目。 int

TensorFlow

名稱 描述
distributionType [必要]指定散發架構的類型。 需要 『TensorFlow』 ()
parameterServerCount 參數伺服器工作的數目。 int
workerCount 背景工作數目。 如果未指定,則會預設為實例計數。 int

CommandJobEnvironmentVariables

名稱 描述
{自定義屬性} 字串

CommandJobInputs

名稱 描述
{自定義屬性} JobInput

JobInput

名稱 描述
description 輸入的描述。 字串
jobInputType 設定物件類型 CustomModel
文字
MLFlowModel
MLTable
TritonModel
UriFile
需要 UriFolder ()

CustomModelJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 需要 『CustomModel』 ()
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]

LiteralJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 必要) 'Literal' (
[必要]輸入的常值。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]

MLFlowModelJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 必要) 的 'MLFlowModel' (
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

TritonModelJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 'TritonModel' (必要)
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

UriFileJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 'UriFile' (必要)
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

UriFolderJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 'UriFolder' (必要)
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

CommandJobLimits

名稱 描述
jobLimitsType [必要]JobLimit 類型。 'Command'
需要 「掃掠」 ()
timeout ISO 8601 格式的最大執行持續時間,之後作業將會取消。 僅支援有效位數為秒數的持續時間。 字串

CommandJobOutputs

名稱 描述
{自定義屬性} JobOutput

PipelineJob

名稱 描述
jobType [必要]指定作業的類型。 「管線」 (必要)
輸入 管線作業的輸入。 PipelineJobInputs
jobs 作業會建構管線作業。 PipelineJobJobs
outputs 管線作業的輸出 PipelineJobOutputs
設定 管線設定,適用於 ContinueRunOnStepFailure 等專案。 針對 Bicep,您可以使用 any () 函式。

PipelineJobInputs

名稱 描述
{自定義屬性} JobInput

PipelineJobJobs

名稱 描述
{自定義屬性} 針對 Bicep,您可以使用 any () 函式。

PipelineJobOutputs

名稱 描述
{自定義屬性} JobOutput

SweepJob

名稱 描述
jobType [必要]指定作業的類型。 需要 「掃掠」 ()
earlyTermination 早期終止原則可在完成之前取消效能不佳的執行 EarlyTerminationPolicy
輸入 對應作業中使用的輸入數據系結。 SweepJobInputs
限制 掃掠作業限制。 SweepJobLimits
objective [必要]優化目標。 必要目標 ()
outputs 對應作業中使用的輸出數據系結。 SweepJobOutputs
samplingAlgorithm [必要]超參數取樣演算法 SamplingAlgorithm (必要)
searchSpace [必要]包含每個參數及其分佈的字典。 字典索引鍵是參數的名稱 針對 Bicep,您可以使用 任何 () function. (必要)
試用版 [必要]試用版元件定義。 需要 TrialComponent ()

SweepJobInputs

名稱 描述
{自定義屬性} JobInput

SweepJobLimits

名稱 描述
jobLimitsType [必要]JobLimit 類型。 'Command'
需要 「掃掠」 ()
maxConcurrentTrials 掃掠作業最大並行試用版。 int
maxTotalTrials 掃掠作業總數上限。 int
timeout ISO 8601 格式的最大執行持續時間,之後作業將會取消。 僅支援有效位數為秒數的持續時間。 字串
trialTimeout 掃掠作業試用版逾時值。 字串

目標

名稱 描述
goal [必要]定義超參數微調支援的計量目標 '最大化'
需要 「最小化」 ()
primaryMetric [必要]要優化的計量名稱。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]

SweepJobOutputs

名稱 描述
{自定義屬性} JobOutput

SamplingAlgorithm

名稱 描述
samplingAlgorithmType 設定物件類型 貝 葉 斯
方格
需要隨機 ()

BayesianSamplingAlgorithm

名稱 描述
samplingAlgorithmType [必要]用來產生超參數值的演算法,以及組態屬性 需要 『Bayesian』 ()

GridSamplingAlgorithm

名稱 描述
samplingAlgorithmType [必要]用來產生超參數值的演算法,以及組態屬性 'Grid' (必要)

RandomSamplingAlgorithm

名稱 描述
samplingAlgorithmType [必要]用來產生超參數值的演算法,以及組態屬性 需要 'Random' ()
rule (規則) 隨機演算法的特定類型 'Random'
'Sobol'
seed 要作為隨機數產生種子的選擇性整數 int

TrialComponent

名稱 描述
codeId 程式代碼資產的 ARM 資源識別碼。 字串
命令 [必要]在作業啟動時執行的命令。 例如 “python train.py” 需要字串 ()

約束:
最小長度 = 1
模式 = [a-zA-Z0-9_]
散發 (distribution) 作業的散發組態。 如果設定,這應該是 Mpi、Tensorflow、PyTorch 或 null 的其中一個。 DistributionConfiguration
environmentId [必要]作業之環境規格的 ARM 資源識別碼。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]
environmentVariables 作業中包含的環境變數。 TrialComponentEnvironmentVariables
resources 作業的計算資源組態。 ResourceConfiguration

TrialComponentEnvironmentVariables

名稱 描述
{自定義屬性} 字串

快速入門範本

下列快速入門範本會部署此資源類型。

範本 描述
建立 Azure Machine Learning AutoML 分類作業

部署至 Azure
此範本會建立 Azure Machine Learning AutoML 分類作業,以找出預測用戶端是否會訂閱與金融機構的固定期限金的最佳模型。
建立 Azure Machine Learning 命令作業

部署至 Azure
此範本會建立具有基本hello_world腳本的 Azure Machine Learning 命令作業
建立 Azure Machine Learning 掃掠作業

部署至 Azure
此範本會建立用於超參數微調的 Azure Machine Learning 掃掠作業。

ARM 範本資源定義

工作區/作業資源類型可以使用目標作業進行部署:

如需每個 API 版本中已變更屬性的清單,請參閱 變更記錄

資源格式

若要建立 Microsoft.MachineLearningServices/workspaces/jobs 資源,請將下列 JSON 新增至您的範本。

{
  "type": "Microsoft.MachineLearningServices/workspaces/jobs",
  "apiVersion": "2022-02-01-preview",
  "name": "string",
  "properties": {
    "computeId": "string",
    "description": "string",
    "displayName": "string",
    "experimentName": "string",
    "identity": {
      "identityType": "string"
      // For remaining properties, see IdentityConfiguration objects
    },
    "isArchived": "bool",
    "properties": {
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string"
    },
    "schedule": {
      "endTime": "string",
      "scheduleStatus": "string",
      "startTime": "string",
      "timeZone": "string",
      "scheduleType": "string"
      // For remaining properties, see ScheduleBase objects
    },
    "services": {
      "{customized property}": {
        "endpoint": "string",
        "jobServiceType": "string",
        "port": "int",
        "properties": {
          "{customized property}": "string"
        }
      }
    },
    "tags": {
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string",
      "{customized property}": "string"
    },
    "jobType": "string"
    // For remaining properties, see JobBaseDetails objects
  }
}

JobBaseDetails 物件

設定 jobType 屬性以指定物件的類型。

針對 AutoML,請使用:

  "jobType": "AutoML",
  "environmentId": "string",
  "environmentVariables": {
    "{customized property}": "string"
  },
  "outputs": {
    "{customized property}": {
      "description": "string",
      "jobOutputType": "string"
      // For remaining properties, see JobOutput objects
    }
  },
  "resources": {
    "instanceCount": "int",
    "instanceType": "string",
    "properties": {
      "{customized property}": {}
    }
  },
  "taskDetails": {
    "logVerbosity": "string",
    "taskType": "string"
    // For remaining properties, see AutoMLVertical objects
  }

針對 [命令],請使用:

  "jobType": "Command",
  "codeId": "string",
  "command": "string",
  "distribution": {
    "distributionType": "string"
    // For remaining properties, see DistributionConfiguration objects
  },
  "environmentId": "string",
  "environmentVariables": {
    "{customized property}": "string"
  },
  "inputs": {
    "{customized property}": {
      "description": "string",
      "jobInputType": "string"
      // For remaining properties, see JobInput objects
    }
  },
  "limits": {
    "jobLimitsType": "string",
    "timeout": "string"
  },
  "outputs": {
    "{customized property}": {
      "description": "string",
      "jobOutputType": "string"
      // For remaining properties, see JobOutput objects
    }
  },
  "resources": {
    "instanceCount": "int",
    "instanceType": "string",
    "properties": {
      "{customized property}": {}
    }
  }

針對 管線,請使用:

  "jobType": "Pipeline",
  "inputs": {
    "{customized property}": {
      "description": "string",
      "jobInputType": "string"
      // For remaining properties, see JobInput objects
    }
  },
  "jobs": {
    "{customized property}": {}
  },
  "outputs": {
    "{customized property}": {
      "description": "string",
      "jobOutputType": "string"
      // For remaining properties, see JobOutput objects
    }
  },
  "settings": {}

針對 掃掠,請使用:

  "jobType": "Sweep",
  "earlyTermination": {
    "delayEvaluation": "int",
    "evaluationInterval": "int",
    "policyType": "string"
    // For remaining properties, see EarlyTerminationPolicy objects
  },
  "inputs": {
    "{customized property}": {
      "description": "string",
      "jobInputType": "string"
      // For remaining properties, see JobInput objects
    }
  },
  "limits": {
    "jobLimitsType": "string",
    "maxConcurrentTrials": "int",
    "maxTotalTrials": "int",
    "timeout": "string",
    "trialTimeout": "string"
  },
  "objective": {
    "goal": "string",
    "primaryMetric": "string"
  },
  "outputs": {
    "{customized property}": {
      "description": "string",
      "jobOutputType": "string"
      // For remaining properties, see JobOutput objects
    }
  },
  "samplingAlgorithm": {
    "samplingAlgorithmType": "string"
    // For remaining properties, see SamplingAlgorithm objects
  },
  "searchSpace": {},
  "trial": {
    "codeId": "string",
    "command": "string",
    "distribution": {
      "distributionType": "string"
      // For remaining properties, see DistributionConfiguration objects
    },
    "environmentId": "string",
    "environmentVariables": {
      "{customized property}": "string"
    },
    "resources": {
      "instanceCount": "int",
      "instanceType": "string",
      "properties": {
        "{customized property}": {}
      }
    }
  }

IdentityConfiguration 物件

設定 identityType 屬性以指定物件的類型。

針對 AMLToken,請使用:

  "identityType": "AMLToken"

針對 Managed,請使用:

  "identityType": "Managed",
  "clientId": "string",
  "objectId": "string",
  "resourceId": "string"

針對 UserIdentity,請使用:

  "identityType": "UserIdentity"

ScheduleBase 物件

設定 scheduleType 屬性以指定物件的類型。

針對 Cron,請使用:

  "scheduleType": "Cron",
  "expression": "string"

針對 [週期],請使用:

  "scheduleType": "Recurrence",
  "frequency": "string",
  "interval": "int",
  "pattern": {
    "hours": [ "int" ],
    "minutes": [ "int" ],
    "weekdays": [ "string" ]
  }

JobOutput 物件

設定 jobOutputType 屬性以指定對象的類型。

針對 CustomModel,請使用:

  "jobOutputType": "CustomModel",
  "mode": "string",
  "uri": "string"

針對 MLFlowModel,請使用:

  "jobOutputType": "MLFlowModel",
  "mode": "string",
  "uri": "string"

針對 MLTable,請使用:

  "jobOutputType": "MLTable",
  "mode": "string",
  "uri": "string"

針對 TritonModel,請使用:

  "jobOutputType": "TritonModel",
  "mode": "string",
  "uri": "string"

針對 UriFile,請使用:

  "jobOutputType": "UriFile",
  "mode": "string",
  "uri": "string"

針對 UriFolder,請使用:

  "jobOutputType": "UriFolder",
  "mode": "string",
  "uri": "string"

AutoMLVertical 物件

設定 taskType 屬性以指定物件的類型。

針對 [分類],請使用:

  "taskType": "Classification",
  "allowedModels": [ "string" ],
  "blockedModels": [ "string" ],
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "cvSplitColumnNames": [ "string" ],
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "nCrossValidations": {
        "mode": "string"
        // For remaining properties, see NCrossValidations objects
      },
      "validationDataSize": "int"
    },
    "weightColumnName": "string"
  },
  "featurizationSettings": {
    "blockedTransformers": [ "string" ],
    "columnNameAndTypes": {
      "{customized property}": "string"
    },
    "datasetLanguage": "string",
    "dropColumns": [ "string" ],
    "enableDnnFeaturization": "bool",
    "mode": "string",
    "transformerParams": {
      "{customized property}": [
        {
          "fields": [ "string" ],
          "parameters": {}
        }
      ]
    }
  },
  "limitSettings": {
    "enableEarlyTermination": "bool",
    "exitScore": "int",
    "maxConcurrentTrials": "int",
    "maxCoresPerTrial": "int",
    "maxTrials": "int",
    "timeout": "string",
    "trialTimeout": "string"
  },
  "primaryMetric": "string",
  "trainingSettings": {
    "enableDnnTraining": "bool",
    "enableModelExplainability": "bool",
    "enableOnnxCompatibleModels": "bool",
    "enableStackEnsemble": "bool",
    "enableVoteEnsemble": "bool",
    "ensembleModelDownloadTimeout": "string",
    "stackEnsembleSettings": {
      "stackMetaLearnerKWargs": {},
      "stackMetaLearnerTrainPercentage": "int",
      "stackMetaLearnerType": "string"
    }
  }

針對 預測,請使用:

  "taskType": "Forecasting",
  "allowedModels": [ "string" ],
  "blockedModels": [ "string" ],
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "cvSplitColumnNames": [ "string" ],
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "nCrossValidations": {
        "mode": "string"
        // For remaining properties, see NCrossValidations objects
      },
      "validationDataSize": "int"
    },
    "weightColumnName": "string"
  },
  "featurizationSettings": {
    "blockedTransformers": [ "string" ],
    "columnNameAndTypes": {
      "{customized property}": "string"
    },
    "datasetLanguage": "string",
    "dropColumns": [ "string" ],
    "enableDnnFeaturization": "bool",
    "mode": "string",
    "transformerParams": {
      "{customized property}": [
        {
          "fields": [ "string" ],
          "parameters": {}
        }
      ]
    }
  },
  "forecastingSettings": {
    "countryOrRegionForHolidays": "string",
    "cvStepSize": "int",
    "featureLags": "string",
    "forecastHorizon": {
      "mode": "string"
      // For remaining properties, see ForecastHorizon objects
    },
    "frequency": "string",
    "seasonality": {
      "mode": "string"
      // For remaining properties, see Seasonality objects
    },
    "shortSeriesHandlingConfig": "string",
    "targetAggregateFunction": "string",
    "targetLags": {
      "mode": "string"
      // For remaining properties, see TargetLags objects
    },
    "targetRollingWindowSize": {
      "mode": "string"
      // For remaining properties, see TargetRollingWindowSize objects
    },
    "timeColumnName": "string",
    "timeSeriesIdColumnNames": [ "string" ],
    "useStl": "string"
  },
  "limitSettings": {
    "enableEarlyTermination": "bool",
    "exitScore": "int",
    "maxConcurrentTrials": "int",
    "maxCoresPerTrial": "int",
    "maxTrials": "int",
    "timeout": "string",
    "trialTimeout": "string"
  },
  "primaryMetric": "string",
  "trainingSettings": {
    "enableDnnTraining": "bool",
    "enableModelExplainability": "bool",
    "enableOnnxCompatibleModels": "bool",
    "enableStackEnsemble": "bool",
    "enableVoteEnsemble": "bool",
    "ensembleModelDownloadTimeout": "string",
    "stackEnsembleSettings": {
      "stackMetaLearnerKWargs": {},
      "stackMetaLearnerTrainPercentage": "int",
      "stackMetaLearnerType": "string"
    }
  }

針對 ImageClassification,請使用:

  "taskType": "ImageClassification",
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "validationDataSize": "int"
    }
  },
  "limitSettings": {
    "maxConcurrentTrials": "int",
    "maxTrials": "int",
    "timeout": "string"
  },
  "modelSettings": {
    "advancedSettings": "string",
    "amsGradient": "bool",
    "augmentations": "string",
    "beta1": "int",
    "beta2": "int",
    "checkpointDatasetId": "string",
    "checkpointFilename": "string",
    "checkpointFrequency": "int",
    "checkpointRunId": "string",
    "distributed": "bool",
    "earlyStopping": "bool",
    "earlyStoppingDelay": "int",
    "earlyStoppingPatience": "int",
    "enableOnnxNormalization": "bool",
    "evaluationFrequency": "int",
    "gradientAccumulationStep": "int",
    "layersToFreeze": "int",
    "learningRate": "int",
    "learningRateScheduler": "string",
    "modelName": "string",
    "momentum": "int",
    "nesterov": "bool",
    "numberOfEpochs": "int",
    "numberOfWorkers": "int",
    "optimizer": "string",
    "randomSeed": "int",
    "splitRatio": "int",
    "stepLRGamma": "int",
    "stepLRStepSize": "int",
    "trainingBatchSize": "int",
    "trainingCropSize": "int",
    "validationBatchSize": "int",
    "validationCropSize": "int",
    "validationResizeSize": "int",
    "warmupCosineLRCycles": "int",
    "warmupCosineLRWarmupEpochs": "int",
    "weightDecay": "int",
    "weightedLoss": "int"
  },
  "primaryMetric": "string",
  "searchSpace": [
    {
      "amsGradient": "string",
      "augmentations": "string",
      "beta1": "string",
      "beta2": "string",
      "distributed": "string",
      "earlyStopping": "string",
      "earlyStoppingDelay": "string",
      "earlyStoppingPatience": "string",
      "enableOnnxNormalization": "string",
      "evaluationFrequency": "string",
      "gradientAccumulationStep": "string",
      "layersToFreeze": "string",
      "learningRate": "string",
      "learningRateScheduler": "string",
      "modelName": "string",
      "momentum": "string",
      "nesterov": "string",
      "numberOfEpochs": "string",
      "numberOfWorkers": "string",
      "optimizer": "string",
      "randomSeed": "string",
      "splitRatio": "string",
      "stepLRGamma": "string",
      "stepLRStepSize": "string",
      "trainingBatchSize": "string",
      "trainingCropSize": "string",
      "validationBatchSize": "string",
      "validationCropSize": "string",
      "validationResizeSize": "string",
      "warmupCosineLRCycles": "string",
      "warmupCosineLRWarmupEpochs": "string",
      "weightDecay": "string",
      "weightedLoss": "string"
    }
  ],
  "sweepSettings": {
    "earlyTermination": {
      "delayEvaluation": "int",
      "evaluationInterval": "int",
      "policyType": "string"
      // For remaining properties, see EarlyTerminationPolicy objects
    },
    "limits": {
      "maxConcurrentTrials": "int",
      "maxTrials": "int"
    },
    "samplingAlgorithm": "string"
  }

針對 ImageClassificationMultilabel,請使用:

  "taskType": "ImageClassificationMultilabel",
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "validationDataSize": "int"
    }
  },
  "limitSettings": {
    "maxConcurrentTrials": "int",
    "maxTrials": "int",
    "timeout": "string"
  },
  "modelSettings": {
    "advancedSettings": "string",
    "amsGradient": "bool",
    "augmentations": "string",
    "beta1": "int",
    "beta2": "int",
    "checkpointDatasetId": "string",
    "checkpointFilename": "string",
    "checkpointFrequency": "int",
    "checkpointRunId": "string",
    "distributed": "bool",
    "earlyStopping": "bool",
    "earlyStoppingDelay": "int",
    "earlyStoppingPatience": "int",
    "enableOnnxNormalization": "bool",
    "evaluationFrequency": "int",
    "gradientAccumulationStep": "int",
    "layersToFreeze": "int",
    "learningRate": "int",
    "learningRateScheduler": "string",
    "modelName": "string",
    "momentum": "int",
    "nesterov": "bool",
    "numberOfEpochs": "int",
    "numberOfWorkers": "int",
    "optimizer": "string",
    "randomSeed": "int",
    "splitRatio": "int",
    "stepLRGamma": "int",
    "stepLRStepSize": "int",
    "trainingBatchSize": "int",
    "trainingCropSize": "int",
    "validationBatchSize": "int",
    "validationCropSize": "int",
    "validationResizeSize": "int",
    "warmupCosineLRCycles": "int",
    "warmupCosineLRWarmupEpochs": "int",
    "weightDecay": "int",
    "weightedLoss": "int"
  },
  "primaryMetric": "string",
  "searchSpace": [
    {
      "amsGradient": "string",
      "augmentations": "string",
      "beta1": "string",
      "beta2": "string",
      "distributed": "string",
      "earlyStopping": "string",
      "earlyStoppingDelay": "string",
      "earlyStoppingPatience": "string",
      "enableOnnxNormalization": "string",
      "evaluationFrequency": "string",
      "gradientAccumulationStep": "string",
      "layersToFreeze": "string",
      "learningRate": "string",
      "learningRateScheduler": "string",
      "modelName": "string",
      "momentum": "string",
      "nesterov": "string",
      "numberOfEpochs": "string",
      "numberOfWorkers": "string",
      "optimizer": "string",
      "randomSeed": "string",
      "splitRatio": "string",
      "stepLRGamma": "string",
      "stepLRStepSize": "string",
      "trainingBatchSize": "string",
      "trainingCropSize": "string",
      "validationBatchSize": "string",
      "validationCropSize": "string",
      "validationResizeSize": "string",
      "warmupCosineLRCycles": "string",
      "warmupCosineLRWarmupEpochs": "string",
      "weightDecay": "string",
      "weightedLoss": "string"
    }
  ],
  "sweepSettings": {
    "earlyTermination": {
      "delayEvaluation": "int",
      "evaluationInterval": "int",
      "policyType": "string"
      // For remaining properties, see EarlyTerminationPolicy objects
    },
    "limits": {
      "maxConcurrentTrials": "int",
      "maxTrials": "int"
    },
    "samplingAlgorithm": "string"
  }

針對 ImageInstanceSegmentation,請使用:

  "taskType": "ImageInstanceSegmentation",
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "validationDataSize": "int"
    }
  },
  "limitSettings": {
    "maxConcurrentTrials": "int",
    "maxTrials": "int",
    "timeout": "string"
  },
  "modelSettings": {
    "advancedSettings": "string",
    "amsGradient": "bool",
    "augmentations": "string",
    "beta1": "int",
    "beta2": "int",
    "boxDetectionsPerImage": "int",
    "boxScoreThreshold": "int",
    "checkpointDatasetId": "string",
    "checkpointFilename": "string",
    "checkpointFrequency": "int",
    "checkpointRunId": "string",
    "distributed": "bool",
    "earlyStopping": "bool",
    "earlyStoppingDelay": "int",
    "earlyStoppingPatience": "int",
    "enableOnnxNormalization": "bool",
    "evaluationFrequency": "int",
    "gradientAccumulationStep": "int",
    "imageSize": "int",
    "layersToFreeze": "int",
    "learningRate": "int",
    "learningRateScheduler": "string",
    "maxSize": "int",
    "minSize": "int",
    "modelName": "string",
    "modelSize": "string",
    "momentum": "int",
    "multiScale": "bool",
    "nesterov": "bool",
    "nmsIouThreshold": "int",
    "numberOfEpochs": "int",
    "numberOfWorkers": "int",
    "optimizer": "string",
    "randomSeed": "int",
    "splitRatio": "int",
    "stepLRGamma": "int",
    "stepLRStepSize": "int",
    "tileGridSize": "string",
    "tileOverlapRatio": "int",
    "tilePredictionsNmsThreshold": "int",
    "trainingBatchSize": "int",
    "validationBatchSize": "int",
    "validationIouThreshold": "int",
    "validationMetricType": "string",
    "warmupCosineLRCycles": "int",
    "warmupCosineLRWarmupEpochs": "int",
    "weightDecay": "int"
  },
  "primaryMetric": "MeanAveragePrecision",
  "searchSpace": [
    {
      "amsGradient": "string",
      "augmentations": "string",
      "beta1": "string",
      "beta2": "string",
      "boxDetectionsPerImage": "string",
      "boxScoreThreshold": "string",
      "distributed": "string",
      "earlyStopping": "string",
      "earlyStoppingDelay": "string",
      "earlyStoppingPatience": "string",
      "enableOnnxNormalization": "string",
      "evaluationFrequency": "string",
      "gradientAccumulationStep": "string",
      "imageSize": "string",
      "layersToFreeze": "string",
      "learningRate": "string",
      "learningRateScheduler": "string",
      "maxSize": "string",
      "minSize": "string",
      "modelName": "string",
      "modelSize": "string",
      "momentum": "string",
      "multiScale": "string",
      "nesterov": "string",
      "nmsIouThreshold": "string",
      "numberOfEpochs": "string",
      "numberOfWorkers": "string",
      "optimizer": "string",
      "randomSeed": "string",
      "splitRatio": "string",
      "stepLRGamma": "string",
      "stepLRStepSize": "string",
      "tileGridSize": "string",
      "tileOverlapRatio": "string",
      "tilePredictionsNmsThreshold": "string",
      "trainingBatchSize": "string",
      "validationBatchSize": "string",
      "validationIouThreshold": "string",
      "validationMetricType": "string",
      "warmupCosineLRCycles": "string",
      "warmupCosineLRWarmupEpochs": "string",
      "weightDecay": "string"
    }
  ],
  "sweepSettings": {
    "earlyTermination": {
      "delayEvaluation": "int",
      "evaluationInterval": "int",
      "policyType": "string"
      // For remaining properties, see EarlyTerminationPolicy objects
    },
    "limits": {
      "maxConcurrentTrials": "int",
      "maxTrials": "int"
    },
    "samplingAlgorithm": "string"
  }

針對 ImageObjectDetection,請使用:

  "taskType": "ImageObjectDetection",
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "validationDataSize": "int"
    }
  },
  "limitSettings": {
    "maxConcurrentTrials": "int",
    "maxTrials": "int",
    "timeout": "string"
  },
  "modelSettings": {
    "advancedSettings": "string",
    "amsGradient": "bool",
    "augmentations": "string",
    "beta1": "int",
    "beta2": "int",
    "boxDetectionsPerImage": "int",
    "boxScoreThreshold": "int",
    "checkpointDatasetId": "string",
    "checkpointFilename": "string",
    "checkpointFrequency": "int",
    "checkpointRunId": "string",
    "distributed": "bool",
    "earlyStopping": "bool",
    "earlyStoppingDelay": "int",
    "earlyStoppingPatience": "int",
    "enableOnnxNormalization": "bool",
    "evaluationFrequency": "int",
    "gradientAccumulationStep": "int",
    "imageSize": "int",
    "layersToFreeze": "int",
    "learningRate": "int",
    "learningRateScheduler": "string",
    "maxSize": "int",
    "minSize": "int",
    "modelName": "string",
    "modelSize": "string",
    "momentum": "int",
    "multiScale": "bool",
    "nesterov": "bool",
    "nmsIouThreshold": "int",
    "numberOfEpochs": "int",
    "numberOfWorkers": "int",
    "optimizer": "string",
    "randomSeed": "int",
    "splitRatio": "int",
    "stepLRGamma": "int",
    "stepLRStepSize": "int",
    "tileGridSize": "string",
    "tileOverlapRatio": "int",
    "tilePredictionsNmsThreshold": "int",
    "trainingBatchSize": "int",
    "validationBatchSize": "int",
    "validationIouThreshold": "int",
    "validationMetricType": "string",
    "warmupCosineLRCycles": "int",
    "warmupCosineLRWarmupEpochs": "int",
    "weightDecay": "int"
  },
  "primaryMetric": "MeanAveragePrecision",
  "searchSpace": [
    {
      "amsGradient": "string",
      "augmentations": "string",
      "beta1": "string",
      "beta2": "string",
      "boxDetectionsPerImage": "string",
      "boxScoreThreshold": "string",
      "distributed": "string",
      "earlyStopping": "string",
      "earlyStoppingDelay": "string",
      "earlyStoppingPatience": "string",
      "enableOnnxNormalization": "string",
      "evaluationFrequency": "string",
      "gradientAccumulationStep": "string",
      "imageSize": "string",
      "layersToFreeze": "string",
      "learningRate": "string",
      "learningRateScheduler": "string",
      "maxSize": "string",
      "minSize": "string",
      "modelName": "string",
      "modelSize": "string",
      "momentum": "string",
      "multiScale": "string",
      "nesterov": "string",
      "nmsIouThreshold": "string",
      "numberOfEpochs": "string",
      "numberOfWorkers": "string",
      "optimizer": "string",
      "randomSeed": "string",
      "splitRatio": "string",
      "stepLRGamma": "string",
      "stepLRStepSize": "string",
      "tileGridSize": "string",
      "tileOverlapRatio": "string",
      "tilePredictionsNmsThreshold": "string",
      "trainingBatchSize": "string",
      "validationBatchSize": "string",
      "validationIouThreshold": "string",
      "validationMetricType": "string",
      "warmupCosineLRCycles": "string",
      "warmupCosineLRWarmupEpochs": "string",
      "weightDecay": "string"
    }
  ],
  "sweepSettings": {
    "earlyTermination": {
      "delayEvaluation": "int",
      "evaluationInterval": "int",
      "policyType": "string"
      // For remaining properties, see EarlyTerminationPolicy objects
    },
    "limits": {
      "maxConcurrentTrials": "int",
      "maxTrials": "int"
    },
    "samplingAlgorithm": "string"
  }

針對 回歸,請使用:

  "taskType": "Regression",
  "allowedModels": [ "string" ],
  "blockedModels": [ "string" ],
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "cvSplitColumnNames": [ "string" ],
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "nCrossValidations": {
        "mode": "string"
        // For remaining properties, see NCrossValidations objects
      },
      "validationDataSize": "int"
    },
    "weightColumnName": "string"
  },
  "featurizationSettings": {
    "blockedTransformers": [ "string" ],
    "columnNameAndTypes": {
      "{customized property}": "string"
    },
    "datasetLanguage": "string",
    "dropColumns": [ "string" ],
    "enableDnnFeaturization": "bool",
    "mode": "string",
    "transformerParams": {
      "{customized property}": [
        {
          "fields": [ "string" ],
          "parameters": {}
        }
      ]
    }
  },
  "limitSettings": {
    "enableEarlyTermination": "bool",
    "exitScore": "int",
    "maxConcurrentTrials": "int",
    "maxCoresPerTrial": "int",
    "maxTrials": "int",
    "timeout": "string",
    "trialTimeout": "string"
  },
  "primaryMetric": "string",
  "trainingSettings": {
    "enableDnnTraining": "bool",
    "enableModelExplainability": "bool",
    "enableOnnxCompatibleModels": "bool",
    "enableStackEnsemble": "bool",
    "enableVoteEnsemble": "bool",
    "ensembleModelDownloadTimeout": "string",
    "stackEnsembleSettings": {
      "stackMetaLearnerKWargs": {},
      "stackMetaLearnerTrainPercentage": "int",
      "stackMetaLearnerType": "string"
    }
  }

針對 TextClassification,請使用:

  "taskType": "TextClassification",
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "validationDataSize": "int"
    }
  },
  "featurizationSettings": {
    "datasetLanguage": "string"
  },
  "limitSettings": {
    "maxConcurrentTrials": "int",
    "maxTrials": "int",
    "timeout": "string"
  },
  "primaryMetric": "string"

針對 TextClassificationMultilabel,請使用:

  "taskType": "TextClassificationMultilabel",
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "validationDataSize": "int"
    }
  },
  "featurizationSettings": {
    "datasetLanguage": "string"
  },
  "limitSettings": {
    "maxConcurrentTrials": "int",
    "maxTrials": "int",
    "timeout": "string"
  }

針對 TextNER,請使用:

  "taskType": "TextNER",
  "dataSettings": {
    "targetColumnName": "string",
    "testData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "testDataSize": "int"
    },
    "trainingData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      }
    },
    "validationData": {
      "data": {
        "description": "string",
        "jobInputType": "string",
        "mode": "string",
        "uri": "string"
      },
      "validationDataSize": "int"
    }
  },
  "featurizationSettings": {
    "datasetLanguage": "string"
  },
  "limitSettings": {
    "maxConcurrentTrials": "int",
    "maxTrials": "int",
    "timeout": "string"
  }

NCrossValidations 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  "mode": "Auto"

針對 [自定義],請使用:

  "mode": "Custom",
  "value": "int"

ForecastHorizon 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  "mode": "Auto"

針對 [自定義],請使用:

  "mode": "Custom",
  "value": "int"

季節性物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  "mode": "Auto"

針對 [自定義],請使用:

  "mode": "Custom",
  "value": "int"

TargetLags 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  "mode": "Auto"

針對 [自定義],請使用:

  "mode": "Custom",
  "values": [ "int" ]

TargetRollingWindowSize 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  "mode": "Auto"

針對 [自定義],請使用:

  "mode": "Custom",
  "value": "int"

EarlyTerminationPolicy 物件

設定 policyType 屬性以指定物件的類型。

針對 Bandit,請使用:

  "policyType": "Bandit",
  "slackAmount": "int",
  "slackFactor": "int"

針對 MedianStopping,請使用:

  "policyType": "MedianStopping"

針對 TruncationSelection,請使用:

  "policyType": "TruncationSelection",
  "truncationPercentage": "int"

DistributionConfiguration 物件

設定 distributionType 屬性以指定物件的類型。

針對 Mpi,請使用:

  "distributionType": "Mpi",
  "processCountPerInstance": "int"

針對 PyTorch,請使用:

  "distributionType": "PyTorch",
  "processCountPerInstance": "int"

針對 TensorFlow,請使用:

  "distributionType": "TensorFlow",
  "parameterServerCount": "int",
  "workerCount": "int"

JobInput 物件

設定 jobInputType 屬性以指定物件的類型。

針對 CustomModel,請使用:

  "jobInputType": "CustomModel",
  "mode": "string",
  "uri": "string"

針對 常值,請使用:

  "jobInputType": "Literal",
  "value": "string"

針對 MLFlowModel,請使用:

  "jobInputType": "MLFlowModel",
  "mode": "string",
  "uri": "string"

針對 MLTable,請使用:

  "jobInputType": "MLTable",
  "mode": "string",
  "uri": "string"

針對 TritonModel,請使用:

  "jobInputType": "TritonModel",
  "mode": "string",
  "uri": "string"

針對 UriFile,請使用:

  "jobInputType": "UriFile",
  "mode": "string",
  "uri": "string"

針對 UriFolder,請使用:

  "jobInputType": "UriFolder",
  "mode": "string",
  "uri": "string"

SamplingAlgorithm 物件

設定 samplingAlgorithmType 屬性以指定物件的類型。

針對 貝氏,請使用:

  "samplingAlgorithmType": "Bayesian"

針對 Grid,請使用:

  "samplingAlgorithmType": "Grid"

針對 隨機,請使用:

  "samplingAlgorithmType": "Random",
  "rule": "string",
  "seed": "int"

屬性值

workspaces/jobs

名稱 描述
類型 資源類型 'Microsoft.MachineLearningServices/workspaces/jobs'
apiVersion 資源 API 版本 '2022-02-01-preview'
NAME 資源名稱

請參閱如何在 JSON ARM 樣本中設定子資源的名稱和類型。
字串 (必要)
properties [必要]實體的其他屬性。 JobBaseDetails (必要)

JobBaseDetails

名稱 描述
computeId 計算資源的 ARM 資源識別碼。 字串
description 資產描述文字。 字串
displayName 工作的顯示名稱。 字串
experimentName 作業所屬的實驗名稱。 如果未設定,作業會放在「預設」實驗中。 字串
身分識別 身分識別設定。 如果設定,這應該是 AmlToken、ManagedIdentity、UserIdentity 或 null 的其中一個。
如果為 null,則預設為 AmlToken。
IdentityConfiguration
isArchived 資產是否已封存? bool
properties 資產屬性字典。 ResourceBaseProperties
schedule 作業的排程定義。
如果未提供排程,作業會在提交后立即執行一次。
ScheduleBase
服務 JobEndpoints 的清單。
針對本機作業,作業端點會有 FileStreamObject 的端點值。
JobBaseServices
tags 標記字典。 標記可以新增、移除和更新。 物件 (object)
jobType 設定物件類型 AutoML
命令
管線
需要掃掠 ()

IdentityConfiguration

名稱 描述
identityType 設定物件類型 AMLToken
受控
必要) 的 UserIdentity (

AmlToken

名稱 描述
identityType [必要]指定身分識別架構的類型。 必要的 『AMLToken』 ()

ManagedIdentity

名稱 描述
identityType [必要]指定身分識別架構的類型。 「受控」 (必要)
clientId 依用戶端識別元指定使用者指派的身分識別。 針對系統指派,請勿設定此欄位。 字串

約束:
最小長度 = 36
最大長度 = 36
模式 = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$
objectId 依物件識別元指定使用者指派的身分識別。 針對系統指派,請勿設定此欄位。 字串

約束:
最小長度 = 36
最大長度 = 36
模式 = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$
resourceId 依 ARM 資源識別碼指定使用者指派的身分識別。 針對系統指派,請勿設定此欄位。 字串

UserIdentity

名稱 描述
identityType [必要]指定身分識別架構的類型。 'UserIdentity' (必要)

ResourceBaseProperties

名稱 描述
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串

ScheduleBase

名稱 描述
EndTime 以 ISO 8601 格式指定排程的結束時間。
如果沒有,排程將會無限期執行
字串
scheduleStatus 指定排程的狀態 'Disabled'
'Enabled'
startTime 以 ISO 8601 格式指定排程的開始時間。 字串
timeZone 指定排程執行所在的時區。
TimeZone應遵循 Windows 時區格式。
字串
scheduleType 設定物件類型 Cron
需要迴圈 ()

CronSchedule

名稱 描述
scheduleType [必要]指定排程類型 需要 『Cron』 ()
expression [必要]指定排程的cron運算式。
表達式應遵循 NCronTab 格式。
需要字串 ()

約束:
模式 = [a-zA-Z0-9_]

RecurrenceSchedule

名稱 描述
scheduleType [必要]指定排程類型 必要) 「週期」 (
frequency [必要]指定要觸發排程的頻率 'Day'
'Hour'
'Minute'
'Month'
需要 『Week』 ()
interval [必要]指定與頻率搭配的排程間隔 需要 int ()
模式 指定週期排程模式 RecurrencePattern

RecurrencePattern

名稱 描述
小時 [必要]週期排程模式的時數清單 int[] (必要)
分鐘 [必要]週期排程模式的分鐘數清單 int[] (必要)
平日 週期排程模式的工作日清單 包含任何項目的字串數組:
'Friday'
'Monday'
'Saturday'
'Sunday'
'Thursday'
'Tuesday'
'Wednesday'

JobBaseServices

名稱 描述
{自定義屬性} JobService

JobService

名稱 描述
端點 端點的 URL。 字串
jobServiceType 端點類型。 string
連接埠 端點的埠。 int
properties 在端點上設定的其他屬性。 JobServiceProperties

JobServiceProperties

名稱 描述
{自定義屬性} 字串

AutoMLJob

名稱 描述
jobType [必要]指定作業的類型。 必要的 『AutoML』 ()
environmentId 作業環境規格的 ARM 資源識別碼。
這是選擇性值,如果未提供,AutoML 會在執行作業時,將此預設為生產 AutoML 策劃的環境版本。
字串
environmentVariables 作業中包含的環境變數。 AutoMLJobEnvironmentVariables
outputs 對應作業中使用的輸出數據系結。 AutoMLJobOutputs
resources 作業的計算資源組態。 ResourceConfiguration
taskDetails [必要]這代表可以是其中一個數據表/NLP/Image 的案例 需要 AutoMLVertical ()

AutoMLJobEnvironmentVariables

名稱 描述
{自定義屬性} 字串

AutoMLJobOutputs

名稱 描述
{自定義屬性} JobOutput

JobOutput

名稱 描述
description 輸出的描述。 字串
jobOutputType 設定物件類型 CustomModel
MLFlowModel
MLTable
TritonModel
UriFile
需要 uriFolder ()

CustomModelJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 'CustomModel' (必要)
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

MLFlowModelJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 必要) 的 'MLFlowModel' (
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

MLTableJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 需要 『MLTable』 ()
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

TritonModelJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 'TritonModel' (必要)
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

UriFileJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 需要 『UriFile』 ()
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

UriFolderJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 'UriFolder' (必要)
mode 輸出資產傳遞模式。 'ReadWriteMount'
'Upload'
uri 輸出資產 URI。 字串

ResourceConfiguration

名稱 描述
instanceCount 計算目標所使用的選用執行個體或節點數目。 int
instanceType 計算目標所支持的選擇性 VM 類型。 字串
properties 其他屬性包。 ResourceConfigurationProperties

ResourceConfigurationProperties

名稱 描述
{自定義屬性}

AutoMLVertical

名稱 描述
logVerbosity 作業的記錄詳細資訊。 'Critical'
'Debug'
'Error'
'Info'
'NotSet'
'Warning'
taskType 設定物件類型 分類
預測
ImageClassification
ImageClassificationMultilabel
ImageInstanceSegmentation
ImageObjectDetection
迴歸
TextClassification
TextClassificationMultilabel
需要 TextNER ()

分類

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 「分類」 (必要)
allowedModels 分類工作的允許模型。 包含任何項目的字串數組:
'BernoulliNaive設定'
'DecisionTree'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LightGBM'
'LinearSVM'
'LogisticRegression'
'MultinomialNaiveBayes'
'RandomForest'
'SGD'
'SVM'
'XGBoostClassifier'
blockedModels 分類工作的封鎖模型。 包含任何項目的字串數組:
'BernoulliNaive設定'
'DecisionTree'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LightGBM'
'LinearSVM'
'LogisticRegression'
'MultinomialNaiveBayes'
'RandomForest'
'SGD'
'SVM'
'XGBoostClassifier'
dataSettings AutoMLJob 的數據輸入。 TableVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 TableVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 TableVerticalLimitSettings
primaryMetric 工作的主要計量。 'AUCWeighted'
「精確度」
'AveragePrecisionScoreWeighted'
'NormMacroRecall'
'PrecisionScoreWeighted'
trainingSettings AutoML 作業定型階段的輸入。 TrainingSettings

TableVerticalDataSettings

名稱 描述
targetColumnName [必要]目標數據行名稱:這是預測值數據行。
也稱為分類工作內容中的標籤數據行名稱。
需要字串 ()

約束:
模式 = [a-zA-Z0-9_]
testData 測試數據輸入。 TestDataSettings
trainingData [必要]定型數據輸入。 TrainingDataSettings (必要)
validationData 驗證數據輸入。 TableVerticalValidationDataSettings
weightColumnName 範例加權數據行的名稱。 自動化 ML 支援加權數據行做為輸入,導致數據中的數據列向上或向下加權。 字串

TestDataSettings

名稱 描述
資料 測試數據 MLTable。 MLTableJobInput
testDataSize 測試數據集的分數,必須保留以供驗證之用。
(0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

MLTableJobInput

名稱 描述
description 輸入的描述。 字串
jobInputType [必要]指定作業的類型。 'CustomModel'
'Literal'
'MLFlowModel'
'MLTable'
'TritonModel'
'UriFile'
'UriFolder' (必要)
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

TrainingDataSettings

名稱 描述
資料 [必要]訓練數據MLTable。 MLTableJobInput (必要)

TableVerticalValidationDataSettings

名稱 描述
cvSplitColumnNames 要用於 CVSplit 資料的數據行。 string[]
資料 驗證數據 MLTable。 MLTableJobInput
nCrossValidations 要套用至定型數據集的交叉驗證折疊數目
未提供驗證數據集時。
NCrossValidations
validationDataSize 必須針對驗證目的將定型數據集的分數。
介於 (0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

NCrossValidations

名稱 描述
mode 設定物件類型 Auto
自訂 (必要)

AutoNCrossValidations

名稱 描述
mode [必要]判斷 N-Cross 驗證的模式。 需要 『Auto』 ()

CustomNCrossValidations

名稱 描述
mode [必要]判斷 N-Cross 驗證的模式。 需要 『Custom』 ()
[必要]N-Cross 驗證值。 int (必要)

TableVerticalFeaturizationSettings

名稱 描述
blockedTransformers 這些轉換器不得用於特徵化。 string[]
columnNameAndTypes 數據行名稱的字典及其類型 (int、float、string、datetime 等) 。 TableVerticalFeaturizationSettingsColumnNameAndTypes
datasetLanguage 數據集語言,適用於文字數據。 字串
dropColumns 特徵化期間要從數據卸除的數據行。 string[]
enableDnnFeaturization 判斷是否要使用 Dnn 型特徵化程式進行數據特徵化。 bool
mode 特徵化模式 - 使用者可以保留預設的「自動」模式,而 AutoML 會負責在特徵化階段中轉換數據的必要轉換。
如果選取 [關閉],則不會完成特徵化。
如果選取 [自定義],則使用者可以指定其他輸入,以自定義特徵化完成的方式。
'Auto'
'Custom'
'Off'
transformerParams 用戶可以指定要搭配其套用的數據行和轉換程式建構函式參數使用的其他轉換器。 TableVerticalFeaturizationSettingsTransformerParams

TableVerticalFeaturizationSettingsColumnNameAndTypes

名稱 描述
{自定義屬性} 字串

TableVerticalFeaturizationSettingsTransformerParams

名稱 描述
{自定義屬性} ColumnTransformer[]

ColumnTransformer

名稱 描述
fields 要套用轉換器邏輯的欄位。 string[]
參數 要傳遞至轉換器的不同屬性。
預期的輸入是 JSON 格式的索引鍵、值組字典。

TableVerticalLimitSettings

名稱 描述
enableEarlyTermination 啟用提早終止,判斷如果最後 20 個反覆項目沒有分數改善,AutoMLJob 是否提早終止。 bool
exitScore AutoML 作業的結束分數。 int
maxConcurrentTrials 並行反覆專案上限。 int
maxCoresPerTrial 每次反覆運算的核心數上限。 int
maxTrials 反覆運算次數。 int
timeout AutoML 作業逾時。 字串
trialTimeout 反覆專案逾時。 字串

TrainingSettings

名稱 描述
enableDnnTraining 啟用 DNN 模型的建議。 bool
enableModelExplainability 在最佳模型上開啟可解釋性的旗標。 bool
enableOnnxCompatibleModels 啟用 onnx 相容模型的旗標。 bool
enableStackEnsemble 啟用堆疊Ensemble 執行。 bool
enableVoteEnsemble 啟用投票Ensemble 執行。 bool
ensembleModelDownloadTimeout 在 VotingEnsemble 和 StackEnsemble 模型產生期間,會下載先前子回合中的多個配適模型。
如果需要更多時間,請將此參數設定為高於 300 秒的值。
字串
stackEnsembleSettings 堆疊Ensemble 執行的堆疊Ensemble 設定。 StackEnsembleSettings

StackEnsembleSettings

名稱 描述
stackMetaLearnerKWargs 要傳遞至中繼學習模組初始化表達式的選擇性參數。
stackMetaLearnerTrainPercentage 指定在選擇訓練和驗證類型的訓練) 保留給訓練中繼學習工具時,訓練集 (的比例。 預設值為 0.2。 int
stackMetaLearnerType 中繼學習模組是針對個別異質模型的輸出定型的模型。 'ElasticNet'
'ElasticNetCV'
'LightGBMClassifier'
'LightGBMRegressor'
'LinearRegression'
'LogisticRegression'
'LogisticRegressionCV'
'None'

預測

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 「預測」 (必要)
allowedModels 用於預測工作的允許模型。 包含任何項目的字串數組:
'Arimax'
'AutoArima'
'Average'
'DecisionTree'
'ElasticNet'
'ExponentialSmoothing'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LassoLars'
'LightGBM'
'Naive'
'以新月'
'RandomForest'
'SGD'
'SeasonalAverage'
'SeasonalNaive'
'TCNForecaster'
'XGBoostRegressor'
blockedModels 用於預測工作的封鎖模型。 包含任何項目的字串數組:
'Arimax'
'AutoArima'
'Average'
'DecisionTree'
'ElasticNet'
'ExponentialSmoothing'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LassoLars'
'LightGBM'
'Naive'
'以新月'
'RandomForest'
'SGD'
'SeasonalAverage'
'SeasonalNaive'
'TCNForecaster'
'XGBoostRegressor'
dataSettings AutoMLJob 的數據輸入。 TableVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 TableVerticalFeaturizationSettings
forecastingSettings 預測工作特定輸入。 ForecastingSettings
limitSettings AutoMLJob 的執行條件約束。 TableVerticalLimitSettings
primaryMetric 預測工作的主要計量。 'NormalizedMeanAbsoluteError'
'NormalizedRootMeanSquaredError'
'R2Score'
'SpearmanCorrelation'
trainingSettings AutoML 作業定型階段的輸入。 TrainingSettings

ForecastingSettings

名稱 描述
countryOrRegionForHolidays 用於預測工作的假日國家或地區。
這些應該是 ISO 3166 兩個字母的國家/地區代碼,例如 'US' 或 'GB'。
字串
cvStepSize 一個 CV 折疊的原點時間與下一個折疊之間的期間數。 For
例如,如果 CVStepSize = 3 表示每日數據,則每個折疊的源時間將會是
相隔三天。
int
featureLags 旗標,用於產生具有 'auto' 或 null 之數值特徵的延隔時間。 'Auto'
'None'
forecastHorizon 時間序列頻率單位所需的最大預測層級。 ForecastHorizon
frequency 預測時,此參數代表需要預測的期間,例如每日、每周、每年等等。預測頻率預設為數據集頻率。 字串
季節性 將時間序列季節性設定為數列頻率的整數倍數。
如果季節性設定為 『auto』,則會推斷它。
季節性
shortSeriesHandlingConfig 參數,定義 AutoML 是否應該處理簡短的時間序列。 'Auto'
'Drop'
'None'
'Pad'
targetAggregateFunction 用來匯總時間序列目標數據行以符合使用者指定頻率的函式。
如果 TargetAggregateFunction 設定為不是 『None』,但未設定 freq 參數,就會引發錯誤。 可能的目標匯總函數包括:「sum」、“ max”
'Max'
'Mean'
'Min'
'None'
'Sum'
targetLags 要從目標數據行延遲的過去期間數。 TargetLags
targetRollingWindowSize 過去期間數,用來建立目標數據行的滾動視窗平均值。 TargetRollingWindowSize
timeColumnName 時間數據行的名稱。 當預測指定用於建置時間序列並推斷其頻率的輸入數據中的 datetime 數據行時,需要此參數。 字串
timeSeriesIdColumnNames 用來分組時間範圍的數據行名稱。 它可以用來建立多個數列。
如果未定義粒紋,則會假設資料集為一個時間序列。 此參數會與工作類型預測搭配使用。
string[]
useStl 設定時間序列目標數據行的 STL 分解。 'None'
'Season'
'SeasonTrend'

ForecastHorizon

名稱 描述
mode 設定物件類型 Auto
自訂 (必要)

AutoForecastHorizon

名稱 描述
mode [必要]設定預測水準值選取模式。 需要 『Auto』 ()

CustomForecastHorizon

名稱 描述
mode [必要]設定預測水準值選取模式。 需要 『Custom』 ()
[必要]預測水準值。 int (必要)

季節性

名稱 描述
mode 設定物件類型 Auto
自訂 (必要)

AutoSeasonality

名稱 描述
mode [必要]季節性模式。 需要 『Auto』 ()

CustomSeasonality

名稱 描述
mode [必要]季節性模式。 需要 『Custom』 ()
[必要]季節性值。 int (必要)

TargetLags

名稱 描述
mode 設定物件類型 Auto
自訂 (必要)

AutoTargetLags

名稱 描述
mode [必要]設定目標延遲模式 - 自動/自定義 需要 『Auto』 ()

CustomTargetLags

名稱 描述
mode [必要]設定目標延隔模式 - 自動/自定義 必要) 'Custom' (
[必要]設定目標延隔值。 int[] (必要)

TargetRollingWindowSize

名稱 描述
mode 設定物件類型 Auto
需要自訂 ()

AutoTargetRollingWindowSize

名稱 描述
mode [必要]TargetRollingWindowSiz 偵測模式。 需要 『Auto』 ()

CustomTargetRollingWindowSize

名稱 描述
mode [必要]TargetRollingWindowSiz 偵測模式。 必要) 'Custom' (
[必要]TargetRollingWindowSize 值。 需要 int ()

ImageClassification

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'ImageClassification' (必要)
dataSettings [必要]已註冊表格式數據集標識碼的集合,以及定型和驗證模型所需的其他數據設定。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsClassification
primaryMetric 要針對這項工作優化的主要計量。 'AUCWeighted'
「精確度」
'AveragePrecisionScoreWeighted'
'NormMacroRecall'
'PrecisionScoreWeighted'
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsClassification[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

ImageVerticalDataSettings

名稱 描述
targetColumnName [必要]目標數據行名稱:這是預測值數據行。
也稱為分類工作內容中的標籤數據行名稱。
需要字串 ()

約束:
模式 = [a-zA-Z0-9_]
testData 測試數據輸入。 TestDataSettings
trainingData [必要]定型數據輸入。 TrainingDataSettings (必要)
validationData 驗證數據集的設定。 ImageVerticalValidationDataSettings

ImageVerticalValidationDataSettings

名稱 描述
資料 驗證數據 MLTable。 MLTableJobInput
validationDataSize 需要為驗證目的保留的定型數據集分數。
(0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

ImageLimitSettings

名稱 描述
maxConcurrentTrials 並行 AutoML 反覆項目的數目上限。 int
maxTrials AutoML 反覆項目的數目上限。 int
timeout AutoML 作業逾時。 字串

ImageModelSettingsClassification

名稱 描述
advancedSettings 進階案例的設定。 字串
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 bool
擴增 使用擴增的設定。 字串
beta1 當優化器為 'adam' 或 'adamw' 時,'beta1' 的值。 必須是範圍 [0, 1] 內的浮點數。 int
beta2 當優化器為 'adam' 或 'adamw' 時,'beta2' 的值。 必須是範圍 [0, 1] 內的浮點數。 int
checkpointDatasetId 預先定型檢查點的 FileDataset 識別碼, (累加式定型的) 。
請務必傳遞 CheckpointFilename 以及 CheckpointDatasetId。
字串
checkpointFilename FileDataset 中用於累加式定型的預先定型檢查點檔案名稱。
請務必傳遞 CheckpointDatasetId 以及 CheckpointFilename。
字串
checkpointFrequency 儲存模型檢查點的頻率。 必須是正整數。 int
checkpointRunId 先前執行之標識碼,其具有累加式定型的預先定型檢查點。 字串
分散式 是否要使用分散式定型。 bool
earlyStopping 在定型期間啟用提早停止邏輯。 bool
earlyStoppingDelay 在主要計量改進之前等待的最小 Epoch 或驗證評估數目
會追蹤提早停止。 必須是正整數。
int
earlyStoppingPatience 在之前沒有主要計量改善的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
int
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 bool
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 int
gradientAccumulationStep 漸層累積表示執行未設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
int
layersToFreeze 要凍結模型的層數。 必須是正整數。
例如,傳遞 2 作為 'seresnext' 值表示
凍結 layer0 和 layer1。 如需所支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
int
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 int
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 'None'
'Step'
'WarmupCosine'
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請造訪官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
動量 優化工具為 『sgd』 時的動向值。 必須是範圍 [0, 1] 內的浮點數。 int
nesterov 當優化器為 'sgd' 時,請啟用 nesterov。 bool
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 int
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 int
optimizer 最佳化工具的類型。 'Adam'
'Adamw'
'None'
'Sgd'
randomSeed 使用決定性定型時要使用的隨機種子。 int
splitRatio 如果未定義驗證數據,這會指定分割的分割比率
將數據定型成隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
int
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 int
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 int
trainingBatchSize 定型批次大小。 必須是正整數。 int
trainingCropSize 影像裁剪大小,為定型數據集的類神經網路輸入。 必須是正整數。 int
validationBatchSize 驗證批次大小。 必須是正整數。 int
validationCropSize 輸入至驗證數據集之類神經網路的影像裁剪大小。 必須是正整數。 int
validationResizeSize 在針對驗證資料集進行裁切之前,要調整大小的影像大小。 必須是正整數。 int
warmupCosineLRCycles 當學習速率排程器為 『warmup_cosine』 時,餘弦迴圈的值。 必須是範圍 [0, 1] 內的浮點數。 int
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』時,熱身 Epoch 的值。 必須是正整數。 int
weightDecay 優化器為 'sgd'、'adam' 或 'adamw' 時的權數衰減值。 必須是範圍[0, 1] 中的浮點數。 int
weightedLoss 加權損失。 接受的值是 0,不會遺失加權。
1 用於 sqrt. (class_weights) 加權損失。 2 表示具有 class_weights 的加權遺失。 必須是 0 或是 1 或 2。
int

ImageModelDistributionSettingsClassification

名稱 描述
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 字串
擴增 使用擴增的設定。 字串
beta1 優化器為 'adam' 或 'adamw' 時的 'beta1' 值。 必須是範圍 [0, 1] 內的浮點數。 字串
beta2 優化器為 'adam' 或 'adamw' 時的 'beta2' 值。 必須是範圍 [0, 1] 內的浮點數。 字串
分散式 是否要使用分散式程式定型。 字串
earlyStopping 在定型期間啟用提早停止邏輯。 字串
earlyStoppingDelay 在主要計量改進之前等候的最小 Epoch 或驗證評估數目
會追蹤早期停止。 必須是正整數。
字串
earlyStoppingPatience 在之前沒有主要計量改善的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
字串
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 字串
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 字串
gradientAccumulationStep 漸層累積表示執行未設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
字串
layersToFreeze 要凍結模型的層數。 必須是正整數。
例如,傳遞 2 作為 'seresnext' 值表示
凍結 layer0 和 layer1。 如需所支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 字串
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 字串
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請造訪官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
動量 優化工具為 『sgd』 時的動向值。 必須是範圍 [0, 1] 內的浮點數。 字串
nesterov 當優化器為 『sgd』 時啟用 nesterov。 字串
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 字串
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 字串
optimizer 最佳化工具的類型。 必須是 'sgd'、'adam' 或 'adamw'。 字串
randomSeed 使用決定性定型時要使用的隨機種子。 字串
splitRatio 如果未定義驗證數據,這會指定分割的分割比例
將數據定型為隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
字串
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 字串
trainingBatchSize 定型批次大小。 必須是正整數。 字串
trainingCropSize 影像裁剪大小,為定型數據集輸入類神經網路。 必須是正整數。 字串
validationBatchSize 驗證批次大小。 必須是正整數。 字串
validationCropSize 影像裁剪大小,這是驗證數據集之類神經網路的輸入。 必須是正整數。 字串
validationResizeSize 在針對驗證資料集進行裁切之前,要調整大小的影像大小。 必須是正整數。 字串
warmupCosineLRCycles 學習速率排程器為「warmup_cosine」時的餘弦週期值。 必須是範圍 [0, 1] 內的浮點數。 字串
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』 時,暖 Epoch 的值。 必須是正整數。 字串
weightDecay 當優化器為 'sgd'、'adam' 或 'adamw' 時,加權衰減的值。 必須是 range[0, 1] 中的浮點數。 字串
weightedLoss 加權損失。 接受的值為 0,不會遺失加權。
1 用於 sqrt. (class_weights) 的加權損失。 2 表示具有 class_weights 的加權遺失。 必須是 0 或是 1 或 2。
字串

ImageSweepSettings

名稱 描述
earlyTermination 早期終止原則的類型。 EarlyTerminationPolicy
限制 [必要]限制模型掃掠和超參數掃掠的設定。 ImageSweepLimitSettings (必要)
samplingAlgorithm [必要]超參數取樣演算法的類型。 'Bayesian'
'Grid'
需要 'Random' ()

EarlyTerminationPolicy

名稱 描述
delayEvaluation 延遲第一次評估的間隔數目。 int
evaluationInterval 原則評估之間) 執行的間隔 (數目。 int
policyType 設定物件類型 Bandit
MedianStopping
截斷選取 必要 ()

BanditPolicy

名稱 描述
policyType [必要]原則設定的名稱 需要 『Bandit』 ()
slackAmount 最佳執行所允許的絕對距離。 int
slackFactor 允許與最佳執行距離的比例。 int

MedianStoppingPolicy

名稱 描述
policyType [必要]原則設定的名稱 必要 ('MedianStopping')

TruncationSelectionPolicy

名稱 描述
policyType [必要]原則設定的名稱 'TruncationSelection' (必要)
truncationPercentage 要在每個評估間隔取消的執行百分比。 int

ImageSweepLimitSettings

名稱 描述
maxConcurrentTrials 基礎掃掠作業的並行反覆項目數目上限。 int
maxTrials 基礎掃掠作業的反覆項目數目上限。 int

ImageClassificationMultilabel

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'ImageClassificationMultilabel' (必要)
dataSettings [必要]已註冊表格式數據集標識碼的集合,以及定型和驗證模型所需的其他數據設定。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsClassification
primaryMetric 要針對這項工作優化的主要計量。 'AUCWeighted'
'Accuracy'
'AveragePrecisionScoreWeighted'
'IOU'
'NormMacroRecall'
'PrecisionScoreWeighted'
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsClassification[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

ImageInstanceSegmentation

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'ImageInstanceSegmentation' (必要)
dataSettings [必要]註冊的表格式數據集標識碼和其他定型和驗證模型所需的數據設定集合。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsObjectDetection
primaryMetric 要針對這項工作優化的主要計量。 'MeanAveragePrecision'
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsObjectDetection[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

ImageModelSettingsObjectDetection

名稱 描述
advancedSettings 進階案例的設定。 字串
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 bool
擴增 使用擴增的設定。 字串
beta1 優化器為 'adam' 或 'adamw' 時的 'beta1' 值。 必須是範圍 [0, 1] 內的浮點數。 int
beta2 優化器為 'adam' 或 'adamw' 時的 'beta2' 值。 必須是範圍 [0, 1] 內的浮點數。 int
boxDetectionsPerImage 針對所有類別,每個影像的偵測數目上限。 必須是正整數。
注意:'yolov5' 演算法不支援此設定。
int
boxScoreThreshold 在推斷期間,只傳回分類分數大於的建議
BoxScoreThreshold。 必須是範圍[0, 1] 中的浮點數。
int
checkpointDatasetId 預先定型檢查點的 FileDataset 識別碼, (累加式定型的) 。
請務必傳遞 CheckpointFilename 以及 CheckpointDatasetId。
字串
checkpointFilename FileDataset 中用於累加式定型的預先定型檢查點檔案名稱。
請務必傳遞 CheckpointDatasetId 以及 CheckpointFilename。
字串
checkpointFrequency 儲存模型檢查點的頻率。 必須是正整數。 int
checkpointRunId 先前執行的標識碼,其具有累加定型的預先定型檢查點。 字串
分散式 是否要使用分散式定型。 bool
earlyStopping 在定型期間啟用提早停止邏輯。 bool
earlyStoppingDelay 在主要計量改進之前等候的最小 Epoch 或驗證評估數目
會追蹤早期停止。 必須是正整數。
int
earlyStoppingPatience 之前沒有主要計量改進的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
int
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 bool
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 int
gradientAccumulationStep 漸層累積表示在沒有的情況下執行已設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
int
imageSize 用於定型和驗證的影像大小。 必須是正整數。
注意:如果大小太大,定型回合可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
int
layersToFreeze 要凍結模型的圖層數目。 必須是正整數。
例如,傳遞 2 做為 'seresnext' 值表示
凍結第0層和第1層。 如需支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
int
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 int
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 'None'
'Step'
'WarmupCosine'
maxSize 將影像饋送至骨幹之前要重新調整的影像大小上限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
int
minSize 將影像饋送至骨幹之前要重新調整的影像大小下限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
int
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請流覽官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
modelSize 模型大小。 必須是 'small'、'medium'、'large' 或 'xlarge'。
注意:如果模型大小太大,則定型執行可能會進入 CUDA OOM.
注意:只有 'yolov5' 演算法才支援此設定。
'ExtraLarge'
'Large'
'Medium'
'None'
'Small'
動量 優化工具為 『sgd』 時的時刻值。 必須是範圍 [0, 1] 內的浮點數。 int
尺度 依 +/- 50% 的不同影像大小啟用多縮放映射。
注意:如果沒有足夠的 GPU 記憶體,則定型執行可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
bool
nesterov 當優化器為 'sgd' 時,請啟用 nesterov。 bool
nmsIouThreshold 在NMS後置處理中推斷期間使用的IOU臨界值。 必須是範圍 [0, 1] 內的浮點數。 int
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 int
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 int
optimizer 最佳化工具的類型。 'Adam'
'Adamw'
'None'
'Sgd'
randomSeed 使用決定性定型時要使用的隨機種子。 int
splitRatio 如果未定義驗證數據,這會指定分割的分割比率
將數據定型成隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
int
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 int
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 int
tileGridSize 要用於每個並排影像的格線大小。 注意:TileGridSize 不得為
無可啟用小型物件偵測邏輯。 字串,包含兩個整數,格式為一個。
注意:'yolov5' 演算法不支援此設定。
字串
tileOverlapRatio 每個維度中相鄰圖格之間的重疊比例。 必須是範圍 [0, 1) 的 float。
注意:'yolov5' 演算法不支援此設定。
int
tilePredictionsNmsThreshold 從圖格和影像合併預測時用來執行 NMS 的 IOU 閾值。
用於驗證/推斷。 必須是範圍 [0, 1] 中的浮點數。
注意:'yolov5' 演算法不支援此設定。
int
trainingBatchSize 定型批次大小。 必須是正整數。 int
validationBatchSize 驗證批次大小。 必須是正整數。 int
validationIouThreshold 計算驗證計量時要使用的IOU臨界值。 必須是範圍 [0, 1] 中的浮點數。 int
validationMetricType 要用於驗證計量的計量計算方法。 'Coco'
'CocoVoc'
'None'
'Voc'
warmupCosineLRCycles 學習速率排程器為「warmup_cosine」時的餘弦週期值。 必須是範圍 [0, 1] 內的浮點數。 int
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』 時,暖 Epoch 的值。 必須是正整數。 int
weightDecay 當優化器為 'sgd'、'adam' 或 'adamw' 時,加權衰減的值。 必須是 range[0, 1] 中的浮點數。 int

ImageModelDistributionSettingsObjectDetection

名稱 描述
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 字串
擴增 使用擴增的設定。 字串
beta1 當優化器為 'adam' 或 'adamw' 時,'beta1' 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
beta2 當優化器為 'adam' 或 'adamw' 時,'beta2' 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
boxDetectionsPerImage 針對所有類別,每個影像的偵測數目上限。 必須是正整數。
注意:'yolov5' 演算法不支援此設定。
字串
boxScoreThreshold 在推斷期間,只傳回分類分數大於的建議
BoxScoreThreshold。 必須是 range[0, 1] 中的浮點數。
字串
分散式 是否要使用散發器定型。 字串
earlyStopping 在定型期間啟用提早停止邏輯。 字串
earlyStoppingDelay 在主要計量改進之前等待的最小 Epoch 或驗證評估數目
會追蹤提早停止。 必須是正整數。
字串
earlyStoppingPatience 在之前沒有主要計量改善的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
字串
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 字串
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 字串
gradientAccumulationStep 漸層累積表示在沒有的情況下執行已設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
字串
imageSize 用於定型和驗證的影像大小。 必須是正整數。
注意:如果大小太大,定型回合可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
字串
layersToFreeze 要凍結模型的圖層數目。 必須是正整數。
例如,傳遞 2 做為 'seresnext' 值表示
凍結第0層和第1層。 如需支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 字串
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 字串
maxSize 將影像饋送至骨幹之前要重新調整的影像大小上限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
字串
minSize 將影像饋送至骨幹之前要重新調整的影像大小下限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
字串
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請流覽官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
modelSize 模型大小。 必須是 'small'、'medium'、'large' 或 'xlarge'。
注意:如果模型大小太大,則定型執行可能會進入 CUDA OOM.
注意:只有 'yolov5' 演算法才支援此設定。
字串
動量 優化工具為 『sgd』 時的時刻值。 必須是範圍 [0, 1] 內的浮點數。 字串
尺度 依 +/- 50% 的不同影像大小啟用多縮放映射。
注意:如果沒有足夠的 GPU 記憶體,則定型執行可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
字串
nesterov 當優化器為 'sgd' 時,請啟用 nesterov。 字串
nmsIouThreshold 在NMS後置處理中推斷期間使用的IOU臨界值。 必須是範圍 [0, 1] 中的 float。 字串
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 字串
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 字串
optimizer 最佳化工具的類型。 必須是 『sgd』、『adam』或 『adamw』。 字串
randomSeed 使用決定性定型時要使用的隨機種子。 字串
splitRatio 如果未定義驗證數據,這會指定分割的分割比率
將數據定型成隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
字串
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 字串
tileGridSize 要用於每個並排影像的格線大小。 注意:TileGridSize 不得為
無可啟用小型物件偵測邏輯。 包含兩個整數的字串,其格式為#。
注意:'yolov5' 演算法不支援此設定。
字串
tileOverlapRatio 每個維度中相鄰圖格之間的重疊比例。 範圍 [0, 1) 必須是 float。
注意:'yolov5' 演算法不支援此設定。
字串
tilePredictionsNmsThreshold 從圖格和影像合併預測時用來執行 NMS 的 IOU 閾值。
用於驗證/推斷。 必須是範圍 [0, 1] 中的 float。
注意:'yolov5' 演算法不支援此設定。
NMS:非最大隱藏
字串
trainingBatchSize 定型批次大小。 必須是正整數。 字串
validationBatchSize 驗證批次大小。 必須是正整數。 字串
validationIouThreshold 計算驗證計量時要使用的IOU臨界值。 必須是範圍 [0, 1] 中的 float。 字串
validationMetricType 要用於驗證計量的計量計算方法。 必須是 『none』、'coco'、'voc' 或 'coco_voc'。 字串
warmupCosineLRCycles 當學習速率排程器為 『warmup_cosine』 時,餘弦迴圈的值。 必須是範圍 [0, 1] 內的浮點數。 字串
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』時,熱身 Epoch 的值。 必須是正整數。 字串
weightDecay 優化器為 'sgd'、'adam' 或 'adamw' 時的權數衰減值。 必須是範圍[0, 1] 中的浮點數。 字串

ImageObjectDetection

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'ImageObjectDetection' (必要)
dataSettings [必要]註冊的表格式數據集標識碼和其他定型和驗證模型所需的數據設定集合。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsObjectDetection
primaryMetric 要針對這項工作優化的主要計量。 'MeanAveragePrecision'
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsObjectDetection[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

迴歸

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 需要 「回歸」 ()
allowedModels 回歸工作的允許模型。 包含任何項目的字串數組:
'DecisionTree'
'ElasticNet'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LassoLars'
'LightGBM'
'RandomForest'
'SGD'
'XGBoostRegressor'
blockedModels 回歸工作的封鎖模型。 包含任何項目的字串數組:
'DecisionTree'
'ElasticNet'
'ExtremeRandomTrees'
'GradientBoosting'
'KNN'
'LassoLars'
'LightGBM'
'RandomForest'
'SGD'
'XGBoostRegressor'
dataSettings AutoMLJob 的數據輸入。 TableVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 TableVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 TableVerticalLimitSettings
primaryMetric 回歸工作的主要計量。 'NormalizedMeanAbsoluteError'
'NormalizedRootMeanSquaredError'
'R2Score'
'SpearmanCorrelation'
trainingSettings AutoML 作業定型階段的輸入。 TrainingSettings

TextClassification

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'TextClassification' (必要)
dataSettings AutoMLJob 的數據輸入。 NlpVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 NlpVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 NlpVerticalLimitSettings
primaryMetric Text-Classification 工作的主要計量。 'AUCWeighted'
「精確度」
'AveragePrecisionScoreWeighted'
'NormMacroRecall'
'PrecisionScoreWeighted'

NlpVerticalDataSettings

名稱 描述
targetColumnName [必要]目標數據行名稱:這是預測值數據行。
也稱為分類工作內容中的標籤數據行名稱。
需要字串 ()

約束:
模式 = [a-zA-Z0-9_]
testData 測試數據輸入。 TestDataSettings
trainingData [必要]定型數據輸入。 TrainingDataSettings (必要)
validationData 驗證數據輸入。 NlpVerticalValidationDataSettings

NlpVerticalValidationDataSettings

名稱 描述
資料 驗證數據 MLTable。 MLTableJobInput
validationDataSize 需要為驗證目的保留的定型數據集分數。
(0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

NlpVerticalFeaturizationSettings

名稱 描述
datasetLanguage 數據集語言,適用於文字數據。 字串

NlpVerticalLimitSettings

名稱 描述
maxConcurrentTrials 並行 AutoML 反覆專案上限。 int
maxTrials AutoML 反覆項目的數目。 int
timeout AutoML 作業逾時。 字串

TextClassificationMultilabel

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 'TextClassificationMultilabel' (必要)
dataSettings AutoMLJob 的數據輸入。 NlpVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 NlpVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 NlpVerticalLimitSettings

TextNer

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 需要 『TextNER』 ()
dataSettings AutoMLJob 的數據輸入。 NlpVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 NlpVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 NlpVerticalLimitSettings

CommandJob

名稱 描述
jobType [必要]指定作業的類型。 需要 『Command』 ()
codeId 程式代碼資產的 ARM 資源識別碼。 字串
命令 [必要]在作業啟動時執行的命令。 例如 “python train.py” 需要字串 ()

約束:
最小長度 = 1
模式 = [a-zA-Z0-9_]
散發 (distribution) 作業的散發組態。 如果設定,這應該是 Mpi、Tensorflow、PyTorch 或 null 的其中一個。 DistributionConfiguration
environmentId [必要]作業之環境規格的 ARM 資源識別碼。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]
environmentVariables 作業中包含的環境變數。 CommandJobEnvironmentVariables
輸入 對應作業中使用的輸入數據系結。 CommandJobInputs
限制 命令作業限制。 CommandJobLimits
outputs 對應作業中使用的輸出數據系結。 CommandJobOutputs
resources 作業的計算資源組態。 ResourceConfiguration

DistributionConfiguration

名稱 描述
distributionType 設定物件類型 Mpi
PyTorch
TensorFlow (必要)

Mpi

名稱 描述
distributionType [必要]指定散發架構的類型。 需要 『Mpi』 ()
processCountPerInstance 每個 MPI 節點的進程數目。 int

PyTorch

名稱 描述
distributionType [必要]指定散發架構的類型。 需要 『PyTorch』 ()
processCountPerInstance 每個節點的進程數目。 int

TensorFlow

名稱 描述
distributionType [必要]指定散發架構的類型。 必要) 'TensorFlow' (
parameterServerCount 參數伺服器工作的數目。 int
workerCount 背景工作數目。 如果未指定,則會預設為實例計數。 int

CommandJobEnvironmentVariables

名稱 描述
{自定義屬性} 字串

CommandJobInputs

名稱 描述
{自定義屬性} JobInput

JobInput

名稱 描述
description 輸入的描述。 字串
jobInputType 設定物件類型 CustomModel
文字
MLFlowModel
MLTable
TritonModel
UriFile
需要 uriFolder ()

CustomModelJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 'CustomModel' (必要)
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

LiteralJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 必要) ('Literal'
[必要]輸入的常值。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

MLFlowModelJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 必要) 的 'MLFlowModel' (
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

TritonModelJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 'TritonModel' (必要)
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

UriFileJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 'UriFile' (必要)
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

UriFolderJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 'UriFolder' (必要)
mode 輸入資產傳遞模式。 'Direct'
'Download'
'EvalDownload'
'EvalMount'
'ReadOnlyMount'
'ReadWriteMount'
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

CommandJobLimits

名稱 描述
jobLimitsType [必要]JobLimit 類型。 'Command'
需要 「掃掠」 ()
timeout ISO 8601 格式的最大執行持續時間,之後作業將會取消。 僅支援有效位數為秒數的持續時間。 字串

CommandJobOutputs

名稱 描述
{自定義屬性} JobOutput

PipelineJob

名稱 描述
jobType [必要]指定作業的類型。 需要 『Pipeline』 ()
輸入 管線作業的輸入。 PipelineJobInputs
jobs 作業會建構管線作業。 PipelineJobJobs
outputs 管線作業的輸出 PipelineJobOutputs
設定 管線設定,例如 ContinueRunOnStepFailure 等專案。

PipelineJobInputs

名稱 描述
{自定義屬性} JobInput

PipelineJobJobs

名稱 描述
{自定義屬性}

PipelineJobOutputs

名稱 描述
{自定義屬性} JobOutput

SweepJob

名稱 描述
jobType [必要]指定作業的類型。 需要 「掃掠」 ()
earlyTermination 早期終止原則可在完成之前取消效能不佳的執行 EarlyTerminationPolicy
輸入 對應作業中使用的輸入數據系結。 SweepJobInputs
限制 掃掠作業限制。 SweepJobLimits
objective [必要]優化目標。 必要 目標 ()
outputs 對應作業中使用的輸出數據系結。 SweepJobOutputs
samplingAlgorithm [必要]超參數取樣演算法 SamplingAlgorithm (必要)
searchSpace [必要]包含每個參數及其分佈的字典。 字典索引鍵是參數的名稱
試用版 [必要]試用版元件定義。 必要) 的 TrialComponent (

SweepJobInputs

名稱 描述
{自定義屬性} JobInput

SweepJobLimits

名稱 描述
jobLimitsType [必要]JobLimit 類型。 'Command'
需要 「掃掠」 ()
maxConcurrentTrials 掃掠作業最大並行試用版。 int
maxTotalTrials 掃掠作業總試用版數上限。 int
timeout ISO 8601 格式的最大執行持續時間,之後作業將會取消。 僅支援有效位數為秒數的持續時間。 字串
trialTimeout 掃掠作業試用逾時值。 字串

目標

名稱 描述
goal [必要]定義超參數微調支援的計量目標 '最大化'
必要) 『最小化 (」。
primaryMetric [必要]要優化之計量的名稱。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

SweepJobOutputs

名稱 描述
{自定義屬性} JobOutput

SamplingAlgorithm

名稱 描述
samplingAlgorithmType 設定物件類型 貝 葉 斯
方格
需要隨機 ()

BayesianSamplingAlgorithm

名稱 描述
samplingAlgorithmType [必要]用來產生超參數值的演算法,以及組態屬性 需要 『Bayesian』 ()

GridSamplingAlgorithm

名稱 描述
samplingAlgorithmType [必要]用來產生超參數值的演算法,以及組態屬性 'Grid' (必要)

RandomSamplingAlgorithm

名稱 描述
samplingAlgorithmType [必要]用來產生超參數值的演算法,以及組態屬性 必要) 的 'Random' (
rule (規則) 隨機演算法的特定類型 'Random'
'Sobol'
seed 要作為隨機數產生種子的選擇性整數 int

TrialComponent

名稱 描述
codeId 程式代碼資產的 ARM 資源識別碼。 字串
命令 [必要]在作業啟動時執行的命令。 例如 “python train.py” 字串 (必要)

約束:
最小長度 = 1
模式 = [a-zA-Z0-9_]
散發 (distribution) 作業的散發組態。 如果設定,這應該是 Mpi、Tensorflow、PyTorch 或 null 的其中一個。 DistributionConfiguration
environmentId [必要]作業環境規格的 ARM 資源識別碼。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]
environmentVariables 作業中包含的環境變數。 TrialComponentEnvironmentVariables
resources 作業的計算資源組態。 ResourceConfiguration

TrialComponentEnvironmentVariables

名稱 描述
{自定義屬性} 字串

快速入門範本

下列快速入門範本會部署此資源類型。

範本 描述
建立 Azure Machine Learning AutoML 分類作業

部署至 Azure
此範本會建立 Azure Machine Learning AutoML 分類作業,以找出預測用戶端是否會向金融機構訂閱固定期存金的最佳模型。
建立 Azure Machine Learning 命令作業

部署至 Azure
此範本會建立具有基本hello_world腳本的 Azure Machine Learning 命令作業
建立 Azure Machine Learning 掃掠作業

部署至 Azure
此範本會建立 Azure Machine Learning 掃掠作業以進行超參數微調。

Terraform (AzAPI 提供者) 資源定義

工作區/作業資源類型可以使用目標作業來部署:

  • 資源群組

如需每個 API 版本中已變更屬性的清單,請參閱 變更記錄

資源格式

若要建立 Microsoft.MachineLearningServices/workspaces/jobs 資源,請將下列 Terraform 新增至您的範本。

resource "azapi_resource" "symbolicname" {
  type = "Microsoft.MachineLearningServices/workspaces/jobs@2022-02-01-preview"
  name = "string"
  parent_id = "string"
  body = jsonencode({
    properties = {
      computeId = "string"
      description = "string"
      displayName = "string"
      experimentName = "string"
      identity = {
        identityType = "string"
        // For remaining properties, see IdentityConfiguration objects
      }
      isArchived = bool
      properties = {
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
      }
      schedule = {
        endTime = "string"
        scheduleStatus = "string"
        startTime = "string"
        timeZone = "string"
        scheduleType = "string"
        // For remaining properties, see ScheduleBase objects
      }
      services = {
        {customized property} = {
          endpoint = "string"
          jobServiceType = "string"
          port = int
          properties = {
            {customized property} = "string"
          }
        }
      }
      tags = {
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
        {customized property} = "string"
      }
      jobType = "string"
      // For remaining properties, see JobBaseDetails objects
    }
  })
}

JobBaseDetails 物件

設定 jobType 屬性以指定物件的類型。

針對 AutoML,請使用:

  jobType = "AutoML"
  environmentId = "string"
  environmentVariables = {
    {customized property} = "string"
  }
  outputs = {
    {customized property} = {
      description = "string"
      jobOutputType = "string"
      // For remaining properties, see JobOutput objects
    }
  }
  resources = {
    instanceCount = int
    instanceType = "string"
    properties = {}
  }
  taskDetails = {
    logVerbosity = "string"
    taskType = "string"
    // For remaining properties, see AutoMLVertical objects
  }

針對 [命令],請使用:

  jobType = "Command"
  codeId = "string"
  command = "string"
  distribution = {
    distributionType = "string"
    // For remaining properties, see DistributionConfiguration objects
  }
  environmentId = "string"
  environmentVariables = {
    {customized property} = "string"
  }
  inputs = {
    {customized property} = {
      description = "string"
      jobInputType = "string"
      // For remaining properties, see JobInput objects
    }
  }
  limits = {
    jobLimitsType = "string"
    timeout = "string"
  }
  outputs = {
    {customized property} = {
      description = "string"
      jobOutputType = "string"
      // For remaining properties, see JobOutput objects
    }
  }
  resources = {
    instanceCount = int
    instanceType = "string"
    properties = {}
  }

針對 管線,請使用:

  jobType = "Pipeline"
  inputs = {
    {customized property} = {
      description = "string"
      jobInputType = "string"
      // For remaining properties, see JobInput objects
    }
  }
  jobs = {}
  outputs = {
    {customized property} = {
      description = "string"
      jobOutputType = "string"
      // For remaining properties, see JobOutput objects
    }
  }

針對 掃掠,請使用:

  jobType = "Sweep"
  earlyTermination = {
    delayEvaluation = int
    evaluationInterval = int
    policyType = "string"
    // For remaining properties, see EarlyTerminationPolicy objects
  }
  inputs = {
    {customized property} = {
      description = "string"
      jobInputType = "string"
      // For remaining properties, see JobInput objects
    }
  }
  limits = {
    jobLimitsType = "string"
    maxConcurrentTrials = int
    maxTotalTrials = int
    timeout = "string"
    trialTimeout = "string"
  }
  objective = {
    goal = "string"
    primaryMetric = "string"
  }
  outputs = {
    {customized property} = {
      description = "string"
      jobOutputType = "string"
      // For remaining properties, see JobOutput objects
    }
  }
  samplingAlgorithm = {
    samplingAlgorithmType = "string"
    // For remaining properties, see SamplingAlgorithm objects
  }
  trial = {
    codeId = "string"
    command = "string"
    distribution = {
      distributionType = "string"
      // For remaining properties, see DistributionConfiguration objects
    }
    environmentId = "string"
    environmentVariables = {
      {customized property} = "string"
    }
    resources = {
      instanceCount = int
      instanceType = "string"
      properties = {}
    }
  }

IdentityConfiguration 物件

設定 identityType 屬性以指定物件的類型。

針對 AMLToken,請使用:

  identityType = "AMLToken"

針對 Managed,請使用:

  identityType = "Managed"
  clientId = "string"
  objectId = "string"
  resourceId = "string"

針對 UserIdentity,請使用:

  identityType = "UserIdentity"

ScheduleBase 物件

設定 scheduleType 屬性以指定物件的類型。

針對 Cron,請使用:

  scheduleType = "Cron"
  expression = "string"

針對 [週期],請使用:

  scheduleType = "Recurrence"
  frequency = "string"
  interval = int
  pattern = {
    hours = [
      int
    ]
    minutes = [
      int
    ]
    weekdays = [
      "string"
    ]
  }

JobOutput 物件

設定 jobOutputType 屬性以指定對象的類型。

針對 CustomModel,請使用:

  jobOutputType = "CustomModel"
  mode = "string"
  uri = "string"

針對 MLFlowModel,請使用:

  jobOutputType = "MLFlowModel"
  mode = "string"
  uri = "string"

針對 MLTable,請使用:

  jobOutputType = "MLTable"
  mode = "string"
  uri = "string"

針對 TritonModel,請使用:

  jobOutputType = "TritonModel"
  mode = "string"
  uri = "string"

針對 UriFile,請使用:

  jobOutputType = "UriFile"
  mode = "string"
  uri = "string"

針對 UriFolder,請使用:

  jobOutputType = "UriFolder"
  mode = "string"
  uri = "string"

AutoMLVertical 物件

設定 taskType 屬性以指定物件的類型。

針對 [分類],請使用:

  taskType = "Classification"
  allowedModels = [
    "string"
  ]
  blockedModels = [
    "string"
  ]
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      cvSplitColumnNames = [
        "string"
      ]
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      nCrossValidations = {
        mode = "string"
        // For remaining properties, see NCrossValidations objects
      }
      validationDataSize = int
    }
    weightColumnName = "string"
  }
  featurizationSettings = {
    blockedTransformers = [
      "string"
    ]
    columnNameAndTypes = {
      {customized property} = "string"
    }
    datasetLanguage = "string"
    dropColumns = [
      "string"
    ]
    enableDnnFeaturization = bool
    mode = "string"
    transformerParams = {
      {customized property} = [
        {
          fields = [
            "string"
          ]
        }
      ]
    }
  }
  limitSettings = {
    enableEarlyTermination = bool
    exitScore = int
    maxConcurrentTrials = int
    maxCoresPerTrial = int
    maxTrials = int
    timeout = "string"
    trialTimeout = "string"
  }
  primaryMetric = "string"
  trainingSettings = {
    enableDnnTraining = bool
    enableModelExplainability = bool
    enableOnnxCompatibleModels = bool
    enableStackEnsemble = bool
    enableVoteEnsemble = bool
    ensembleModelDownloadTimeout = "string"
    stackEnsembleSettings = {
      stackMetaLearnerTrainPercentage = int
      stackMetaLearnerType = "string"
    }
  }

針對 預測,請使用:

  taskType = "Forecasting"
  allowedModels = [
    "string"
  ]
  blockedModels = [
    "string"
  ]
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      cvSplitColumnNames = [
        "string"
      ]
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      nCrossValidations = {
        mode = "string"
        // For remaining properties, see NCrossValidations objects
      }
      validationDataSize = int
    }
    weightColumnName = "string"
  }
  featurizationSettings = {
    blockedTransformers = [
      "string"
    ]
    columnNameAndTypes = {
      {customized property} = "string"
    }
    datasetLanguage = "string"
    dropColumns = [
      "string"
    ]
    enableDnnFeaturization = bool
    mode = "string"
    transformerParams = {
      {customized property} = [
        {
          fields = [
            "string"
          ]
        }
      ]
    }
  }
  forecastingSettings = {
    countryOrRegionForHolidays = "string"
    cvStepSize = int
    featureLags = "string"
    forecastHorizon = {
      mode = "string"
      // For remaining properties, see ForecastHorizon objects
    }
    frequency = "string"
    seasonality = {
      mode = "string"
      // For remaining properties, see Seasonality objects
    }
    shortSeriesHandlingConfig = "string"
    targetAggregateFunction = "string"
    targetLags = {
      mode = "string"
      // For remaining properties, see TargetLags objects
    }
    targetRollingWindowSize = {
      mode = "string"
      // For remaining properties, see TargetRollingWindowSize objects
    }
    timeColumnName = "string"
    timeSeriesIdColumnNames = [
      "string"
    ]
    useStl = "string"
  }
  limitSettings = {
    enableEarlyTermination = bool
    exitScore = int
    maxConcurrentTrials = int
    maxCoresPerTrial = int
    maxTrials = int
    timeout = "string"
    trialTimeout = "string"
  }
  primaryMetric = "string"
  trainingSettings = {
    enableDnnTraining = bool
    enableModelExplainability = bool
    enableOnnxCompatibleModels = bool
    enableStackEnsemble = bool
    enableVoteEnsemble = bool
    ensembleModelDownloadTimeout = "string"
    stackEnsembleSettings = {
      stackMetaLearnerTrainPercentage = int
      stackMetaLearnerType = "string"
    }
  }

針對 ImageClassification,請使用:

  taskType = "ImageClassification"
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      validationDataSize = int
    }
  }
  limitSettings = {
    maxConcurrentTrials = int
    maxTrials = int
    timeout = "string"
  }
  modelSettings = {
    advancedSettings = "string"
    amsGradient = bool
    augmentations = "string"
    beta1 = int
    beta2 = int
    checkpointDatasetId = "string"
    checkpointFilename = "string"
    checkpointFrequency = int
    checkpointRunId = "string"
    distributed = bool
    earlyStopping = bool
    earlyStoppingDelay = int
    earlyStoppingPatience = int
    enableOnnxNormalization = bool
    evaluationFrequency = int
    gradientAccumulationStep = int
    layersToFreeze = int
    learningRate = int
    learningRateScheduler = "string"
    modelName = "string"
    momentum = int
    nesterov = bool
    numberOfEpochs = int
    numberOfWorkers = int
    optimizer = "string"
    randomSeed = int
    splitRatio = int
    stepLRGamma = int
    stepLRStepSize = int
    trainingBatchSize = int
    trainingCropSize = int
    validationBatchSize = int
    validationCropSize = int
    validationResizeSize = int
    warmupCosineLRCycles = int
    warmupCosineLRWarmupEpochs = int
    weightDecay = int
    weightedLoss = int
  }
  primaryMetric = "string"
  searchSpace = [
    {
      amsGradient = "string"
      augmentations = "string"
      beta1 = "string"
      beta2 = "string"
      distributed = "string"
      earlyStopping = "string"
      earlyStoppingDelay = "string"
      earlyStoppingPatience = "string"
      enableOnnxNormalization = "string"
      evaluationFrequency = "string"
      gradientAccumulationStep = "string"
      layersToFreeze = "string"
      learningRate = "string"
      learningRateScheduler = "string"
      modelName = "string"
      momentum = "string"
      nesterov = "string"
      numberOfEpochs = "string"
      numberOfWorkers = "string"
      optimizer = "string"
      randomSeed = "string"
      splitRatio = "string"
      stepLRGamma = "string"
      stepLRStepSize = "string"
      trainingBatchSize = "string"
      trainingCropSize = "string"
      validationBatchSize = "string"
      validationCropSize = "string"
      validationResizeSize = "string"
      warmupCosineLRCycles = "string"
      warmupCosineLRWarmupEpochs = "string"
      weightDecay = "string"
      weightedLoss = "string"
    }
  ]
  sweepSettings = {
    earlyTermination = {
      delayEvaluation = int
      evaluationInterval = int
      policyType = "string"
      // For remaining properties, see EarlyTerminationPolicy objects
    }
    limits = {
      maxConcurrentTrials = int
      maxTrials = int
    }
    samplingAlgorithm = "string"
  }

針對 ImageClassificationMultilabel,請使用:

  taskType = "ImageClassificationMultilabel"
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      validationDataSize = int
    }
  }
  limitSettings = {
    maxConcurrentTrials = int
    maxTrials = int
    timeout = "string"
  }
  modelSettings = {
    advancedSettings = "string"
    amsGradient = bool
    augmentations = "string"
    beta1 = int
    beta2 = int
    checkpointDatasetId = "string"
    checkpointFilename = "string"
    checkpointFrequency = int
    checkpointRunId = "string"
    distributed = bool
    earlyStopping = bool
    earlyStoppingDelay = int
    earlyStoppingPatience = int
    enableOnnxNormalization = bool
    evaluationFrequency = int
    gradientAccumulationStep = int
    layersToFreeze = int
    learningRate = int
    learningRateScheduler = "string"
    modelName = "string"
    momentum = int
    nesterov = bool
    numberOfEpochs = int
    numberOfWorkers = int
    optimizer = "string"
    randomSeed = int
    splitRatio = int
    stepLRGamma = int
    stepLRStepSize = int
    trainingBatchSize = int
    trainingCropSize = int
    validationBatchSize = int
    validationCropSize = int
    validationResizeSize = int
    warmupCosineLRCycles = int
    warmupCosineLRWarmupEpochs = int
    weightDecay = int
    weightedLoss = int
  }
  primaryMetric = "string"
  searchSpace = [
    {
      amsGradient = "string"
      augmentations = "string"
      beta1 = "string"
      beta2 = "string"
      distributed = "string"
      earlyStopping = "string"
      earlyStoppingDelay = "string"
      earlyStoppingPatience = "string"
      enableOnnxNormalization = "string"
      evaluationFrequency = "string"
      gradientAccumulationStep = "string"
      layersToFreeze = "string"
      learningRate = "string"
      learningRateScheduler = "string"
      modelName = "string"
      momentum = "string"
      nesterov = "string"
      numberOfEpochs = "string"
      numberOfWorkers = "string"
      optimizer = "string"
      randomSeed = "string"
      splitRatio = "string"
      stepLRGamma = "string"
      stepLRStepSize = "string"
      trainingBatchSize = "string"
      trainingCropSize = "string"
      validationBatchSize = "string"
      validationCropSize = "string"
      validationResizeSize = "string"
      warmupCosineLRCycles = "string"
      warmupCosineLRWarmupEpochs = "string"
      weightDecay = "string"
      weightedLoss = "string"
    }
  ]
  sweepSettings = {
    earlyTermination = {
      delayEvaluation = int
      evaluationInterval = int
      policyType = "string"
      // For remaining properties, see EarlyTerminationPolicy objects
    }
    limits = {
      maxConcurrentTrials = int
      maxTrials = int
    }
    samplingAlgorithm = "string"
  }

針對 ImageInstanceSegmentation,請使用:

  taskType = "ImageInstanceSegmentation"
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      validationDataSize = int
    }
  }
  limitSettings = {
    maxConcurrentTrials = int
    maxTrials = int
    timeout = "string"
  }
  modelSettings = {
    advancedSettings = "string"
    amsGradient = bool
    augmentations = "string"
    beta1 = int
    beta2 = int
    boxDetectionsPerImage = int
    boxScoreThreshold = int
    checkpointDatasetId = "string"
    checkpointFilename = "string"
    checkpointFrequency = int
    checkpointRunId = "string"
    distributed = bool
    earlyStopping = bool
    earlyStoppingDelay = int
    earlyStoppingPatience = int
    enableOnnxNormalization = bool
    evaluationFrequency = int
    gradientAccumulationStep = int
    imageSize = int
    layersToFreeze = int
    learningRate = int
    learningRateScheduler = "string"
    maxSize = int
    minSize = int
    modelName = "string"
    modelSize = "string"
    momentum = int
    multiScale = bool
    nesterov = bool
    nmsIouThreshold = int
    numberOfEpochs = int
    numberOfWorkers = int
    optimizer = "string"
    randomSeed = int
    splitRatio = int
    stepLRGamma = int
    stepLRStepSize = int
    tileGridSize = "string"
    tileOverlapRatio = int
    tilePredictionsNmsThreshold = int
    trainingBatchSize = int
    validationBatchSize = int
    validationIouThreshold = int
    validationMetricType = "string"
    warmupCosineLRCycles = int
    warmupCosineLRWarmupEpochs = int
    weightDecay = int
  }
  primaryMetric = "MeanAveragePrecision"
  searchSpace = [
    {
      amsGradient = "string"
      augmentations = "string"
      beta1 = "string"
      beta2 = "string"
      boxDetectionsPerImage = "string"
      boxScoreThreshold = "string"
      distributed = "string"
      earlyStopping = "string"
      earlyStoppingDelay = "string"
      earlyStoppingPatience = "string"
      enableOnnxNormalization = "string"
      evaluationFrequency = "string"
      gradientAccumulationStep = "string"
      imageSize = "string"
      layersToFreeze = "string"
      learningRate = "string"
      learningRateScheduler = "string"
      maxSize = "string"
      minSize = "string"
      modelName = "string"
      modelSize = "string"
      momentum = "string"
      multiScale = "string"
      nesterov = "string"
      nmsIouThreshold = "string"
      numberOfEpochs = "string"
      numberOfWorkers = "string"
      optimizer = "string"
      randomSeed = "string"
      splitRatio = "string"
      stepLRGamma = "string"
      stepLRStepSize = "string"
      tileGridSize = "string"
      tileOverlapRatio = "string"
      tilePredictionsNmsThreshold = "string"
      trainingBatchSize = "string"
      validationBatchSize = "string"
      validationIouThreshold = "string"
      validationMetricType = "string"
      warmupCosineLRCycles = "string"
      warmupCosineLRWarmupEpochs = "string"
      weightDecay = "string"
    }
  ]
  sweepSettings = {
    earlyTermination = {
      delayEvaluation = int
      evaluationInterval = int
      policyType = "string"
      // For remaining properties, see EarlyTerminationPolicy objects
    }
    limits = {
      maxConcurrentTrials = int
      maxTrials = int
    }
    samplingAlgorithm = "string"
  }

針對 ImageObjectDetection,請使用:

  taskType = "ImageObjectDetection"
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      validationDataSize = int
    }
  }
  limitSettings = {
    maxConcurrentTrials = int
    maxTrials = int
    timeout = "string"
  }
  modelSettings = {
    advancedSettings = "string"
    amsGradient = bool
    augmentations = "string"
    beta1 = int
    beta2 = int
    boxDetectionsPerImage = int
    boxScoreThreshold = int
    checkpointDatasetId = "string"
    checkpointFilename = "string"
    checkpointFrequency = int
    checkpointRunId = "string"
    distributed = bool
    earlyStopping = bool
    earlyStoppingDelay = int
    earlyStoppingPatience = int
    enableOnnxNormalization = bool
    evaluationFrequency = int
    gradientAccumulationStep = int
    imageSize = int
    layersToFreeze = int
    learningRate = int
    learningRateScheduler = "string"
    maxSize = int
    minSize = int
    modelName = "string"
    modelSize = "string"
    momentum = int
    multiScale = bool
    nesterov = bool
    nmsIouThreshold = int
    numberOfEpochs = int
    numberOfWorkers = int
    optimizer = "string"
    randomSeed = int
    splitRatio = int
    stepLRGamma = int
    stepLRStepSize = int
    tileGridSize = "string"
    tileOverlapRatio = int
    tilePredictionsNmsThreshold = int
    trainingBatchSize = int
    validationBatchSize = int
    validationIouThreshold = int
    validationMetricType = "string"
    warmupCosineLRCycles = int
    warmupCosineLRWarmupEpochs = int
    weightDecay = int
  }
  primaryMetric = "MeanAveragePrecision"
  searchSpace = [
    {
      amsGradient = "string"
      augmentations = "string"
      beta1 = "string"
      beta2 = "string"
      boxDetectionsPerImage = "string"
      boxScoreThreshold = "string"
      distributed = "string"
      earlyStopping = "string"
      earlyStoppingDelay = "string"
      earlyStoppingPatience = "string"
      enableOnnxNormalization = "string"
      evaluationFrequency = "string"
      gradientAccumulationStep = "string"
      imageSize = "string"
      layersToFreeze = "string"
      learningRate = "string"
      learningRateScheduler = "string"
      maxSize = "string"
      minSize = "string"
      modelName = "string"
      modelSize = "string"
      momentum = "string"
      multiScale = "string"
      nesterov = "string"
      nmsIouThreshold = "string"
      numberOfEpochs = "string"
      numberOfWorkers = "string"
      optimizer = "string"
      randomSeed = "string"
      splitRatio = "string"
      stepLRGamma = "string"
      stepLRStepSize = "string"
      tileGridSize = "string"
      tileOverlapRatio = "string"
      tilePredictionsNmsThreshold = "string"
      trainingBatchSize = "string"
      validationBatchSize = "string"
      validationIouThreshold = "string"
      validationMetricType = "string"
      warmupCosineLRCycles = "string"
      warmupCosineLRWarmupEpochs = "string"
      weightDecay = "string"
    }
  ]
  sweepSettings = {
    earlyTermination = {
      delayEvaluation = int
      evaluationInterval = int
      policyType = "string"
      // For remaining properties, see EarlyTerminationPolicy objects
    }
    limits = {
      maxConcurrentTrials = int
      maxTrials = int
    }
    samplingAlgorithm = "string"
  }

針對 回歸,請使用:

  taskType = "Regression"
  allowedModels = [
    "string"
  ]
  blockedModels = [
    "string"
  ]
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      cvSplitColumnNames = [
        "string"
      ]
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      nCrossValidations = {
        mode = "string"
        // For remaining properties, see NCrossValidations objects
      }
      validationDataSize = int
    }
    weightColumnName = "string"
  }
  featurizationSettings = {
    blockedTransformers = [
      "string"
    ]
    columnNameAndTypes = {
      {customized property} = "string"
    }
    datasetLanguage = "string"
    dropColumns = [
      "string"
    ]
    enableDnnFeaturization = bool
    mode = "string"
    transformerParams = {
      {customized property} = [
        {
          fields = [
            "string"
          ]
        }
      ]
    }
  }
  limitSettings = {
    enableEarlyTermination = bool
    exitScore = int
    maxConcurrentTrials = int
    maxCoresPerTrial = int
    maxTrials = int
    timeout = "string"
    trialTimeout = "string"
  }
  primaryMetric = "string"
  trainingSettings = {
    enableDnnTraining = bool
    enableModelExplainability = bool
    enableOnnxCompatibleModels = bool
    enableStackEnsemble = bool
    enableVoteEnsemble = bool
    ensembleModelDownloadTimeout = "string"
    stackEnsembleSettings = {
      stackMetaLearnerTrainPercentage = int
      stackMetaLearnerType = "string"
    }
  }

針對 TextClassification,請使用:

  taskType = "TextClassification"
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      validationDataSize = int
    }
  }
  featurizationSettings = {
    datasetLanguage = "string"
  }
  limitSettings = {
    maxConcurrentTrials = int
    maxTrials = int
    timeout = "string"
  }
  primaryMetric = "string"

針對 TextClassificationMultilabel,請使用:

  taskType = "TextClassificationMultilabel"
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      validationDataSize = int
    }
  }
  featurizationSettings = {
    datasetLanguage = "string"
  }
  limitSettings = {
    maxConcurrentTrials = int
    maxTrials = int
    timeout = "string"
  }

針對 TextNER,請使用:

  taskType = "TextNER"
  dataSettings = {
    targetColumnName = "string"
    testData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      testDataSize = int
    }
    trainingData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
    }
    validationData = {
      data = {
        description = "string"
        jobInputType = "string"
        mode = "string"
        uri = "string"
      }
      validationDataSize = int
    }
  }
  featurizationSettings = {
    datasetLanguage = "string"
  }
  limitSettings = {
    maxConcurrentTrials = int
    maxTrials = int
    timeout = "string"
  }

NCrossValidations 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode = "Auto"

針對 [自定義],請使用:

  mode = "Custom"
  value = int

ForecastHorizon 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode = "Auto"

針對 [自定義],請使用:

  mode = "Custom"
  value = int

季節性物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode = "Auto"

針對 [自定義],請使用:

  mode = "Custom"
  value = int

TargetLags 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode = "Auto"

針對 [自定義],請使用:

  mode = "Custom"
  values = [
    int
  ]

TargetRollingWindowSize 物件

設定 mode 屬性以指定物件的類型。

針對 [自動],請使用:

  mode = "Auto"

針對 [自定義],請使用:

  mode = "Custom"
  value = int

EarlyTerminationPolicy 物件

設定 policyType 屬性以指定物件的類型。

針對 Bandit,請使用:

  policyType = "Bandit"
  slackAmount = int
  slackFactor = int

針對 MedianStopping,請使用:

  policyType = "MedianStopping"

針對 TruncationSelection,請使用:

  policyType = "TruncationSelection"
  truncationPercentage = int

DistributionConfiguration 物件

設定 distributionType 屬性以指定物件的類型。

針對 Mpi,請使用:

  distributionType = "Mpi"
  processCountPerInstance = int

針對 PyTorch,請使用:

  distributionType = "PyTorch"
  processCountPerInstance = int

針對 TensorFlow,請使用:

  distributionType = "TensorFlow"
  parameterServerCount = int
  workerCount = int

JobInput 物件

設定 jobInputType 屬性以指定物件的類型。

針對 CustomModel,請使用:

  jobInputType = "CustomModel"
  mode = "string"
  uri = "string"

針對 常值,請使用:

  jobInputType = "Literal"
  value = "string"

針對 MLFlowModel,請使用:

  jobInputType = "MLFlowModel"
  mode = "string"
  uri = "string"

針對 MLTable,請使用:

  jobInputType = "MLTable"
  mode = "string"
  uri = "string"

針對 TritonModel,請使用:

  jobInputType = "TritonModel"
  mode = "string"
  uri = "string"

針對 UriFile,請使用:

  jobInputType = "UriFile"
  mode = "string"
  uri = "string"

針對 UriFolder,請使用:

  jobInputType = "UriFolder"
  mode = "string"
  uri = "string"

SamplingAlgorithm 物件

設定 samplingAlgorithmType 屬性以指定物件的類型。

針對 貝氏,請使用:

  samplingAlgorithmType = "Bayesian"

針對 Grid,請使用:

  samplingAlgorithmType = "Grid"

針對 [隨機],請使用:

  samplingAlgorithmType = "Random"
  rule = "string"
  seed = int

屬性值

workspaces/jobs

名稱 描述
類型 資源類型 “Microsoft.MachineLearningServices/workspaces/jobs@2022-02-01-preview”
NAME 資源名稱 需要字串 ()
parent_id 此資源父系之資源的標識碼。 類型資源的標識碼: 工作區
properties [必要]實體的其他屬性。 JobBaseDetails (必要)

JobBaseDetails

名稱 描述
computeId 計算資源的 ARM 資源識別碼。 字串
description 資產描述文字。 字串
displayName 工作的顯示名稱。 字串
experimentName 作業所屬的實驗名稱。 如果未設定,作業會放在「預設」實驗中。 字串
身分識別 身分識別組態。 如果設定,這應該是其中一個 AmlToken、ManagedIdentity、UserIdentity 或 null。
如果為 null,則預設為 AmlToken。
IdentityConfiguration
isArchived 資產是否已封存? bool
properties 資產屬性字典。 ResourceBaseProperties
schedule 作業的排程定義。
如果未提供排程,作業會在提交后立即執行一次。
ScheduleBase
服務 JobEndpoints 的清單。
針對本機作業,作業端點會有 FileStreamObject 的端點值。
JobBaseServices
tags 標記字典。 標記可以新增、移除和更新。 物件 (object)
jobType 設定物件類型 AutoML
命令
管線
需要掃掠 ()

IdentityConfiguration

名稱 描述
identityType 設定物件類型 AMLToken
受控
必要) 的 UserIdentity (

AmlToken

名稱 描述
identityType [必要]指定身分識別架構的類型。 “AMLToken” (必要)

ManagedIdentity

名稱 描述
identityType [必要]指定身分識別架構的類型。 「受控」 (必要)
clientId 依用戶端識別元指定使用者指派的身分識別。 針對系統指派,請勿設定此欄位。 字串

約束:
最小長度 = 36
最大長度 = 36
模式 = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$
objectId 依物件識別元指定使用者指派的身分識別。 針對系統指派,請勿設定此欄位。 字串

約束:
最小長度 = 36
最大長度 = 36
模式 = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$
resourceId 依 ARM 資源識別碼指定使用者指派的身分識別。 針對系統指派,請勿設定此欄位。 字串

UserIdentity

名稱 描述
identityType [必要]指定身分識別架構的類型。 必要) “UserIdentity” (

ResourceBaseProperties

名稱 描述
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串
{自定義屬性} 字串

ScheduleBase

名稱 描述
EndTime 以 ISO 8601 格式指定排程的結束時間。
如果沒有,排程將會無限期執行
字串
scheduleStatus 指定排程的狀態 “Disabled”
“Enabled”
startTime 指定 ISO 8601 格式的排程開始時間。 字串
timeZone 指定排程執行所在的時區。
TimeZone應遵循 Windows 時區格式。
字串
scheduleType 設定物件類型 Cron
需要週期 性 ()

CronSchedule

名稱 描述
scheduleType [必要]指定排程類型 “Cron” (必要)
expression [必要]指定排程的cron運算式。
表達式應該遵循 NCronTab 格式。
字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

RecurrenceSchedule

名稱 描述
scheduleType [必要]指定排程類型 必要的「週期」 ()
frequency [必要]指定要觸發排程的頻率 “Day”
“Hour”
“Minute”
“Month”
「周」 (必要)
interval [必要]指定排程間隔與頻率 int (必要)
模式 指定週期排程模式 RecurrencePattern

RecurrencePattern

名稱 描述
小時 [必要]週期排程模式的時數清單 int[] (必要)
分鐘 [必要]週期排程模式的分鐘清單 int[] (必要)
平日 週期排程模式的工作日清單 包含任何項目的字串數組:
"Friday"
"Monday"
"Saturday"
"Sunday"
"Thursday"
"Tuesday"
"Wednesday"

JobBaseServices

名稱 描述
{自定義屬性} JobService

JobService

名稱 描述
端點 端點的 URL。 字串
jobServiceType 端點類型。 string
連接埠 端點的埠。 int
properties 在端點上設定的其他屬性。 JobServiceProperties

JobServiceProperties

名稱 描述
{自定義屬性} 字串

AutoMLJob

名稱 描述
jobType [必要]指定作業的類型。 必要) “AutoML” (
environmentId 作業環境規格的 ARM 資源識別碼。
這是選擇性值,如果未提供,AutoML 會在執行作業時,將此預設為生產 AutoML 策劃的環境版本。
字串
environmentVariables 作業中包含的環境變數。 AutoMLJobEnvironmentVariables
outputs 對應作業中使用的輸出數據系結。 AutoMLJobOutputs
resources 作業的計算資源組態。 ResourceConfiguration
taskDetails [必要]這代表可以是其中一個數據表/NLP/Image 的案例 需要 AutoMLVertical ()

AutoMLJobEnvironmentVariables

名稱 描述
{自定義屬性} 字串

AutoMLJobOutputs

名稱 描述
{自定義屬性} JobOutput

JobOutput

名稱 描述
description 輸出的描述。 字串
jobOutputType 設定物件類型 CustomModel
MLFlowModel
MLTable
TritonModel
UriFile
需要 uriFolder ()

CustomModelJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 必要) 的 「CustomModel」 (
mode 輸出資產傳遞模式。 “ReadWriteMount”
“Upload”
uri 輸出資產 URI。 字串

MLFlowModelJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 “MLFlowModel” (必要)
mode 輸出資產傳遞模式。 “ReadWriteMount”
“Upload”
uri 輸出資產 URI。 字串

MLTableJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 “MLTable” (必要)
mode 輸出資產傳遞模式。 “ReadWriteMount”
“Upload”
uri 輸出資產 URI。 字串

TritonModelJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 “TritonModel” (必要)
mode 輸出資產傳遞模式。 “ReadWriteMount”
“上傳”
uri 輸出資產 URI。 字串

UriFileJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 “UriFile” (必要)
mode 輸出資產傳遞模式。 “ReadWriteMount”
“上傳”
uri 輸出資產 URI。 字串

UriFolderJobOutput

名稱 描述
jobOutputType [必要]指定作業的類型。 “UriFolder” (必要)
mode 輸出資產傳遞模式。 “ReadWriteMount”
“上傳”
uri 輸出資產 URI。 字串

ResourceConfiguration

名稱 描述
instanceCount 計算目標所使用的選用執行個體或節點數目。 int
instanceType 計算目標所支持的選擇性 VM 類型。 字串
properties 其他屬性包。 ResourceConfigurationProperties

ResourceConfigurationProperties

名稱 描述
{自定義屬性}

AutoMLVertical

名稱 描述
logVerbosity 作業的記錄詳細資訊。 “Critical”
“Debug”
“錯誤”
“Info”
“NotSet”
「警告」
taskType 設定物件類型 分類
預測
ImageClassification
ImageClassificationMultilabel
ImageInstanceSegmentation
ImageObjectDetection
迴歸
TextClassification
TextClassificationMultilabel
需要 TextNER ()

分類

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 「分類」 (必要)
allowedModels 分類工作的允許模型。 包含任何項目的字串數組:
“BernoulliNaive設定”
“DecisionTree”
“ExtremeRandomTrees”
“GradientBoosting”
“KNN”
“LightGBM”
“LinearSVM”
“LogisticRegression”
“MultinomialNaiveBayes”
“RandomForest”
“SGD”
“SVM”
“XGBoostClassifier”
blockedModels 分類工作的封鎖模型。 包含任何項目的字串數組:
“BernoulliNaiveBaes”
“DecisionTree”
“ExtremeRandomTrees”
“GradientBoosting”
“KNN”
“LightGBM”
“LinearSVM”
“LogisticRegression”
“MultinomialNaiveBaes”
“RandomForest”
“SGD”
“SVM”
“XGBoostClassifier”
dataSettings AutoMLJob 的數據輸入。 TableVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 TableVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 TableVerticalLimitSettings
primaryMetric 工作的主要計量。 “AUCWeighted”
「精確度」
“AveragePrecisionScoreWeighted”
“NormMacroRecall”
“PrecisionScoreWeighted”
trainingSettings AutoML 作業定型階段的輸入。 TrainingSettings

TableVerticalDataSettings

名稱 描述
targetColumnName [必要]目標數據行名稱:這是預測值數據行。
也稱為分類工作內容中的標籤數據行名稱。
字串 (必要)

約束:
模式 = [a-zA-Z0-9_]
testData 測試數據輸入。 TestDataSettings
trainingData [必要]定型數據輸入。 TrainingDataSettings (必要)
validationData 驗證數據輸入。 TableVerticalValidationDataSettings
weightColumnName 範例加權數據行的名稱。 自動化 ML 支援加權數據行做為輸入,導致數據中的數據列向上或向下加權。 字串

TestDataSettings

名稱 描述
資料 測試數據 MLTable。 MLTableJobInput
testDataSize 必須針對驗證目的將測試數據集的分數放在一起。
介於 (0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

MLTableJobInput

名稱 描述
description 輸入的描述。 字串
jobInputType [必要]指定作業的類型。 “CustomModel”
“Literal”
“MLFlowModel”
“MLTable”
“TritonModel”
“UriFile”
必要) “UriFolder” (
mode 輸入資產傳遞模式。 “Direct”
“下載”
“EvalDownload”
“EvalMount”
“ReadOnlyMount”
“ReadWriteMount”
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

TrainingDataSettings

名稱 描述
資料 [必要]訓練數據MLTable。 MLTableJobInput (必要)

TableVerticalValidationDataSettings

名稱 描述
cvSplitColumnNames 要用於 CVSplit 資料的數據行。 string[]
資料 驗證數據 MLTable。 MLTableJobInput
nCrossValidations 要套用至定型數據集的交叉驗證折疊數目
未提供驗證數據集時。
NCrossValidations
validationDataSize 必須針對驗證目的將定型數據集的分數。
介於 (0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

NCrossValidations

名稱 描述
mode 設定物件類型 Auto
自訂 (必要)

AutoNCrossValidations

名稱 描述
mode [必要]判斷 N-Cross 驗證的模式。 必要) 「自動」 (

CustomNCrossValidations

名稱 描述
mode [必要]判斷 N-Cross 驗證的模式。 「自定義」 (必要)
[必要]N-Cross 驗證值。 int (必要)

TableVerticalFeaturizationSettings

名稱 描述
blockedTransformers 這些轉換器不得用於特徵化。 string[]
columnNameAndTypes 數據行名稱的字典及其類型 (int、float、string、datetime 等) 。 TableVerticalFeaturizationSettingsColumnNameAndTypes
datasetLanguage 數據集語言,適用於文字數據。 字串
dropColumns 特徵化期間要從數據卸除的數據行。 string[]
enableDnnFeaturization 判斷是否要使用 Dnn 型特徵化程式進行數據特徵化。 bool
mode 特徵化模式 - 使用者可以保留預設的「自動」模式,而 AutoML 會負責在特徵化階段中轉換數據的必要轉換。
如果選取 [關閉],則不會完成特徵化。
如果選取 [自定義],則使用者可以指定其他輸入,以自定義特徵化完成的方式。
“Auto”
"Custom"
“Off”
transformerParams 使用者可以指定其他要用於的轉換器,以及要套用它的數據行,以及轉換程式建構函式的參數。 TableVerticalFeaturizationSettingsTransformerParams

TableVerticalFeaturizationSettingsColumnNameAndTypes

名稱 描述
{自定義屬性} 字串

TableVerticalFeaturizationSettingsTransformerParams

名稱 描述
{自定義屬性} ColumnTransformer[]

ColumnTransformer

名稱 描述
fields 要套用轉換器邏輯的欄位。 string[]
參數 要傳遞至轉換器的不同屬性。
預期的輸入是 JSON 格式的索引鍵、值組字典。

TableVerticalLimitSettings

名稱 描述
enableEarlyTermination 啟用提早終止,判斷如果最後 20 個反覆項目沒有分數改善,AutoMLJob 是否提早終止。 bool
exitScore AutoML 作業的結束分數。 int
maxConcurrentTrials 並行反覆專案上限。 int
maxCoresPerTrial 每次反覆運算的核心數上限。 int
maxTrials 反覆運算次數。 int
timeout AutoML 作業逾時。 字串
trialTimeout 反覆專案逾時。 字串

TrainingSettings

名稱 描述
enableDnnTraining 啟用 DNN 模型的建議。 bool
enableModelExplainability 在最佳模型上開啟可解釋性的旗標。 bool
enableOnnxCompatibleModels 啟用 onnx 相容模型的旗標。 bool
enableStackEnsemble 啟用堆疊Ensemble 執行。 bool
enableVoteEnsemble 啟用投票Ensemble 執行。 bool
ensembleModelDownloadTimeout 在 VotingEnsemble 和 StackEnsemble 模型產生期間,會下載先前子回合中的多個配適模型。
如果需要更多時間,請將此參數設定為高於 300 秒的值。
字串
stackEnsembleSettings 堆疊Ensemble 執行的堆疊Ensemble 設定。 StackEnsembleSettings

StackEnsembleSettings

名稱 描述
stackMetaLearnerKWargs 要傳遞至中繼學習模組初始化表達式的選擇性參數。
stackMetaLearnerTrainPercentage 指定在選擇訓練和驗證類型的訓練) 保留給訓練中繼學習工具時,訓練集 (的比例。 預設值為 0.2。 int
stackMetaLearnerType 中繼學習模組是針對個別異質模型的輸出定型的模型。 “ElasticNet”
“ElasticNetCV”
“LightGBMClassifier”
“LightGBMRegressor”
“LinearRegression”
“LogisticRegression”
“LogisticRegressionCV”
"None"

預測

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 「預測」 (必要)
allowedModels 用於預測工作的允許模型。 包含任何項目的字串數組:
“Arimax”
“AutoArima”
“Average”
“DecisionTree”
“ElasticNet”
“ExponentialSmoothing”
“ExtremeRandomTrees”
“GradientBoosting”
“KNN”
“LassoLars”
“LightGBM”
“Naive”
“聯機”
“RandomForest”
“SGD”
“SeasonalAverage”
“SeasonalNaive”
“TCNForecaster”
“XGBoostRegressor”
blockedModels 用於預測工作的封鎖模型。 包含任何項目的字串數組:
“Arimax”
“AutoArima”
“Average”
“DecisionTree”
“ElasticNet”
“ExponentialSmoothing”
“ExtremeRandomTrees”
“GradientBoosting”
“KNN”
“LassoLars”
“LightGBM”
“Naive”
“聯機”
“RandomForest”
“SGD”
“SeasonalAverage”
“SeasonalNaive”
“TCNForecaster”
“XGBoostRegressor”
dataSettings AutoMLJob 的數據輸入。 TableVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 TableVerticalFeaturizationSettings
forecastingSettings 預測工作特定輸入。 ForecastingSettings
limitSettings AutoMLJob 的執行條件約束。 TableVerticalLimitSettings
primaryMetric 預測工作的主要計量。 “NormalizedMeanAbsoluteError”
“NormalizedRootMeanSquaredError”
“R2Score”
“SpearmanCorrelation”
trainingSettings AutoML 作業定型階段的輸入。 TrainingSettings

ForecastingSettings

名稱 描述
countryOrRegionForHolidays 用於預測工作的假日國家或地區。
這些應該是 ISO 3166 兩個字母的國家/地區代碼,例如 'US' 或 'GB'。
字串
cvStepSize 一個 CV 折疊的原點時間與下一個折疊之間的期間數。 For
例如,如果 CVStepSize = 3 表示每日數據,則每個折疊的源時間將會是
相隔三天。
int
featureLags 旗標,用於產生具有 'auto' 或 null 之數值特徵的延隔時間。 “Auto”
"None"
forecastHorizon 時間序列頻率單位所需的最大預測層級。 ForecastHorizon
frequency 預測時,此參數代表需要預測的期間,例如每日、每周、每年等等。預測頻率預設為數據集頻率。 字串
季節性 將時間序列季節性設定為數列頻率的整數倍數。
如果季節性設定為 『auto』,則會推斷它。
季節性
shortSeriesHandlingConfig 定義 AutoML 是否應該處理短時間序列的參數。 “Auto”
“Drop”
"None"
“Pad”
targetAggregateFunction 用來匯總時間序列目標數據行以符合使用者指定頻率的函式。
如果已設定 TargetAggregateFunction,亦即不是 『None』,但未設定 freq 參數,則會引發錯誤。 可能的目標聚合函數包括:「sum」、“max”、“min” 和 “mean”。
“Max”
“Mean”
“Min”
"None"
“Sum”
targetLags 要從目標數據行延隔的過去期間數。 TargetLags
targetRollingWindowSize 用來建立目標數據行滾動視窗平均值的過去期間數目。 TargetRollingWindowSize
timeColumnName 時間數據行的名稱。 當預測指定用於建置時間序列和推斷其頻率之輸入數據中的 datetime 數據行時,需要此參數。 字串
timeSeriesIdColumnNames 用來分組時間範圍的數據行名稱。 它可以用來建立多個數列。
如果未定義粒紋,則會假設資料集為一個時間序列。 此參數會與工作類型預測搭配使用。
string[]
useStl 設定時間序列目標數據行的 STL 分解。 "None"
“季”
“SeasonTrend”

ForecastHorizon

名稱 描述
mode 設定物件類型 Auto
需要自訂 ()

AutoForecastHorizon

名稱 描述
mode [必要]設定預測層級值選取模式。 “Auto” (必要)

CustomForecastHorizon

名稱 描述
mode [必要]設定預測層級值選取模式。 「自定義」 (必要)
[必要]預測層級值。 需要 int ()

季節性

名稱 描述
mode 設定物件類型 Auto
需要自訂 ()

AutoSeasonality

名稱 描述
mode [必要]季節性模式。 “Auto” (必要)

CustomSeasonality

名稱 描述
mode [必要]季節性模式。 「自定義」 (必要)
[必要]季節性值。 需要 int ()

TargetLags

名稱 描述
mode 設定物件類型 Auto
需要自訂 ()

AutoTargetLags

名稱 描述
mode [必要]設定目標延隔模式 - 自動/自定義 “Auto” (必要)

CustomTargetLags

名稱 描述
mode [必要]設定目標延隔模式 - 自動/自定義 「自定義」 (必要)
[必要]設定目標延隔值。 int[] (必要)

TargetRollingWindowSize

名稱 描述
mode 設定物件類型 Auto
需要自訂 ()

AutoTargetRollingWindowSize

名稱 描述
mode [必要]TargetRollingWindowSiz 偵測模式。 “Auto” (必要)

CustomTargetRollingWindowSize

名稱 描述
mode [必要]TargetRollingWindowSiz 偵測模式。 「自定義」 (必要)
[必要]TargetRollingWindowSize 值。 需要 int ()

ImageClassification

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 “ImageClassification” (必要)
dataSettings [必要]已註冊表格式數據集標識碼的集合,以及定型和驗證模型所需的其他數據設定。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsClassification
primaryMetric 要針對這項工作優化的主要計量。 “AUCWeighted”
“精確度”
“AveragePrecisionScoreWeighted”
“NormMacroRecall”
“PrecisionScoreWeighted”
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsClassification[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

ImageVerticalDataSettings

名稱 描述
targetColumnName [必要]目標數據行名稱:這是預測值數據行。
也稱為分類工作內容中的標籤數據行名稱。
需要字串 ()

約束:
模式 = [a-zA-Z0-9_]
testData 測試數據輸入。 TestDataSettings
trainingData [必要]定型數據輸入。 TrainingDataSettings (必要)
validationData 驗證數據集的設定。 ImageVerticalValidationDataSettings

ImageVerticalValidationDataSettings

名稱 描述
資料 驗證數據 MLTable。 MLTableJobInput
validationDataSize 需要為驗證目的保留的定型數據集分數。
(0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

ImageLimitSettings

名稱 描述
maxConcurrentTrials 並行 AutoML 反覆項目的數目上限。 int
maxTrials AutoML 反覆項目的數目上限。 int
timeout AutoML 作業逾時。 字串

ImageModelSettingsClassification

名稱 描述
advancedSettings 進階案例的設定。 字串
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 bool
擴增 使用擴增的設定。 字串
beta1 當優化器為 'adam' 或 'adamw' 時,'beta1' 的值。 必須是範圍 [0, 1] 內的浮點數。 int
beta2 當優化器為 'adam' 或 'adamw' 時,'beta2' 的值。 必須是範圍 [0, 1] 內的浮點數。 int
checkpointDatasetId 預先定型檢查點的 FileDataset 識別碼, (累加式定型的) 。
請務必傳遞 CheckpointFilename 以及 CheckpointDatasetId。
字串
checkpointFilename FileDataset 中用於累加式定型的預先定型檢查點檔案名稱。
請務必傳遞 CheckpointDatasetId 以及 CheckpointFilename。
字串
checkpointFrequency 儲存模型檢查點的頻率。 必須是正整數。 int
checkpointRunId 先前執行之標識碼,其具有累加式定型的預先定型檢查點。 字串
分散式 是否要使用分散式定型。 bool
earlyStopping 在定型期間啟用提早停止邏輯。 bool
earlyStoppingDelay 在主要計量改進之前等待的最小 Epoch 或驗證評估數目
會追蹤提早停止。 必須是正整數。
int
earlyStoppingPatience 在之前沒有主要計量改善的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
int
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 bool
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 int
gradientAccumulationStep 漸層累積表示執行未設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
int
layersToFreeze 要凍結模型的層數。 必須是正整數。
例如,傳遞 2 作為 'seresnext' 值表示
凍結 layer0 和 layer1。 如需所支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
int
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 int
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 "None"
“步驟”
“WarmupCosine”
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請造訪官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
動量 優化工具為 『sgd』 時的動向值。 必須是範圍 [0, 1] 內的浮點數。 int
nesterov 當優化器為 『sgd』 時啟用 nesterov。 bool
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 int
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 int
optimizer 最佳化工具的類型。 “Adam”
“Adamw”
"None"
“Sgd”
randomSeed 使用決定性定型時要使用的隨機種子。 int
splitRatio 如果未定義驗證數據,這會指定分割的分割比例
將數據定型為隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
int
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 int
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 int
trainingBatchSize 定型批次大小。 必須是正整數。 int
trainingCropSize 影像裁剪大小,為定型數據集的類神經網路輸入。 必須是正整數。 int
validationBatchSize 驗證批次大小。 必須是正整數。 int
validationCropSize 輸入至驗證數據集之類神經網路的影像裁剪大小。 必須是正整數。 int
validationResizeSize 在針對驗證資料集進行裁切之前,要調整大小的影像大小。 必須是正整數。 int
warmupCosineLRCycles 當學習速率排程器為 『warmup_cosine』 時,餘弦迴圈的值。 必須是範圍 [0, 1] 內的浮點數。 int
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』時,熱身 Epoch 的值。 必須是正整數。 int
weightDecay 優化器為 'sgd'、'adam' 或 'adamw' 時的權數衰減值。 必須是範圍[0, 1] 中的浮點數。 int
weightedLoss 加權損失。 接受的值是 0,不會遺失加權。
1 用於 sqrt. (class_weights) 加權損失。 2 表示具有 class_weights 的加權遺失。 必須是 0 或是 1 或 2。
int

ImageModelDistributionSettingsClassification

名稱 描述
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 字串
擴增 使用擴增的設定。 字串
beta1 優化器為 'adam' 或 'adamw' 時的 'beta1' 值。 必須是範圍 [0, 1] 內的浮點數。 字串
beta2 優化器為 'adam' 或 'adamw' 時的 'beta2' 值。 必須是範圍 [0, 1] 內的浮點數。 字串
分散式 是否要使用分散式程式定型。 字串
earlyStopping 在定型期間啟用提早停止邏輯。 字串
earlyStoppingDelay 在主要計量改進之前等候的最小 Epoch 或驗證評估數目
會追蹤早期停止。 必須是正整數。
字串
earlyStoppingPatience 之前沒有主要計量改進的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
字串
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 字串
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 字串
gradientAccumulationStep 漸層累積表示在沒有的情況下執行已設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
字串
layersToFreeze 要凍結模型的圖層數目。 必須是正整數。
例如,傳遞 2 做為 'seresnext' 值表示
凍結第0層和第1層。 如需支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 字串
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 字串
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請流覽官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
動量 優化工具為 『sgd』 時的時刻值。 必須是範圍 [0, 1] 內的浮點數。 字串
nesterov 當優化器為 'sgd' 時,請啟用 nesterov。 字串
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 字串
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 字串
optimizer 最佳化工具的類型。 必須是 『sgd』、『adam』或 『adamw』。 字串
randomSeed 使用決定性定型時要使用的隨機種子。 字串
splitRatio 如果未定義驗證數據,這會指定分割的分割比率
將數據定型成隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
字串
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 字串
trainingBatchSize 定型批次大小。 必須是正整數。 字串
trainingCropSize 影像裁剪大小,為定型數據集的類神經網路輸入。 必須是正整數。 字串
validationBatchSize 驗證批次大小。 必須是正整數。 字串
validationCropSize 輸入至驗證數據集之類神經網路的影像裁剪大小。 必須是正整數。 字串
validationResizeSize 在針對驗證資料集進行裁切之前,要調整大小的影像大小。 必須是正整數。 字串
warmupCosineLRCycles 當學習速率排程器為 『warmup_cosine』 時,餘弦迴圈的值。 必須是範圍 [0, 1] 內的浮點數。 字串
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』時,熱身 Epoch 的值。 必須是正整數。 字串
weightDecay 優化器為 'sgd'、'adam' 或 'adamw' 時的權數衰減值。 必須是範圍[0, 1] 中的浮點數。 字串
weightedLoss 加權損失。 接受的值是 0,不會遺失加權。
1 用於 sqrt. (class_weights) 加權損失。 2 表示具有 class_weights 的加權遺失。 必須是 0 或是 1 或 2。
字串

ImageSweepSettings

名稱 描述
earlyTermination 早期終止原則的類型。 EarlyTerminationPolicy
限制 [必要]限制模型掃掠和超參數掃掠的設定。 ImageSweepLimitSettings (必要)
samplingAlgorithm [必要]超參數取樣演算法的類型。 “Bayesian”
“Grid”
「隨機」 (必要)

EarlyTerminationPolicy

名稱 描述
delayEvaluation 延遲第一次評估的間隔數目。 int
evaluationInterval 原則評估之間) 執行的間隔 (數目。 int
policyType 設定物件類型 Bandit
MedianStopping
截斷選取 必要 ()

BanditPolicy

名稱 描述
policyType [必要]原則設定的名稱 “Bandit” (必要)
slackAmount 最佳執行所允許的絕對距離。 int
slackFactor 允許與最佳執行距離的比例。 int

MedianStoppingPolicy

名稱 描述
policyType [必要]原則設定的名稱 “MedianStopping” (必要)

TruncationSelectionPolicy

名稱 描述
policyType [必要]原則設定的名稱 “TruncationSelection” (必要)
truncationPercentage 要在每個評估間隔取消的執行百分比。 int

ImageSweepLimitSettings

名稱 描述
maxConcurrentTrials 基礎掃掠作業的並行反覆項目數目上限。 int
maxTrials 基礎掃掠作業的反覆項目數目上限。 int

ImageClassificationMultilabel

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 “ImageClassificationMultilabel” (必要)
dataSettings [必要]已註冊表格式數據集標識碼的集合,以及定型和驗證模型所需的其他數據設定。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsClassification
primaryMetric 要針對這項工作優化的主要計量。 “AUCWeighted”
“精確度”
“AveragePrecisionScoreWeighted”
“IOU”
“NormMacroRecall”
“PrecisionScoreWeighted”
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsClassification[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

ImageInstanceSegmentation

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 “ImageInstanceSegmentation” (必要)
dataSettings [必要]已註冊表格式數據集標識碼的集合,以及定型和驗證模型所需的其他數據設定。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsObjectDetection
primaryMetric 要針對這項工作優化的主要計量。 “MeanAveragePrecision”
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsObjectDetection[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

ImageModelSettingsObjectDetection

名稱 描述
advancedSettings 進階案例的設定。 字串
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 bool
擴增 使用擴增的設定。 字串
beta1 當優化器為 'adam' 或 'adamw' 時,'beta1' 的值。 必須是範圍 [0, 1] 內的浮點數。 int
beta2 當優化器為 'adam' 或 'adamw' 時,'beta2' 的值。 必須是範圍 [0, 1] 內的浮點數。 int
boxDetectionsPerImage 針對所有類別,每個影像的偵測數目上限。 必須是正整數。
注意:'yolov5' 演算法不支援此設定。
int
boxScoreThreshold 在推斷期間,只傳回分類分數大於的建議
BoxScoreThreshold。 必須是 range[0, 1] 中的浮點數。
int
checkpointDatasetId 預先定型檢查點的 FileDataset 識別碼, (累加式定型的) 。
請務必傳遞 CheckpointFilename 以及 CheckpointDatasetId。
字串
checkpointFilename FileDataset 中用於累加式定型的預先定型檢查點檔案名稱。
請務必傳遞 CheckpointDatasetId 以及 CheckpointFilename。
字串
checkpointFrequency 儲存模型檢查點的頻率。 必須是正整數。 int
checkpointRunId 先前執行之標識碼,其具有累加式定型的預先定型檢查點。 字串
分散式 是否要使用分散式定型。 bool
earlyStopping 在定型期間啟用提早停止邏輯。 bool
earlyStoppingDelay 在主要計量改進之前等待的最小 Epoch 或驗證評估數目
會追蹤提早停止。 必須是正整數。
int
earlyStoppingPatience 在之前沒有主要計量改善的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
int
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 bool
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 int
gradientAccumulationStep 漸層累積表示執行未設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
int
imageSize 用於定型和驗證的影像大小。 必須是正整數。
注意:如果大小太大,訓練回合可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
int
layersToFreeze 要凍結模型的層數。 必須是正整數。
例如,傳遞 2 作為 'seresnext' 值表示
凍結 layer0 和 layer1。 如需所支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
int
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 int
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 "None"
“Step”
“WarmupCosine”
maxSize 將影像饋送至骨幹之前要重新調整的影像大小上限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
int
minSize 將影像饋送至骨幹之前要重新調整的影像大小下限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
int
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請流覽官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
modelSize 模型大小。 必須是 'small'、'medium'、'large' 或 'xlarge'。
注意:如果模型大小太大,則定型執行可能會進入 CUDA OOM.
注意:只有 'yolov5' 演算法才支援此設定。
“ExtraLarge”
“Large”
“Medium”
"None"
“Small”
動量 優化工具為 『sgd』 時的時刻值。 必須是範圍 [0, 1] 內的浮點數。 int
尺度 依 +/- 50% 的不同影像大小啟用多縮放映射。
注意:如果沒有足夠的 GPU 記憶體,則定型執行可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
bool
nesterov 當優化器為 'sgd' 時,請啟用 nesterov。 bool
nmsIouThreshold 在NMS後置處理中推斷期間使用的IOU臨界值。 必須是範圍 [0, 1] 內的浮點數。 int
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 int
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 int
optimizer 最佳化工具的類型。 “Adam”
“Adamw”
"None"
“Sgd”
randomSeed 使用決定性定型時要使用的隨機種子。 int
splitRatio 如果未定義驗證數據,這會指定分割的分割比率
將數據定型成隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
int
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 int
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 int
tileGridSize 要用於每個並排影像的格線大小。 注意:TileGridSize 不得為
無可啟用小型物件偵測邏輯。 包含兩個整數的字串,其格式為#。
注意:'yolov5' 演算法不支援此設定。
字串
tileOverlapRatio 每個維度中相鄰圖格之間的重疊比例。 範圍 [0, 1) 必須是 float。
注意:'yolov5' 演算法不支援此設定。
int
tilePredictionsNmsThreshold 從圖格和影像合併預測時用來執行 NMS 的 IOU 閾值。
用於驗證/推斷。 必須是範圍 [0, 1] 中的 float。
注意:'yolov5' 演算法不支援此設定。
int
trainingBatchSize 定型批次大小。 必須是正整數。 int
validationBatchSize 驗證批次大小。 必須是正整數。 int
validationIouThreshold 計算驗證計量時要使用的IOU臨界值。 必須是範圍 [0, 1] 中的浮點數。 int
validationMetricType 要用於驗證計量的計量計算方法。 “Coco”
“CocoVoc”
"None"
“Voc”
warmupCosineLRCycles 學習速率排程器為「warmup_cosine」時的餘弦週期值。 必須是範圍 [0, 1] 內的浮點數。 int
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』 時,暖 Epoch 的值。 必須是正整數。 int
weightDecay 當優化器為 'sgd'、'adam' 或 'adamw' 時,加權衰減的值。 必須是 range[0, 1] 中的浮點數。 int

ImageModelDistributionSettingsObjectDetection

名稱 描述
amsGradient 當優化器為 『adam』 或 'adamw' 時啟用AMSGrad。 字串
擴增 使用擴增的設定。 字串
beta1 當優化器為 'adam' 或 'adamw' 時,'beta1' 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
beta2 當優化器為 'adam' 或 'adamw' 時,'beta2' 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
boxDetectionsPerImage 針對所有類別,每個影像的偵測數目上限。 必須是正整數。
注意:'yolov5' 演算法不支援此設定。
字串
boxScoreThreshold 在推斷期間,只傳回分類分數大於的建議
BoxScoreThreshold。 必須是 range[0, 1] 中的浮點數。
字串
分散式 是否要使用散發器定型。 字串
earlyStopping 在定型期間啟用提早停止邏輯。 字串
earlyStoppingDelay 在主要計量改進之前等待的最小 Epoch 或驗證評估數目
會追蹤提早停止。 必須是正整數。
字串
earlyStoppingPatience 在之前沒有主要計量改善的最小 Epoch 或驗證評估數目
執行已停止。 必須是正整數。
字串
enableOnnxNormalization 匯出 ONNX 模型時啟用正規化。 字串
evaluationFrequency 評估驗證資料集以取得計量分數的頻率。 必須是正整數。 字串
gradientAccumulationStep 漸層累積表示執行未設定數目的 「GradAccumulationStep」 步驟
更新模型權數,同時累積這些步驟的漸層,然後使用
用來計算權數更新的累積漸層。 必須是正整數。
字串
imageSize 用於定型和驗證的影像大小。 必須是正整數。
注意:如果大小太大,訓練回合可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
字串
layersToFreeze 要凍結模型的層數。 必須是正整數。
例如,傳遞 2 作為 'seresnext' 值表示
凍結 layer0 和 layer1。 如需所支援模型的完整清單,以及圖層凍結的詳細數據,請
看到: https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
learningRate 初始學習速率。 必須是範圍 [0, 1] 內的浮點數。 字串
learningRateScheduler 學習率排程器的類型。 必須是 『warmup_cosine』 或 'step'。 字串
maxSize 將影像饋送至骨幹之前要重新調整的影像大小上限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
字串
minSize 將影像饋送至骨幹之前要重新調整的影像大小下限。
必須是正整數。 注意:如果大小太大,則定型執行可能會進入 CUDA OOM.
注意:'yolov5' 演算法不支援此設定。
字串
modelName 要用於定型的模型名稱。
如需可用模型的詳細資訊,請造訪官方檔:
https://docs.microsoft.com/azure/machine-learning/how-to-auto-train-image-models.
字串
modelSize 模型大小。 必須是 'small'、'medium'、'large' 或 'xlarge'。
注意:如果模型大小太大,則定型執行可能會進入 CUDA OOM.
注意:只有 'yolov5' 演算法才支援此設定。
字串
動量 優化工具為 『sgd』 時的動向值。 必須是範圍 [0, 1] 內的浮點數。 字串
尺度 以 +/- 50% 的不同影像大小啟用多縮放影像。
注意:如果沒有足夠的 GPU 記憶體,則定型執行可能會進入 CUDA OOM。
注意:只有 'yolov5' 演算法才支援此設定。
字串
nesterov 當優化器為 『sgd』 時啟用 nesterov。 字串
nmsIouThreshold 在NMS後置處理中的推斷期間使用的IOU臨界值。 必須是範圍 [0, 1] 中的浮點數。 字串
numberOfEpochs 定型 Epoch 的數目。 必須是正整數。 字串
numberOfWorkers 數據載入器背景工作角色的數目。 必須是非負整數。 字串
optimizer 最佳化工具的類型。 必須是 'sgd'、'adam' 或 'adamw'。 字串
randomSeed 使用決定性定型時要使用的隨機種子。 字串
splitRatio 如果未定義驗證數據,這會指定分割的分割比例
將數據定型為隨機定型和驗證子集。 必須是範圍 [0, 1] 內的浮點數。
字串
stepLRGamma 當學習速率排程器為 『step』 時,gamma 的值。 必須是範圍 [0, 1] 內的浮點數。 字串
stepLRStepSize 當學習速率排程器為 『step』 時,步驟大小的值。 必須是正整數。 字串
tileGridSize 要用於每個並排影像的格線大小。 注意:TileGridSize 不得為
無可啟用小型物件偵測邏輯。 字串,包含兩個整數,格式為一個。
注意:'yolov5' 演算法不支援此設定。
字串
tileOverlapRatio 每個維度中相鄰圖格之間的重疊比例。 必須是範圍 [0, 1) 的 float。
注意:'yolov5' 演算法不支援此設定。
字串
tilePredictionsNmsThreshold 從圖格和影像合併預測時用來執行 NMS 的 IOU 閾值。
用於驗證/推斷。 必須是範圍 [0, 1] 中的浮點數。
注意:'yolov5' 演算法不支援此設定。
NMS:非最大歸併
字串
trainingBatchSize 定型批次大小。 必須是正整數。 字串
validationBatchSize 驗證批次大小。 必須是正整數。 字串
validationIouThreshold 計算驗證計量時要使用的IOU臨界值。 必須是範圍 [0, 1] 中的 float。 字串
validationMetricType 要用於驗證計量的計量計算方法。 必須是 『none』、'coco'、'voc' 或 'coco_voc'。 字串
warmupCosineLRCycles 當學習速率排程器為 『warmup_cosine』 時,餘弦迴圈的值。 必須是範圍 [0, 1] 內的浮點數。 字串
warmupCosineLRWarmupEpochs 當學習速率排程器為 『warmup_cosine』時,熱身 Epoch 的值。 必須是正整數。 字串
weightDecay 優化器為 'sgd'、'adam' 或 'adamw' 時的權數衰減值。 必須是範圍[0, 1] 中的浮點數。 字串

ImageObjectDetection

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 “ImageObjectDetection” (必要)
dataSettings [必要]註冊的表格式數據集標識碼和其他定型和驗證模型所需的數據設定集合。 ImageVerticalDataSettings (必要)
limitSettings [必要]限制 AutoML 作業的設定。 ImageLimitSettings (必要)
modelSettings 用於定型模型的設定。 ImageModelSettingsObjectDetection
primaryMetric 要針對這項工作優化的主要計量。 “MeanAveragePrecision”
searchSpace 搜尋空間,以取樣模型及其超參數的不同組合。 ImageModelDistributionSettingsObjectDetection[]
sweepSettings 模型掃掠和超參數掃掠相關設定。 ImageSweepSettings

迴歸

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 必要) 的「回歸」 (
allowedModels 回歸工作的允許模型。 包含任何項目的字串數組:
“DecisionTree”
“ElasticNet”
“ExtremeRandomTrees”
“GradientBoosting”
“KNN”
“LassoLars”
“LightGBM”
“RandomForest”
“SGD”
“XGBoostRegressor”
blockedModels 回歸工作的封鎖模型。 包含任何項目的字串數組:
“DecisionTree”
“ElasticNet”
“ExtremeRandomTrees”
“GradientBoosting”
“KNN”
“LassoLars”
“LightGBM”
“RandomForest”
“SGD”
“XGBoostRegressor”
dataSettings AutoMLJob 的數據輸入。 TableVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 TableVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 TableVerticalLimitSettings
primaryMetric 回歸工作的主要計量。 “NormalizedMeanAbsoluteError”
“NormalizedRootMeanSquaredError”
“R2Score”
“SpearmanCorrelation”
trainingSettings AutoML 作業定型階段的輸入。 TrainingSettings

TextClassification

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 “TextClassification” (必要)
dataSettings AutoMLJob 的數據輸入。 NlpVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 NlpVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 NlpVerticalLimitSettings
primaryMetric Text-Classification 工作的主要計量。 “AUCWeighted”
「精確度」
“AveragePrecisionScoreWeighted”
“NormMacroRecall”
“PrecisionScoreWeighted”

NlpVerticalDataSettings

名稱 描述
targetColumnName [必要]目標數據行名稱:這是預測值數據行。
也稱為分類工作內容中的標籤數據行名稱。
字串 (必要)

約束:
模式 = [a-zA-Z0-9_]
testData 測試數據輸入。 TestDataSettings
trainingData [必要]定型數據輸入。 TrainingDataSettings (必要)
validationData 驗證數據輸入。 NlpVerticalValidationDataSettings

NlpVerticalValidationDataSettings

名稱 描述
資料 驗證數據 MLTable。 MLTableJobInput
validationDataSize 必須針對驗證目的將定型數據集的分數。
介於 (0.0 、1.0) 之間的值
未提供驗證數據集時套用。
int

NlpVerticalFeaturizationSettings

名稱 描述
datasetLanguage 數據集語言,適用於文字數據。 字串

NlpVerticalLimitSettings

名稱 描述
maxConcurrentTrials 並行 AutoML 反覆專案上限。 int
maxTrials AutoML 反覆項目的數目。 int
timeout AutoML 作業逾時。 字串

TextClassificationMultilabel

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 “TextClassificationMultilabel” (必要)
dataSettings AutoMLJob 的數據輸入。 NlpVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 NlpVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 NlpVerticalLimitSettings

TextNer

名稱 描述
taskType [必要]AutoMLJob 的工作類型。 必要) “TextNER” (
dataSettings AutoMLJob 的數據輸入。 NlpVerticalDataSettings
featurizationSettings AutoML 作業所需的特徵化輸入。 NlpVerticalFeaturizationSettings
limitSettings AutoMLJob 的執行條件約束。 NlpVerticalLimitSettings

CommandJob

名稱 描述
jobType [必要]指定作業的類型。 “Command” (必要)
codeId 程式代碼資產的 ARM 資源識別碼。 字串
命令 [必要]在作業啟動時執行的命令。 例如 “python train.py” 字串 (必要)

約束:
最小長度 = 1
模式 = [a-zA-Z0-9_]
散發 (distribution) 作業的散發組態。 如果設定,這應該是 Mpi、Tensorflow、PyTorch 或 null 的其中一個。 DistributionConfiguration
environmentId [必要]作業環境規格的 ARM 資源識別碼。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]
environmentVariables 作業中包含的環境變數。 CommandJobEnvironmentVariables
輸入 對應作業中使用的輸入數據系結。 CommandJobInputs
限制 命令作業限制。 CommandJobLimits
outputs 對應作業中使用的輸出數據系結。 CommandJobOutputs
resources 作業的計算資源組態。 ResourceConfiguration

DistributionConfiguration

名稱 描述
distributionType 設定物件類型 Mpi
PyTorch
TensorFlow (必要)

Mpi

名稱 描述
distributionType [必要]指定散發架構的類型。 “Mpi” (必要)
processCountPerInstance 每個 MPI 節點的進程數目。 int

PyTorch

名稱 描述
distributionType [必要]指定散發架構的類型。 “PyTorch” (必要)
processCountPerInstance 每個節點的進程數目。 int

TensorFlow

名稱 描述
distributionType [必要]指定散發架構的類型。 必要) 「TensorFlow」 (
parameterServerCount 參數伺服器工作的數目。 int
workerCount 背景工作數目。 如果未指定,則會預設為實例計數。 int

CommandJobEnvironmentVariables

名稱 描述
{自定義屬性} 字串

CommandJobInputs

名稱 描述
{自定義屬性} JobInput

JobInput

名稱 描述
description 輸入的描述。 字串
jobInputType 設定物件類型 CustomModel
文字
MLFlowModel
MLTable
TritonModel
UriFile
需要 UriFolder ()

CustomModelJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 “CustomModel” (必要)
mode 輸入資產傳遞模式。 “Direct”
“下載”
“EvalDownload”
“EvalMount”
“ReadOnlyMount”
“ReadWriteMount”
uri [必要]輸入資產 URI。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]

LiteralJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 “Literal” (必要)
[必要]輸入的常值。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]

MLFlowModelJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 “MLFlowModel” (必要)
mode 輸入資產傳遞模式。 “Direct”
“下載”
“EvalDownload”
“EvalMount”
“ReadOnlyMount”
“ReadWriteMount”
uri [必要]輸入資產 URI。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]

TritonModelJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 “TritonModel” (必要)
mode 輸入資產傳遞模式。 “Direct”
“下載”
“EvalDownload”
“EvalMount”
“ReadOnlyMount”
“ReadWriteMount”
uri [必要]輸入資產 URI。 需要字串 ()

約束:
模式 = [a-zA-Z0-9_]

UriFileJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 “UriFile” (必要)
mode 輸入資產傳遞模式。 “Direct”
“下載”
“EvalDownload”
“EvalMount”
“ReadOnlyMount”
“ReadWriteMount”
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

UriFolderJobInput

名稱 描述
jobInputType [必要]指定作業的類型。 必要) “UriFolder” (
mode 輸入資產傳遞模式。 “Direct”
“下載”
“EvalDownload”
“EvalMount”
“ReadOnlyMount”
“ReadWriteMount”
uri [必要]輸入資產 URI。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

CommandJobLimits

名稱 描述
jobLimitsType [必要]JobLimit 類型。 “Command”
「掃掠」 (必要)
timeout ISO 8601 格式的最大執行持續時間,之後作業將會取消。 僅支援有效位數為秒數的持續時間。 字串

CommandJobOutputs

名稱 描述
{自定義屬性} JobOutput

PipelineJob

名稱 描述
jobType [必要]指定作業的類型。 「管線」 (必要)
輸入 管線作業的輸入。 PipelineJobInputs
jobs 作業會建構管線作業。 PipelineJobJobs
outputs 管線作業的輸出 PipelineJobOutputs
設定 管線設定,例如 ContinueRunOnStepFailure 等專案。

PipelineJobInputs

名稱 描述
{自定義屬性} JobInput

PipelineJobJobs

名稱 描述
{自定義屬性}

PipelineJobOutputs

名稱 描述
{自定義屬性} JobOutput

SweepJob

名稱 描述
jobType [必要]指定作業的類型。 「掃掠」 (必要)
earlyTermination 早期終止原則可在完成之前取消效能不佳的執行 EarlyTerminationPolicy
輸入 對應作業中使用的輸入數據系結。 SweepJobInputs
限制 掃掠作業限制。 SweepJobLimits
objective [必要]優化目標。 必要 目標 ()
outputs 對應作業中使用的輸出數據系結。 SweepJobOutputs
samplingAlgorithm [必要]超參數取樣演算法 SamplingAlgorithm (必要)
searchSpace [必要]包含每個參數及其分佈的字典。 字典索引鍵是參數的名稱
試用版 [必要]試用版元件定義。 必要) 的 TrialComponent (

SweepJobInputs

名稱 描述
{自定義屬性} JobInput

SweepJobLimits

名稱 描述
jobLimitsType [必要]JobLimit 類型。 “Command”
「掃掠」 (必要)
maxConcurrentTrials 掃掠作業最大並行試用版。 int
maxTotalTrials 掃掠作業總試用版數上限。 int
timeout ISO 8601 格式的最大執行持續時間,之後作業將會取消。 僅支援有效位數為秒數的持續時間。 字串
trialTimeout 掃掠作業試用逾時值。 字串

目標

名稱 描述
goal [必要]定義超參數微調支援的計量目標 “最大化”
必要) 「最小化」 (
primaryMetric [必要]要優化之計量的名稱。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]

SweepJobOutputs

名稱 描述
{自定義屬性} JobOutput

SamplingAlgorithm

名稱 描述
samplingAlgorithmType 設定物件類型 貝 葉 斯
方格
需要隨機 ()

BayesianSamplingAlgorithm

名稱 描述
samplingAlgorithmType [必要]用來產生超參數值的演算法,以及組態屬性 「貝氏」 (必要)

GridSamplingAlgorithm

名稱 描述
samplingAlgorithmType [必要]用來產生超參數值的演算法,以及組態屬性 「網格線」 (必要)

RandomSamplingAlgorithm

名稱 描述
samplingAlgorithmType [必要]用來產生超參數值的演算法,以及組態屬性 「隨機」 (必要)
rule (規則) 隨機演算法的特定類型 “Random”
“Sobol”
seed 要作為隨機數產生種子的選擇性整數 int

TrialComponent

名稱 描述
codeId 程式代碼資產的 ARM 資源識別碼。 字串
命令 [必要]在作業啟動時執行的命令。 例如 “python train.py” 字串 (必要)

約束:
最小長度 = 1
模式 = [a-zA-Z0-9_]
散發 (distribution) 作業的散發組態。 如果設定,這應該是 Mpi、Tensorflow、PyTorch 或 null 的其中一個。 DistributionConfiguration
environmentId [必要]作業環境規格的 ARM 資源識別碼。 字串 (必要)

約束:
模式 = [a-zA-Z0-9_]
environmentVariables 作業中包含的環境變數。 TrialComponentEnvironmentVariables
resources 作業的計算資源組態。 ResourceConfiguration

TrialComponentEnvironmentVariables

名稱 描述
{自定義屬性} 字串