Microsoft.MachineLearningServices ワークスペース/ジョブ 2022-05-01
Bicep リソース定義
ワークスペース/ジョブ リソースの種類は、次をターゲットとする操作でデプロイできます。
- リソース グループ - リソース グループのデプロイ コマンドを参照してください
各 API バージョンの変更されたプロパティの一覧については、「 変更ログ」を参照してください。
リソース形式
Microsoft.MachineLearningServices/workspaces/jobs リソースを作成するには、次の Bicep をテンプレートに追加します。
resource symbolicname 'Microsoft.MachineLearningServices/workspaces/jobs@2022-05-01' = {
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: {}
services: {}
tags: {}
jobType: 'string'
// For remaining properties, see JobBaseProperties objects
}
}
JobBaseProperties オブジェクト
オブジェクトの種類を指定するには、 jobType プロパティを設定します。
Command の場合は、次のコマンドを使用します。
jobType: 'Command'
codeId: 'string'
command: 'string'
distribution: {
distributionType: 'string'
// For remaining properties, see DistributionConfiguration objects
}
environmentId: 'string'
environmentVariables: {}
inputs: {}
limits: {
jobLimitsType: 'string'
timeout: 'string'
}
outputs: {}
resources: {
instanceCount: int
instanceType: 'string'
properties: {}
}
パイプラインの場合は、次を使用します。
jobType: 'Pipeline'
inputs: {}
jobs: {}
outputs: {}
settings: any()
スイープの場合は、次の値を使用します。
jobType: 'Sweep'
earlyTermination: {
delayEvaluation: int
evaluationInterval: int
policyType: 'string'
// For remaining properties, see EarlyTerminationPolicy objects
}
inputs: {}
limits: {
jobLimitsType: 'string'
maxConcurrentTrials: int
maxTotalTrials: int
timeout: 'string'
trialTimeout: 'string'
}
objective: {
goal: 'string'
primaryMetric: 'string'
}
outputs: {}
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: {}
resources: {
instanceCount: int
instanceType: 'string'
properties: {}
}
}
IdentityConfiguration オブジェクト
identityType プロパティを設定して、オブジェクトの種類を指定します。
AMLToken の場合は、次の値を使用します。
identityType: 'AMLToken'
Managed の場合は、次の値を使用します。
identityType: 'Managed'
clientId: 'string'
objectId: 'string'
resourceId: 'string'
UserIdentity の場合は、次の値を使用します。
identityType: 'UserIdentity'
DistributionConfiguration オブジェクト
distributionType プロパティを設定して、オブジェクトの種類を指定します。
Mpi の場合は、次の値を使用します。
distributionType: 'Mpi'
processCountPerInstance: int
PyTorch の場合は、次の値を使用します。
distributionType: 'PyTorch'
processCountPerInstance: int
TensorFlow の場合は、次の値を使用します。
distributionType: 'TensorFlow'
parameterServerCount: int
workerCount: int
EarlyTerminationPolicy オブジェクト
policyType プロパティを設定して、オブジェクトの種類を指定します。
バンディットの場合は、次の値を使用します。
policyType: 'Bandit'
slackAmount: int
slackFactor: int
MedianStopping の場合は、次の値を使用します。
policyType: 'MedianStopping'
TruncationSelection の場合は、次を使用します。
policyType: 'TruncationSelection'
truncationPercentage: int
SamplingAlgorithm オブジェクト
オブジェクトの種類を指定するには、 samplingAlgorithmType プロパティを設定します。
ベイジアンの場合は、次の値を使用します。
samplingAlgorithmType: 'Bayesian'
Grid の場合は、次の値を使用します。
samplingAlgorithmType: 'Grid'
Random の場合は、次の値を使用します。
samplingAlgorithmType: 'Random'
rule: 'string'
seed: int
プロパティ値
workspaces/jobs
名前 | 説明 | 値 |
---|---|---|
name | リソース名 Bicep で子リソースの名前と型を設定する方法を参照してください。 |
string (必須) |
parent | Bicep では、子リソースの親リソースを指定できます。 このプロパティを追加する必要があるのは、子リソースが親リソースの外部で宣言されている場合のみです。 詳細については、「 親リソースの外部の子リソース」を参照してください。 |
種類のリソースのシンボル名: ワークスペース |
properties | [必須]エンティティの追加の属性。 | JobBaseProperties (必須) |
JobBaseProperties
名前 | 説明 | 値 |
---|---|---|
computeId | コンピューティング リソースの ARM リソース ID。 | string |
description | 資産の説明テキスト。 | string |
displayName | ジョブの表示名。 | string |
experimentName | ジョブが属する実験の名前。 設定されていない場合、ジョブは "Default" 実験に配置されます。 | string |
identity | ID の構成。 設定されている場合、これは AmlToken、ManagedIdentity、UserIdentity、または null のいずれかである必要があります。 null の場合、既定値は AmlToken です。 |
IdentityConfiguration |
isArchived | 資産はアーカイブされていますか? | [bool] |
properties | 資産プロパティディクショナリ。 | object |
services | JobEndpoints の一覧。 ローカル ジョブの場合、ジョブ エンドポイントのエンドポイント値は FileStreamObject になります。 |
object |
tags | タグ辞書。 タグは追加、削除、更新できます。 | object |
jobType | オブジェクトの種類を設定する | コマンド パイプライン スイープ (必須) |
IdentityConfiguration
名前 | 説明 | 値 |
---|---|---|
identityType | オブジェクトの種類を設定する | AMLToken マネージド UserIdentity (必須) |
AmlToken
名前 | 説明 | 値 |
---|---|---|
identityType | [必須]ID フレームワークの種類を指定します。 | 'AMLToken' (必須) |
ManagedIdentity
名前 | 説明 | 値 |
---|---|---|
identityType | [必須]ID フレームワークの種類を指定します。 | 'Managed' (必須) |
clientId | クライアント ID でユーザー割り当て ID を指定します。 システム割り当ての場合は、このフィールドを設定しないでください。 | string 制約: 最小長 = 36 最大長 = 36 Pattern = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$ |
objectId | オブジェクト ID でユーザー割り当て ID を指定します。 システム割り当ての場合は、このフィールドを設定しないでください。 | string 制約: 最小長 = 36 最大長 = 36 Pattern = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$ |
resourceId | ARM リソース ID でユーザー割り当て ID を指定します。 システム割り当ての場合は、このフィールドを設定しないでください。 | string |
UserIdentity
名前 | 説明 | 値 |
---|---|---|
identityType | [必須]ID フレームワークの種類を指定します。 | 'UserIdentity' (必須) |
CommandJob
名前 | 説明 | 値 |
---|---|---|
jobType | [必須]ジョブの種類を指定します。 | 'Command' (必須) |
codeId | コード資産の ARM リソース ID。 | string |
command | [必須]ジョブの起動時に実行するコマンド。 例: "python train.py" | string (必須) 制約: 最小長 = 1 Pattern = [a-zA-Z0-9_] |
分布 (distribution) | ジョブの配布構成。 設定する場合は、Mpi、Tensorflow、PyTorch、または null のいずれかである必要があります。 | DistributionConfiguration |
environmentId | [必須]ジョブの環境仕様の ARM リソース ID。 | string (必須) 制約: Pattern = [a-zA-Z0-9_] |
environmentVariables | ジョブに含まれる環境変数。 | object |
inputs | ジョブで使用される入力データ バインディングのマッピング。 | object |
制限 | コマンド ジョブの制限。 | CommandJobLimits |
outputs | ジョブで使用される出力データ バインディングのマッピング。 | object |
resources | ジョブのコンピューティング リソースの構成。 | ResourceConfiguration |
DistributionConfiguration
名前 | 説明 | 値 |
---|---|---|
distributionType | オブジェクトの種類を設定する | Mpi PyTorch TensorFlow (必須) |
Mpi
名前 | 説明 | 値 |
---|---|---|
distributionType | [必須]ディストリビューション フレームワークの種類を指定します。 | 'Mpi' (必須) |
processCountPerInstance | MPI ノードあたりのプロセス数。 | INT |
PyTorch
名前 | 説明 | 値 |
---|---|---|
distributionType | [必須]ディストリビューション フレームワークの種類を指定します。 | 'PyTorch' (必須) |
processCountPerInstance | ノードあたりのプロセス数。 | INT |
TensorFlow
名前 | 説明 | 値 |
---|---|---|
distributionType | [必須]ディストリビューション フレームワークの種類を指定します。 | 'TensorFlow' (必須) |
parameterServerCount | パラメーター サーバー タスクの数。 | INT |
workerCount | worker 数。 指定しない場合、 は既定でインスタンス数になります。 | INT |
CommandJobLimits
名前 | 説明 | 値 |
---|---|---|
jobLimitsType | [必須]JobLimit 型。 | 'Command' 'Sweep' (必須) |
timeout | ISO 8601 形式の最大実行時間。その後、ジョブは取り消されます。 有効桁数が秒の期間のみをサポートします。 | string |
ResourceConfiguration
名前 | 説明 | 値 |
---|---|---|
instanceCount | コンピューティング 先で使用されるインスタンスまたはノードの数 (省略可能)。 | INT |
instanceType | コンピューティング 先でサポートされている VM のオプションの種類。 | string |
properties | その他のプロパティ バッグ。 | object |
PipelineJob
名前 | 説明 | 値 |
---|---|---|
jobType | [必須]ジョブの種類を指定します。 | 'Pipeline' (必須) |
inputs | パイプライン ジョブの入力。 | object |
jobs | ジョブはパイプライン ジョブを構築します。 | object |
outputs | パイプライン ジョブの出力 | object |
settings | ContinueRunOnStepFailure などのパイプライン設定。 | Bicep の場合は、 any() 関数を使用できます。 |
SweepJob
名前 | 説明 | 値 |
---|---|---|
jobType | [必須]ジョブの種類を指定します。 | 'Sweep' (必須) |
earlyTermination | 早期終了ポリシーを使用すると、実行が完了する前にパフォーマンスの低い実行を取り消す | EarlyTerminationPolicy |
inputs | ジョブで使用される入力データ バインディングのマッピング。 | object |
制限 | スイープ ジョブの制限。 | SweepJobLimits |
objective | [必須]最適化の目標。 | 目標 (必須) |
outputs | ジョブで使用される出力データ バインディングのマッピング。 | object |
samplingAlgorithm | [必須]ハイパーパラメーター サンプリング アルゴリズム | SamplingAlgorithm (必須) |
searchSpace | [必須]各パラメーターとその分布を含むディクショナリ。 ディクショナリ キーは、 パラメーターの名前です | Bicep の場合は、 any() 関数を使用できます。(必須) |
試用 | [必須]試用版コンポーネントの定義。 | TrialComponent (必須) |
EarlyTerminationPolicy
名前 | 説明 | 値 |
---|---|---|
delayEvaluation | 最初の評価を遅延させる間隔の数。 | INT |
evaluationInterval | ポリシー評価間の間隔 (実行回数)。 | INT |
policyType | オブジェクトの種類を設定する | Bandit MedianStopping TruncationSelection (必須) |
BanditPolicy
名前 | 説明 | 値 |
---|---|---|
policyType | [必須]ポリシー構成の名前 | 'Bandit' (必須) |
slackAmount | 最高のパフォーマンスを発揮する実行から許容される絶対距離。 | INT |
slackFactor | 最もパフォーマンスの高い実行からの許可された距離の比率。 | INT |
MedianStoppingPolicy
名前 | 説明 | 値 |
---|---|---|
policyType | [必須]ポリシー構成の名前 | 'MedianStopping' (必須) |
TruncationSelectionPolicy
名前 | 説明 | 値 |
---|---|---|
policyType | [必須]ポリシー構成の名前 | 'TruncationSelection' (必須) |
truncationPercentage | 各評価期間に取り消される実行の割合。 | INT |
SweepJobLimits
名前 | 説明 | 値 |
---|---|---|
jobLimitsType | [必須]JobLimit 型。 | 'Command' 'Sweep' (必須) |
maxConcurrentTrials | スイープ ジョブの最大同時試行数。 | INT |
maxTotalTrials | スイープ ジョブの最大試行回数。 | INT |
timeout | ISO 8601 形式の最大実行時間。その後、ジョブは取り消されます。 秒と同じ精度の期間のみをサポートします。 | string |
trialTimeout | スイープ ジョブ試用版のタイムアウト値。 | string |
目的
名前 | 説明 | 値 |
---|---|---|
goal | [必須]ハイパーパラメーター調整でサポートされるメトリック目標を定義します | '最大化' '最小化' (必須) |
primaryMetric | [必須]最適化するメトリックの名前。 | string (必須) 制約: Pattern = [a-zA-Z0-9_] |
SamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | オブジェクトの種類を設定する | ベイジアン Grid ランダム (必須) |
BayesianSamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | [必須]構成プロパティと共にハイパーパラメーター値を生成するために使用されるアルゴリズム | 'Bayesian' (必須) |
GridSamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | [必須]構成プロパティと共にハイパーパラメーター値を生成するために使用されるアルゴリズム | 'Grid' (必須) |
RandomSamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | [必須]構成プロパティと共にハイパーパラメーター値を生成するために使用されるアルゴリズム | 'Random' (必須) |
ルール | ランダム アルゴリズムの特定の種類 | 'Random' 'Sobol' |
seed | 乱数生成のシードとして使用する省略可能な整数 | INT |
TrialComponent
名前 | 説明 | 値 |
---|---|---|
codeId | コード資産の ARM リソース ID。 | string |
command | [必須]ジョブの起動時に実行するコマンド。 例: "python train.py" | string (必須) 制約: 最小長 = 1 Pattern = [a-zA-Z0-9_] |
分布 (distribution) | ジョブの配布構成。 設定する場合は、Mpi、Tensorflow、PyTorch、または null のいずれかである必要があります。 | DistributionConfiguration |
environmentId | [必須]ジョブの環境仕様の ARM リソース ID。 | string (必須) 制約: Pattern = [a-zA-Z0-9_] |
environmentVariables | ジョブに含まれる環境変数。 | object |
resources | ジョブのコンピューティング リソースの構成。 | ResourceConfiguration |
クイック スタート テンプレート
次のクイックスタート テンプレートでは、このリソースの種類をデプロイします。
Template | 説明 |
---|---|
Azure Machine Learning AutoML 分類ジョブを作成する |
このテンプレートでは、Azure Machine Learning AutoML 分類ジョブを作成して、クライアントが金融機関との固定定期預金をサブスクライブするかどうかを予測するための最適なモデルを見つけます。 |
Azure Machine Learning コマンド ジョブを作成する |
このテンプレートでは、基本的なhello_world スクリプトを使用して Azure Machine Learning コマンド ジョブを作成します |
Azure Machine Learning スイープ ジョブを作成する |
このテンプレートでは、ハイパーパラメーター調整用の Azure Machine Learning スイープ ジョブを作成します。 |
ARM テンプレート リソース定義
ワークスペース/ジョブ リソースの種類は、次を対象とする操作でデプロイできます。
- リソース グループ - リソース グループのデプロイ コマンドに関するページを参照してください
各 API バージョンの変更されたプロパティの一覧については、「 変更ログ」を参照してください。
リソース形式
Microsoft.MachineLearningServices/workspaces/jobs リソースを作成するには、次の JSON をテンプレートに追加します。
{
"type": "Microsoft.MachineLearningServices/workspaces/jobs",
"apiVersion": "2022-05-01",
"name": "string",
"properties": {
"computeId": "string",
"description": "string",
"displayName": "string",
"experimentName": "string",
"identity": {
"identityType": "string"
// For remaining properties, see IdentityConfiguration objects
},
"isArchived": "bool",
"properties": {},
"services": {},
"tags": {},
"jobType": "string"
// For remaining properties, see JobBaseProperties objects
}
}
JobBaseProperties オブジェクト
オブジェクトの種類を指定するには、 jobType プロパティを設定します。
[コマンド] には、次のコマンドを使用します。
"jobType": "Command",
"codeId": "string",
"command": "string",
"distribution": {
"distributionType": "string"
// For remaining properties, see DistributionConfiguration objects
},
"environmentId": "string",
"environmentVariables": {},
"inputs": {},
"limits": {
"jobLimitsType": "string",
"timeout": "string"
},
"outputs": {},
"resources": {
"instanceCount": "int",
"instanceType": "string",
"properties": {}
}
パイプラインの場合は、次の値を使用します。
"jobType": "Pipeline",
"inputs": {},
"jobs": {},
"outputs": {},
"settings": {}
スイープの場合は、次の値を使用します。
"jobType": "Sweep",
"earlyTermination": {
"delayEvaluation": "int",
"evaluationInterval": "int",
"policyType": "string"
// For remaining properties, see EarlyTerminationPolicy objects
},
"inputs": {},
"limits": {
"jobLimitsType": "string",
"maxConcurrentTrials": "int",
"maxTotalTrials": "int",
"timeout": "string",
"trialTimeout": "string"
},
"objective": {
"goal": "string",
"primaryMetric": "string"
},
"outputs": {},
"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": {},
"resources": {
"instanceCount": "int",
"instanceType": "string",
"properties": {}
}
}
IdentityConfiguration オブジェクト
identityType プロパティを設定して、オブジェクトの種類を指定します。
AMLToken の場合は、次の値を使用します。
"identityType": "AMLToken"
[マネージド] には、次の値を使用します。
"identityType": "Managed",
"clientId": "string",
"objectId": "string",
"resourceId": "string"
UserIdentity の場合は、次の値を使用します。
"identityType": "UserIdentity"
DistributionConfiguration オブジェクト
distributionType プロパティを設定して、オブジェクトの種類を指定します。
Mpi の場合は、次の値を使用します。
"distributionType": "Mpi",
"processCountPerInstance": "int"
PyTorch の場合は、次の値を使用します。
"distributionType": "PyTorch",
"processCountPerInstance": "int"
TensorFlow の場合は、次の値を使用します。
"distributionType": "TensorFlow",
"parameterServerCount": "int",
"workerCount": "int"
EarlyTerminationPolicy オブジェクト
policyType プロパティを設定して、オブジェクトの種類を指定します。
Bandit の場合は、次の値を使用します。
"policyType": "Bandit",
"slackAmount": "int",
"slackFactor": "int"
MedianStopping の場合は、次の値を使用します。
"policyType": "MedianStopping"
TruncationSelection の場合は、次を使用します。
"policyType": "TruncationSelection",
"truncationPercentage": "int"
SamplingAlgorithm オブジェクト
オブジェクトの種類を指定するには、 samplingAlgorithmType プロパティを設定します。
ベイジアンの場合は、次の値を使用します。
"samplingAlgorithmType": "Bayesian"
Grid の場合は、次の値を使用します。
"samplingAlgorithmType": "Grid"
Random の場合は、次の値を使用します。
"samplingAlgorithmType": "Random",
"rule": "string",
"seed": "int"
プロパティ値
workspaces/jobs
名前 | 説明 | 値 |
---|---|---|
type | リソースの種類 | 'Microsoft.MachineLearningServices/workspaces/jobs' |
apiVersion | リソース API のバージョン | '2022-05-01' |
name | リソース名 JSON ARM テンプレートで子リソースの名前と型を設定する方法を参照してください。 |
string (必須) |
properties | [必須]エンティティの追加の属性。 | JobBaseProperties (必須) |
JobBaseProperties
名前 | 説明 | 値 |
---|---|---|
computeId | コンピューティング リソースの ARM リソース ID。 | string |
description | 資産の説明テキスト。 | string |
displayName | ジョブの表示名。 | string |
experimentName | ジョブが属する実験の名前。 設定されていない場合、ジョブは "Default" 実験に配置されます。 | string |
identity | ID の構成。 設定されている場合、これは AmlToken、ManagedIdentity、UserIdentity、または null のいずれかである必要があります。 null の場合、既定値は AmlToken です。 |
IdentityConfiguration |
isArchived | 資産はアーカイブされていますか? | [bool] |
properties | 資産プロパティディクショナリ。 | object |
services | JobEndpoints の一覧。 ローカル ジョブの場合、ジョブ エンドポイントのエンドポイント値は FileStreamObject になります。 |
object |
tags | タグ辞書。 タグは追加、削除、更新できます。 | object |
jobType | オブジェクトの種類を設定する | コマンド パイプライン スイープ (必須) |
IdentityConfiguration
名前 | 説明 | 値 |
---|---|---|
identityType | オブジェクトの種類を設定する | AMLToken マネージド UserIdentity (必須) |
AmlToken
名前 | 説明 | 値 |
---|---|---|
identityType | [必須]ID フレームワークの種類を指定します。 | 'AMLToken' (必須) |
ManagedIdentity
名前 | 説明 | 値 |
---|---|---|
identityType | [必須]ID フレームワークの種類を指定します。 | 'Managed' (必須) |
clientId | クライアント ID でユーザー割り当て ID を指定します。 システム割り当ての場合は、このフィールドを設定しないでください。 | string 制約: 最小長 = 36 最大長 = 36 Pattern = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$ |
objectId | オブジェクト ID でユーザー割り当て ID を指定します。 システム割り当ての場合は、このフィールドを設定しないでください。 | string 制約: 最小長 = 36 最大長 = 36 Pattern = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$ |
resourceId | ARM リソース ID でユーザー割り当て ID を指定します。 システム割り当ての場合は、このフィールドを設定しないでください。 | string |
UserIdentity
名前 | 説明 | 値 |
---|---|---|
identityType | [必須]ID フレームワークの種類を指定します。 | 'UserIdentity' (必須) |
CommandJob
名前 | 説明 | 値 |
---|---|---|
jobType | [必須]ジョブの種類を指定します。 | 'Command' (必須) |
codeId | コード資産の ARM リソース ID。 | string |
command | [必須]ジョブの起動時に実行するコマンド。 例: "python train.py" | string (必須) 制約: 最小長 = 1 Pattern = [a-zA-Z0-9_] |
分布 (distribution) | ジョブの配布構成。 設定されている場合、これは Mpi、Tensorflow、PyTorch、または null のいずれかである必要があります。 | DistributionConfiguration |
environmentId | [必須]ジョブの環境仕様の ARM リソース ID。 | string (必須) 制約: Pattern = [a-zA-Z0-9_] |
environmentVariables | ジョブに含まれる環境変数。 | object |
inputs | ジョブで使用される入力データ バインディングのマッピング。 | object |
制限 | コマンド ジョブの制限。 | CommandJobLimits |
outputs | ジョブで使用される出力データ バインディングのマッピング。 | object |
resources | ジョブのコンピューティング リソースの構成。 | ResourceConfiguration |
DistributionConfiguration
名前 | 説明 | 値 |
---|---|---|
distributionType | オブジェクトの種類を設定する | Mpi PyTorch TensorFlow (必須) |
Mpi
名前 | 説明 | 値 |
---|---|---|
distributionType | [必須]ディストリビューション フレームワークの種類を指定します。 | 'Mpi' (必須) |
processCountPerInstance | MPI ノードあたりのプロセス数。 | INT |
PyTorch
名前 | 説明 | 値 |
---|---|---|
distributionType | [必須]ディストリビューション フレームワークの種類を指定します。 | 'PyTorch' (必須) |
processCountPerInstance | ノードあたりのプロセス数。 | INT |
TensorFlow
名前 | 説明 | 値 |
---|---|---|
distributionType | [必須]ディストリビューション フレームワークの種類を指定します。 | 'TensorFlow' (必須) |
parameterServerCount | パラメーター サーバー タスクの数。 | INT |
workerCount | worker 数。 指定しない場合、 は既定でインスタンス数になります。 | INT |
CommandJobLimits
名前 | 説明 | 値 |
---|---|---|
jobLimitsType | [必須]JobLimit 型。 | 'Command' 'Sweep' (必須) |
timeout | ISO 8601 形式の最大実行時間。その後、ジョブは取り消されます。 秒と同じ精度の期間のみをサポートします。 | string |
ResourceConfiguration
名前 | 説明 | 値 |
---|---|---|
instanceCount | コンピューティング 先で使用されるインスタンスまたはノードの数 (省略可能)。 | INT |
instanceType | コンピューティング 先でサポートされている VM の種類 (省略可能)。 | string |
properties | 追加のプロパティ バッグ。 | object |
PipelineJob
名前 | 説明 | 値 |
---|---|---|
jobType | [必須]ジョブの種類を指定します。 | 'Pipeline' (必須) |
inputs | パイプライン ジョブの入力。 | object |
jobs | ジョブはパイプライン ジョブを構築します。 | object |
outputs | パイプライン ジョブの出力 | object |
settings | ContinueRunOnStepFailure などのパイプライン設定。 |
SweepJob
名前 | 説明 | 値 |
---|---|---|
jobType | [必須]ジョブの種類を指定します。 | 'Sweep' (必須) |
earlyTermination | 早期終了ポリシーを使用すると、実行が完了する前にパフォーマンスの低い実行をキャンセルできます | EarlyTerminationPolicy |
inputs | ジョブで使用される入力データ バインディングのマッピング。 | object |
制限 | スイープ ジョブの制限。 | SweepJobLimits |
objective | [必須]最適化の目標。 | 目標 (必須) |
outputs | ジョブで使用される出力データ バインディングのマッピング。 | object |
samplingAlgorithm | [必須]ハイパーパラメーター サンプリング アルゴリズム | SamplingAlgorithm (必須) |
searchSpace | [必須]各パラメーターとその分布を含むディクショナリ。 ディクショナリ キーは、 パラメーターの名前です | |
試用 | [必須]試用版コンポーネントの定義。 | TrialComponent (必須) |
EarlyTerminationPolicy
名前 | 説明 | 値 |
---|---|---|
delayEvaluation | 最初の評価を遅延させる間隔の数。 | INT |
evaluationInterval | ポリシー評価間の間隔 (実行回数)。 | INT |
policyType | オブジェクトの種類を設定する | Bandit MedianStopping TruncationSelection (必須) |
BanditPolicy
名前 | 説明 | 値 |
---|---|---|
policyType | [必須]ポリシー構成の名前 | 'Bandit' (必須) |
slackAmount | 最高のパフォーマンスを発揮する実行から許容される絶対距離。 | INT |
slackFactor | 最もパフォーマンスの高い実行からの許可された距離の比率。 | INT |
MedianStoppingPolicy
名前 | 説明 | 値 |
---|---|---|
policyType | [必須]ポリシー構成の名前 | 'MedianStopping' (必須) |
TruncationSelectionPolicy
名前 | 説明 | 値 |
---|---|---|
policyType | [必須]ポリシー構成の名前 | 'TruncationSelection' (必須) |
truncationPercentage | 各評価期間に取り消される実行の割合。 | INT |
SweepJobLimits
名前 | 説明 | 値 |
---|---|---|
jobLimitsType | [必須]JobLimit 型。 | 'Command' 'Sweep' (必須) |
maxConcurrentTrials | スイープ ジョブの最大同時試行数。 | INT |
maxTotalTrials | スイープ ジョブの最大試行回数。 | INT |
timeout | ISO 8601 形式の最大実行時間。その後、ジョブは取り消されます。 秒と同じ精度の期間のみをサポートします。 | string |
trialTimeout | スイープ ジョブ試用版のタイムアウト値。 | string |
目的
名前 | 説明 | 値 |
---|---|---|
goal | [必須]ハイパーパラメーター調整でサポートされるメトリック目標を定義します | '最大化' '最小化' (必須) |
primaryMetric | [必須]最適化するメトリックの名前。 | string (必須) 制約: Pattern = [a-zA-Z0-9_] |
SamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | オブジェクトの種類を設定する | ベイジアン Grid ランダム (必須) |
BayesianSamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | [必須]構成プロパティと共にハイパーパラメーター値を生成するために使用されるアルゴリズム | 'Bayesian' (必須) |
GridSamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | [必須]構成プロパティと共にハイパーパラメーター値を生成するために使用されるアルゴリズム | 'Grid' (必須) |
RandomSamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | [必須]構成プロパティと共にハイパーパラメーター値を生成するために使用されるアルゴリズム | 'Random' (必須) |
ルール | ランダム アルゴリズムの特定の種類 | 'Random' 'Sobol' |
seed | 乱数生成のシードとして使用する省略可能な整数 | INT |
TrialComponent
名前 | 説明 | 値 |
---|---|---|
codeId | コード資産の ARM リソース ID。 | string |
command | [必須]ジョブの起動時に実行するコマンド。 例: "python train.py" | string (必須) 制約: 最小長 = 1 Pattern = [a-zA-Z0-9_] |
分布 (distribution) | ジョブの配布構成。 設定する場合は、Mpi、Tensorflow、PyTorch、または null のいずれかである必要があります。 | DistributionConfiguration |
environmentId | [必須]ジョブの環境仕様の ARM リソース ID。 | string (必須) 制約: Pattern = [a-zA-Z0-9_] |
environmentVariables | ジョブに含まれる環境変数。 | object |
resources | ジョブのコンピューティング リソースの構成。 | ResourceConfiguration |
クイック スタート テンプレート
次のクイックスタート テンプレートでは、このリソースの種類をデプロイします。
Template | 説明 |
---|---|
Azure Machine Learning AutoML 分類ジョブを作成する |
このテンプレートでは、Azure Machine Learning AutoML 分類ジョブを作成して、クライアントが金融機関との固定定期預金をサブスクライブするかどうかを予測するための最適なモデルを見つけます。 |
Azure Machine Learning コマンド ジョブを作成する |
このテンプレートでは、基本的なhello_world スクリプトを使用して Azure Machine Learning コマンド ジョブを作成します |
Azure Machine Learning スイープ ジョブを作成する |
このテンプレートでは、ハイパーパラメーター調整用の Azure Machine Learning スイープ ジョブを作成します。 |
Terraform (AzAPI プロバイダー) リソース定義
ワークスペース/ジョブ リソースの種類は、次を対象とする操作でデプロイできます。
- リソース グループ
各 API バージョンの変更されたプロパティの一覧については、「 変更ログ」を参照してください。
リソース形式
Microsoft.MachineLearningServices/workspaces/jobs リソースを作成するには、次の Terraform をテンプレートに追加します。
resource "azapi_resource" "symbolicname" {
type = "Microsoft.MachineLearningServices/workspaces/jobs@2022-05-01"
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 = {}
services = {}
tags = {}
jobType = "string"
// For remaining properties, see JobBaseProperties objects
}
})
}
JobBaseProperties オブジェクト
オブジェクトの種類を指定するには、 jobType プロパティを設定します。
[コマンド] には、次のコマンドを使用します。
jobType = "Command"
codeId = "string"
command = "string"
distribution = {
distributionType = "string"
// For remaining properties, see DistributionConfiguration objects
}
environmentId = "string"
environmentVariables = {}
inputs = {}
limits = {
jobLimitsType = "string"
timeout = "string"
}
outputs = {}
resources = {
instanceCount = int
instanceType = "string"
properties = {}
}
パイプラインの場合は、次の値を使用します。
jobType = "Pipeline"
inputs = {}
jobs = {}
outputs = {}
スイープの場合は、次の値を使用します。
jobType = "Sweep"
earlyTermination = {
delayEvaluation = int
evaluationInterval = int
policyType = "string"
// For remaining properties, see EarlyTerminationPolicy objects
}
inputs = {}
limits = {
jobLimitsType = "string"
maxConcurrentTrials = int
maxTotalTrials = int
timeout = "string"
trialTimeout = "string"
}
objective = {
goal = "string"
primaryMetric = "string"
}
outputs = {}
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 = {}
resources = {
instanceCount = int
instanceType = "string"
properties = {}
}
}
IdentityConfiguration オブジェクト
identityType プロパティを設定して、オブジェクトの種類を指定します。
AMLToken の場合は、次の値を使用します。
identityType = "AMLToken"
[マネージド] には、次の値を使用します。
identityType = "Managed"
clientId = "string"
objectId = "string"
resourceId = "string"
UserIdentity の場合は、次の値を使用します。
identityType = "UserIdentity"
DistributionConfiguration オブジェクト
distributionType プロパティを設定して、オブジェクトの種類を指定します。
Mpi の場合は、次の値を使用します。
distributionType = "Mpi"
processCountPerInstance = int
PyTorch の場合は、次の値を使用します。
distributionType = "PyTorch"
processCountPerInstance = int
TensorFlow の場合は、次の値を使用します。
distributionType = "TensorFlow"
parameterServerCount = int
workerCount = int
EarlyTerminationPolicy オブジェクト
policyType プロパティを設定して、オブジェクトの種類を指定します。
Bandit の場合は、次の値を使用します。
policyType = "Bandit"
slackAmount = int
slackFactor = int
MedianStopping の場合は、次の値を使用します。
policyType = "MedianStopping"
TruncationSelection の場合は、次の値を使用します。
policyType = "TruncationSelection"
truncationPercentage = int
SamplingAlgorithm オブジェクト
samplingAlgorithmType プロパティを設定して、オブジェクトの種類を指定します。
ベイジアンの場合は、次の値を使用します。
samplingAlgorithmType = "Bayesian"
Grid の場合は、次の値を使用します。
samplingAlgorithmType = "Grid"
[Random]\(ランダム\) には、次の値を使用します。
samplingAlgorithmType = "Random"
rule = "string"
seed = int
プロパティ値
workspaces/jobs
名前 | 説明 | 値 |
---|---|---|
type | リソースの種類 | "Microsoft.MachineLearningServices/workspaces/jobs@2022-05-01" |
name | リソース名 | string (必須) |
parent_id | このリソースの親であるリソースの ID。 | 種類のリソースの ID: ワークスペース |
properties | [必須]エンティティの追加の属性。 | JobBaseProperties (必須) |
JobBaseProperties
名前 | 説明 | 値 |
---|---|---|
computeId | コンピューティング リソースの ARM リソース ID。 | string |
description | 資産の説明テキスト。 | string |
displayName | ジョブの表示名。 | string |
experimentName | ジョブが属する実験の名前。 設定されていない場合、ジョブは "Default" 実験に配置されます。 | string |
identity | ID の構成。 設定されている場合、これは AmlToken、ManagedIdentity、UserIdentity、または null のいずれかである必要があります。 null の場合、既定値は AmlToken です。 |
IdentityConfiguration |
isArchived | 資産はアーカイブされていますか? | [bool] |
properties | 資産プロパティディクショナリ。 | object |
services | JobEndpoints の一覧。 ローカル ジョブの場合、ジョブ エンドポイントのエンドポイント値は FileStreamObject になります。 |
object |
tags | タグ辞書。 タグは追加、削除、更新できます。 | object |
jobType | オブジェクトの種類を設定する | コマンド パイプライン スイープ (必須) |
IdentityConfiguration
名前 | 説明 | 値 |
---|---|---|
identityType | オブジェクトの種類を設定する | AMLToken マネージド UserIdentity (必須) |
AmlToken
名前 | 説明 | 値 |
---|---|---|
identityType | [必須]ID フレームワークの種類を指定します。 | "AMLToken" (必須) |
ManagedIdentity
名前 | 説明 | 値 |
---|---|---|
identityType | [必須]ID フレームワークの種類を指定します。 | "Managed" (必須) |
clientId | クライアント ID でユーザー割り当て ID を指定します。 システム割り当ての場合は、このフィールドを設定しないでください。 | string 制約: 最小長 = 36 最大長 = 36 Pattern = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$ |
objectId | オブジェクト ID でユーザー割り当て ID を指定します。 システム割り当ての場合は、このフィールドを設定しないでください。 | string 制約: 最小長 = 36 最大長 = 36 Pattern = ^[0-9a-fA-F]{8}-([0-9a-fA-F]{4}-){3}[0-9a-fA-F]{12}$ |
resourceId | ARM リソース ID でユーザー割り当て ID を指定します。 システム割り当ての場合は、このフィールドを設定しないでください。 | string |
UserIdentity
名前 | 説明 | 値 |
---|---|---|
identityType | [必須]ID フレームワークの種類を指定します。 | "UserIdentity" (必須) |
CommandJob
名前 | 説明 | 値 |
---|---|---|
jobType | [必須]ジョブの種類を指定します。 | "Command" (必須) |
codeId | コード資産の ARM リソース ID。 | string |
command | [必須]ジョブの起動時に実行するコマンド。 例: "python train.py" | string (必須) 制約: 最小長 = 1 Pattern = [a-zA-Z0-9_] |
分布 (distribution) | ジョブの配布構成。 設定されている場合、これは Mpi、Tensorflow、PyTorch、または null のいずれかである必要があります。 | DistributionConfiguration |
environmentId | [必須]ジョブの環境仕様の ARM リソース ID。 | string (必須) 制約: Pattern = [a-zA-Z0-9_] |
environmentVariables | ジョブに含まれる環境変数。 | object |
inputs | ジョブで使用される入力データ バインディングのマッピング。 | object |
制限 | コマンド ジョブの制限。 | CommandJobLimits |
outputs | ジョブで使用される出力データ バインディングのマッピング。 | object |
resources | ジョブのコンピューティング リソースの構成。 | ResourceConfiguration |
DistributionConfiguration
名前 | 説明 | 値 |
---|---|---|
distributionType | オブジェクトの種類を設定する | Mpi PyTorch TensorFlow (必須) |
Mpi
名前 | 説明 | 値 |
---|---|---|
distributionType | [必須]ディストリビューション フレームワークの種類を指定します。 | "Mpi" (必須) |
processCountPerInstance | MPI ノードあたりのプロセス数。 | INT |
PyTorch
名前 | 説明 | 値 |
---|---|---|
distributionType | [必須]ディストリビューション フレームワークの種類を指定します。 | "PyTorch" (必須) |
processCountPerInstance | ノードあたりのプロセス数。 | INT |
TensorFlow
名前 | 説明 | 値 |
---|---|---|
distributionType | [必須]ディストリビューション フレームワークの種類を指定します。 | "TensorFlow" (必須) |
parameterServerCount | パラメーター サーバー タスクの数。 | INT |
workerCount | worker 数。 指定しない場合、 は既定でインスタンス数になります。 | INT |
CommandJobLimits
名前 | 説明 | 値 |
---|---|---|
jobLimitsType | [必須]JobLimit 型。 | "Command" "Sweep" (必須) |
timeout | ISO 8601 形式の最大実行時間。その後、ジョブは取り消されます。 有効桁数が秒の期間のみをサポートします。 | string |
ResourceConfiguration
名前 | 説明 | 値 |
---|---|---|
instanceCount | コンピューティング 先で使用されるインスタンスまたはノードの数 (省略可能)。 | INT |
instanceType | コンピューティング 先でサポートされている VM のオプションの種類。 | string |
properties | その他のプロパティ バッグ。 | object |
PipelineJob
名前 | 説明 | 値 |
---|---|---|
jobType | [必須]ジョブの種類を指定します。 | "Pipeline" (必須) |
inputs | パイプライン ジョブの入力。 | object |
jobs | ジョブはパイプライン ジョブを構築します。 | object |
outputs | パイプライン ジョブの出力 | object |
settings | ContinueRunOnStepFailure などのパイプライン設定。 |
SweepJob
名前 | 説明 | 値 |
---|---|---|
jobType | [必須]ジョブの種類を指定します。 | "Sweep" (必須) |
earlyTermination | 早期終了ポリシーを使用すると、実行が完了する前にパフォーマンスの低い実行を取り消す | EarlyTerminationPolicy |
inputs | ジョブで使用される入力データ バインディングのマッピング。 | object |
制限 | スイープ ジョブの制限。 | SweepJobLimits |
objective | [必須]最適化の目的。 | 目標 (必須) |
outputs | ジョブで使用される出力データ バインディングのマッピング。 | object |
samplingAlgorithm | [必須]ハイパーパラメーター サンプリング アルゴリズム | SamplingAlgorithm (必須) |
searchSpace | [必須]各パラメーターとその分布を含むディクショナリ。 ディクショナリ キーは パラメーターの名前です | |
試用 | [必須]試用版コンポーネントの定義。 | TrialComponent (必須) |
EarlyTerminationPolicy
名前 | 説明 | 値 |
---|---|---|
delayEvaluation | 最初の評価を遅らせる間隔の数。 | INT |
evaluationInterval | ポリシー評価間の間隔 (実行数)。 | INT |
policyType | オブジェクトの種類を設定する | Bandit MedianStopping TruncationSelection (必須) |
BanditPolicy
名前 | 説明 | 値 |
---|---|---|
policyType | [必須]ポリシー構成の名前 | "Bandit" (必須) |
slackAmount | 最高のパフォーマンスを発揮する実行から許容される絶対距離。 | INT |
slackFactor | 最もパフォーマンスの高い実行からの許可された距離の比率。 | INT |
MedianStoppingPolicy
名前 | 説明 | 値 |
---|---|---|
policyType | [必須]ポリシー構成の名前 | "MedianStopping" (必須) |
TruncationSelectionPolicy
名前 | 説明 | 値 |
---|---|---|
policyType | [必須]ポリシー構成の名前 | "TruncationSelection" (必須) |
truncationPercentage | 各評価期間に取り消される実行の割合。 | INT |
SweepJobLimits
名前 | 説明 | 値 |
---|---|---|
jobLimitsType | [必須]JobLimit 型。 | "Command" "Sweep" (必須) |
maxConcurrentTrials | スイープ ジョブの最大同時試行回数。 | INT |
maxTotalTrials | スイープ ジョブの最大試行回数。 | INT |
timeout | ISO 8601 形式の最大実行時間。その後、ジョブは取り消されます。 有効桁数が秒の期間のみをサポートします。 | string |
trialTimeout | スイープ ジョブ試用版のタイムアウト値。 | string |
目的
名前 | 説明 | 値 |
---|---|---|
goal | [必須]ハイパーパラメーターチューニングでサポートされるメトリック目標を定義します | "最大化" "最小化" (必須) |
primaryMetric | [必須]最適化するメトリックの名前。 | string (必須) 制約: Pattern = [a-zA-Z0-9_] |
SamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | オブジェクトの種類を設定する | ベイジアン Grid ランダム (必須) |
BayesianSamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | [必須]構成プロパティと共にハイパーパラメーター値を生成するために使用されるアルゴリズム | "Bayesian" (必須) |
GridSamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | [必須]構成プロパティと共にハイパーパラメーター値を生成するために使用されるアルゴリズム | "Grid" (必須) |
RandomSamplingAlgorithm
名前 | 説明 | 値 |
---|---|---|
samplingAlgorithmType | [必須]構成プロパティと共にハイパーパラメーター値を生成するために使用されるアルゴリズム | "Random" (必須) |
ルール | ランダム アルゴリズムの特定の種類 | "Random" "Sobol" |
seed | 乱数生成のシードとして使用する省略可能な整数 | INT |
TrialComponent
名前 | 説明 | 値 |
---|---|---|
codeId | コード資産の ARM リソース ID。 | string |
command | [必須]ジョブの起動時に実行するコマンド。 例: "python train.py" | string (必須) 制約: 最小長 = 1 Pattern = [a-zA-Z0-9_] |
分布 (distribution) | ジョブの配布構成。 設定されている場合、これは Mpi、Tensorflow、PyTorch、または null のいずれかである必要があります。 | DistributionConfiguration |
environmentId | [必須]ジョブの環境仕様の ARM リソース ID。 | string (必須) 制約: Pattern = [a-zA-Z0-9_] |
environmentVariables | ジョブに含まれる環境変数。 | object |
resources | ジョブのコンピューティング リソースの構成。 | ResourceConfiguration |