在 YAML 中定义机器学习管道Define machine learning pipelines in YAML

了解如何在 YAML 中定义机器学习管道。Learn how to define your machine learning pipelines in YAML. 使用适用于 Azure CLI 的机器学习扩展时,许多管道相关命令都要求提供一个用于定义管道的 YAML 文件。When using the machine learning extension for the Azure CLI, many of the pipeline-related commands expect a YAML file that defines the pipeline.

下表列出了在 YAML 中定义管道时目前支持和不支持的内容:The following table lists what is and is not currently supported when defining a pipeline in YAML:

步骤类型Step type 支持?Supported?
PythonScriptStepPythonScriptStep Yes
ParallelRunStepParallelRunStep Yes
AdlaStepAdlaStep Yes
AzureBatchStepAzureBatchStep Yes
DatabricksStepDatabricksStep Yes
DataTransferStepDataTransferStep Yes
AutoMLStepAutoMLStep No
HyperDriveStepHyperDriveStep No
ModuleStepModuleStep Yes
MPIStepMPIStep No
EstimatorStepEstimatorStep No

管道定义Pipeline definition

管道定义使用以下对应于 Pipelines 类的键:A pipeline definition uses the following keys, which correspond to the Pipelines class:

YAML 键YAML key 说明Description
name 管道的说明。The description of the pipeline.
parameters 管道的参数。Parameter(s) to the pipeline.
data_reference 定义在运行中提供数据的方式和位置。Defines how and where data should be made available in a run.
default_compute 默认的计算目标,管道中的所有步骤将在其上运行。Default compute target where all steps in the pipeline run.
steps 管道中使用的步骤。The steps used in the pipeline.

parametersParameters

parameters 节使用以下对应于 PipelineParameter 类的键:The parameters section uses the following keys, which correspond to the PipelineParameter class:

YAML 键YAML key 说明Description
type 参数的值类型。The value type of the parameter. 有效类型为 stringintfloatbooldatapathValid types are string, int, float, bool, or datapath.
default 默认值。The default value.

每个参数已命名。Each parameter is named. 例如,以下 YAML 代码片段定义名为 NumIterationsParameterDataPathParameterNodeCountParameter 的三个参数:For example, the following YAML snippet defines three parameters named NumIterationsParameter, DataPathParameter, and NodeCountParameter:

pipeline:
    name: SamplePipelineFromYaml
    parameters:
        NumIterationsParameter:
            type: int
            default: 40
        DataPathParameter:
            type: datapath
            default:
                datastore: workspaceblobstore
                path_on_datastore: sample2.txt
        NodeCountParameter:
            type: int
            default: 4

数据引用Data reference

data_references 节使用以下对应于 DataReference 的键:The data_references section uses the following keys, which correspond to the DataReference:

YAML 键YAML key 说明Description
datastore 要引用的数据存储。The datastore to reference.
path_on_datastore 数据引用在后备存储中的相对路径。The relative path in the backing storage for the data reference.

每个数据引用包含在一个键中。Each data reference is contained in a key. 例如,以下 YAML 代码片段定义存储在键中的名为 employee_data 的数据引用:For example, the following YAML snippet defines a data reference stored in the key named employee_data:

pipeline:
    name: SamplePipelineFromYaml
    parameters:
        PipelineParam1:
            type: int
            default: 3
    data_references:
        employee_data:
            datastore: adftestadla
            path_on_datastore: "adla_sample/sample_input.csv"

步骤Steps

步骤定义计算环境,以及要在环境中运行的文件。Steps define a computational environment, along with the files to run on the environment. 若要定义步骤的类型,请使用 type 键:To define the type of a step, use the type key:

步骤类型Step type 说明Description
AdlaStep 使用 Azure Data Lake Analytics 运行 U-SQL 脚本。Runs a U-SQL script with Azure Data Lake Analytics. 对应于 AdlaStep 类。Corresponds to the AdlaStep class.
AzureBatchStep 使用 Azure Batch 运行作业。Runs jobs using Azure Batch. 对应于 AzureBatchStep 类。Corresponds to the AzureBatchStep class.
DatabricsStep 添加 Databricks 笔记本、Python 脚本或 JAR。Adds a Databricks notebook, Python script, or JAR. 对应于 DatabricksStep 类。Corresponds to the DatabricksStep class.
DataTransferStep 在存储选项之间传输数据。Transfers data between storage options. 对应于 DataTransferStep 类。Corresponds to the DataTransferStep class.
PythonScriptStep 运行 Python 脚本。Runs a Python script. 对应于 PythonScriptStep 类。Corresponds to the PythonScriptStep class.
ParallelRunStep 运行 Python 脚本,以异步方式并行处理大量数据。Runs a Python script to process large amounts of data asynchronously and in parallel. 对应于 ParallelRunStep 类。Corresponds to the ParallelRunStep class.

ADLA 步骤ADLA step

YAML 键YAML key 说明Description
script_name U-SQL 脚本的名称(相对于 source_directory)。The name of the U-SQL script (relative to the source_directory).
compute_target 用于此步骤的 Azure Data Lake 计算目标。The Azure Data Lake compute target to use for this step.
parameters 管道的参数Parameters to the pipeline.
inputs 输入可以是 InputPortBindingDataReferencePortDataReferencePipelineDataDatasetDatasetDefinitionPipelineDatasetInputs can be InputPortBinding, DataReference, PortDataReference, PipelineData, Dataset, DatasetDefinition, or PipelineDataset.
outputs 输出可以是 PipelineDataOutputPortBindingOutputs can be either PipelineData or OutputPortBinding.
source_directory 包含脚本、程序集等的目录。Directory that contains the script, assemblies, etc.
priority 用于当前作业的优先级值。The priority value to use for the current job.
params 名称/值对的字典。Dictionary of name-value pairs.
degree_of_parallelism 用于此作业的并行度。The degree of parallelism to use for this job.
runtime_version Data Lake Analytics 引擎的运行时版本。The runtime version of the Data Lake Analytics engine.
allow_reuse 确定当使用相同的设置再次运行时,该步骤是否应重用以前的结果。Determines whether the step should reuse previous results when run again with the same settings.

以下示例包含 ADLA 步骤定义:The following example contains an ADLA Step definition:

pipeline:
    name: SamplePipelineFromYaml
    parameters:
        PipelineParam1:
            type: int
            default: 3
    data_references:
        employee_data:
            datastore: adftestadla
            path_on_datastore: "adla_sample/sample_input.csv"
    default_compute: adlacomp
    steps:
        Step1:
            runconfig: "D:\\Yaml\\default_runconfig.yml"
            parameters:
                NUM_ITERATIONS_2:
                    source: PipelineParam1
                NUM_ITERATIONS_1: 7
            type: "AdlaStep"
            name: "MyAdlaStep"
            script_name: "sample_script.usql"
            source_directory: "D:\\scripts\\Adla"
            inputs:
                employee_data:
                    source: employee_data
            outputs:
                OutputData:
                    destination: Output4
                    datastore: adftestadla
                    bind_mode: mount

Azure Batch 步骤Azure Batch step

YAML 键YAML key 说明Description
compute_target 用于此步骤的 Azure Batch 计算目标。The Azure Batch compute target to use for this step.
inputs 输入可以是 InputPortBindingDataReferencePortDataReferencePipelineDataDatasetDatasetDefinitionPipelineDatasetInputs can be InputPortBinding, DataReference, PortDataReference, PipelineData, Dataset, DatasetDefinition, or PipelineDataset.
outputs 输出可以是 PipelineDataOutputPortBindingOutputs can be either PipelineData or OutputPortBinding.
source_directory 包含模块二进制文件、可执行文件、程序集等的目录。Directory that contains the module binaries, executable, assemblies, etc.
executable 要作为此作业的一部分运行的命令/可执行文件的名称。Name of the command/executable that will be ran as part of this job.
create_pool 布尔标志,指示在运行作业之前是否创建池。Boolean flag to indicate whether to create the pool before running the job.
delete_batch_job_after_finish 布尔标志,指示在完成作业后是否从 Batch 帐户中删除该作业。Boolean flag to indicate whether to delete the job from the Batch account after it's finished.
delete_batch_pool_after_finish 布尔标志,指示在完成作业后是否删除池。Boolean flag to indicate whether to delete the pool after the job finishes.
is_positive_exit_code_failure 布尔标志,指示在任务退出并返回正代码时作业是否失败。Boolean flag to indicate if the job fails if the task exits with a positive code.
vm_image_urn 如果 create_poolTrue,则 VM 将使用 VirtualMachineConfigurationIf create_pool is True, and VM uses VirtualMachineConfiguration.
pool_id 要在其中运行作业的池的 ID。The ID of the pool where the job will run.
allow_reuse 确定当使用相同的设置再次运行时,该步骤是否应重用以前的结果。Determines whether the step should reuse previous results when run again with the same settings.

以下示例包含 Azure Batch 步骤定义:The following example contains an Azure Batch step definition:

pipeline:
    name: SamplePipelineFromYaml
    parameters:
        PipelineParam1:
            type: int
            default: 3
    data_references:
        input:
            datastore: workspaceblobstore
            path_on_datastore: "input.txt"
    default_compute: testbatch
    steps:
        Step1:
            runconfig: "D:\\Yaml\\default_runconfig.yml"
            parameters:
                NUM_ITERATIONS_2:
                    source: PipelineParam1
                NUM_ITERATIONS_1: 7
            type: "AzureBatchStep"
            name: "MyAzureBatchStep"
            pool_id: "MyPoolName"
            create_pool: true
            executable: "azurebatch.cmd"
            source_directory: "D:\\scripts\\AureBatch"
            allow_reuse: false
            inputs:
                input:
                    source: input
            outputs:
                output:
                    destination: output
                    datastore: workspaceblobstore

Databricks 步骤Databricks step

YAML 键YAML key 说明Description
compute_target 用于此步骤的 Azure Databricks 计算目标。The Azure Databricks compute target to use for this step.
inputs 输入可以是 InputPortBindingDataReferencePortDataReferencePipelineDataDatasetDatasetDefinitionPipelineDatasetInputs can be InputPortBinding, DataReference, PortDataReference, PipelineData, Dataset, DatasetDefinition, or PipelineDataset.
outputs 输出可以是 PipelineDataOutputPortBindingOutputs can be either PipelineData or OutputPortBinding.
run_name 此运行在 Databricks 中的名称。The name in Databricks for this run.
source_directory 包含脚本和其他文件的目录。Directory that contains the script and other files.
num_workers Databricks 运行群集的辅助角色的静态编号。The static number of workers for the Databricks run cluster.
runconfig .runconfig 文件的路径。The path to a .runconfig file. 此文件是 RunConfiguration 类的 YAML 表示形式。This file is a YAML representation of the RunConfiguration class. 有关此文件的结构的详细信息,请参阅 runconfigschema.jsonFor more information on the structure of this file, see runconfigschema.json.
allow_reuse 确定当使用相同的设置再次运行时,该步骤是否应重用以前的结果。Determines whether the step should reuse previous results when run again with the same settings.

以下示例包含 Databricks 步骤:The following example contains a Databricks step:

pipeline:
    name: SamplePipelineFromYaml
    parameters:
        PipelineParam1:
            type: int
            default: 3
    data_references:
        adls_test_data:
            datastore: adftestadla
            path_on_datastore: "testdata"
        blob_test_data:
            datastore: workspaceblobstore
            path_on_datastore: "dbtest"
    default_compute: mydatabricks
    steps:
        Step1:
            runconfig: "D:\\Yaml\\default_runconfig.yml"
            parameters:
                NUM_ITERATIONS_2:
                    source: PipelineParam1
                NUM_ITERATIONS_1: 7
            type: "DatabricksStep"
            name: "MyDatabrickStep"
            run_name: "DatabricksRun"
            python_script_name: "train-db-local.py"
            source_directory: "D:\\scripts\\Databricks"
            num_workers: 1
            allow_reuse: true
            inputs:
                blob_test_data:
                    source: blob_test_data
            outputs:
                OutputData:
                    destination: Output4
                    datastore: workspaceblobstore
                    bind_mode: mount

数据传输步骤Data transfer step

YAML 键YAML key 说明Description
compute_target 用于此步骤的 Azure 数据工厂计算目标。The Azure Data Factory compute target to use for this step.
source_data_reference 充当数据传输操作的源的输入连接。Input connection that serves as the source of data transfer operations. 支持的值为 InputPortBindingDataReferencePortDataReferencePipelineDataDatasetDatasetDefinitionPipelineDatasetSupported values are InputPortBinding, DataReference, PortDataReference, PipelineData, Dataset, DatasetDefinition, or PipelineDataset.
destination_data_reference 充当数据传输操作的目标的输入连接。Input connection that serves as the destination of data transfer operations. 支持的值为 PipelineDataOutputPortBindingSupported values are PipelineData and OutputPortBinding.
allow_reuse 确定当使用相同的设置再次运行时,该步骤是否应重用以前的结果。Determines whether the step should reuse previous results when run again with the same settings.

以下示例包含数据传输步骤:The following example contains a data transfer step:

pipeline:
    name: SamplePipelineFromYaml
    parameters:
        PipelineParam1:
            type: int
            default: 3
    data_references:
        adls_test_data:
            datastore: adftestadla
            path_on_datastore: "testdata"
        blob_test_data:
            datastore: workspaceblobstore
            path_on_datastore: "testdata"
    default_compute: adftest
    steps:
        Step1:
            runconfig: "D:\\Yaml\\default_runconfig.yml"
            parameters:
                NUM_ITERATIONS_2:
                    source: PipelineParam1
                NUM_ITERATIONS_1: 7
            type: "DataTransferStep"
            name: "MyDataTransferStep"
            adla_compute_name: adftest
            source_data_reference:
                adls_test_data:
                    source: adls_test_data
            destination_data_reference:
                blob_test_data:
                    source: blob_test_data

Python 脚本步骤Python script step

YAML 键YAML key 说明Description
inputs 输入可以是 InputPortBindingDataReferencePortDataReferencePipelineDataDatasetDatasetDefinitionPipelineDatasetInputs can be InputPortBinding, DataReference, PortDataReference, PipelineData, Dataset, DatasetDefinition, or PipelineDataset.
outputs 输出可以是 PipelineDataOutputPortBindingOutputs can be either PipelineData or OutputPortBinding.
script_name Python 脚本的名称(相对于 source_directory)。The name of the Python script (relative to source_directory).
source_directory 包含脚本、Conda 环境等的目录。Directory that contains the script, Conda environment, etc.
runconfig .runconfig 文件的路径。The path to a .runconfig file. 此文件是 RunConfiguration 类的 YAML 表示形式。This file is a YAML representation of the RunConfiguration class. 有关此文件的结构的详细信息,请参阅 runconfig.jsonFor more information on the structure of this file, see runconfig.json.
allow_reuse 确定当使用相同的设置再次运行时,该步骤是否应重用以前的结果。Determines whether the step should reuse previous results when run again with the same settings.

以下示例包含 Python 脚本步骤:The following example contains a Python script step:

pipeline:
    name: SamplePipelineFromYaml
    parameters:
        PipelineParam1:
            type: int
            default: 3
    data_references:
        DataReference1:
            datastore: workspaceblobstore
            path_on_datastore: testfolder/sample.txt
    default_compute: cpu-cluster
    steps:
        Step1:
            runconfig: "D:\\Yaml\\default_runconfig.yml"
            parameters:
                NUM_ITERATIONS_2:
                    source: PipelineParam1
                NUM_ITERATIONS_1: 7
            type: "PythonScriptStep"
            name: "MyPythonScriptStep"
            script_name: "train.py"
            allow_reuse: True
            source_directory: "D:\\scripts\\PythonScript"
            inputs:
                InputData:
                    source: DataReference1
            outputs:
                OutputData:
                    destination: Output4
                    datastore: workspaceblobstore
                    bind_mode: mount

并行运行步骤Parallel run step

YAML 键YAML key 说明Description
inputs 输入可以是 DatasetDatasetDefinitionPipelineDatasetInputs can be Dataset, DatasetDefinition, or PipelineDataset.
outputs 输出可以是 PipelineDataOutputPortBindingOutputs can be either PipelineData or OutputPortBinding.
script_name Python 脚本的名称(相对于 source_directory)。The name of the Python script (relative to source_directory).
source_directory 包含脚本、Conda 环境等的目录。Directory that contains the script, Conda environment, etc.
parallel_run_config parallel_run_config.yml 文件的路径。The path to a parallel_run_config.yml file. 此文件是 ParallelRunConfig 类的 YAML 表示形式。This file is a YAML representation of the ParallelRunConfig class.
allow_reuse 确定当使用相同的设置再次运行时,该步骤是否应重用以前的结果。Determines whether the step should reuse previous results when run again with the same settings.

以下示例包含并行运行步骤:The following example contains a Parallel run step:

pipeline:
    description: SamplePipelineFromYaml
    default_compute: cpu-cluster
    data_references:
        MyMinistInput:
            dataset_name: mnist_sample_data
    parameters:
        PipelineParamTimeout:
            type: int
            default: 600
    steps:        
        Step1:
            parallel_run_config: "yaml/parallel_run_config.yml"
            type: "ParallelRunStep"
            name: "parallel-run-step-1"
            allow_reuse: True
            arguments:
            - "--progress_update_timeout"
            - parameter:timeout_parameter
            - "--side_input"
            - side_input:SideInputData
            parameters:
                timeout_parameter:
                    source: PipelineParamTimeout
            inputs:
                InputData:
                    source: MyMinistInput
            side_inputs:
                SideInputData:
                    source: Output4
                    bind_mode: mount
            outputs:
                OutputDataStep2:
                    destination: Output5
                    datastore: workspaceblobstore
                    bind_mode: mount

包含多个步骤的管道Pipeline with multiple steps

YAML 键YAML key 说明Description
steps 具有一个或多个 PipelineStep 定义的序列。Sequence of one or more PipelineStep definitions. 请注意,步骤 outputsdestination 键将成为下一步 inputssource 键。Note that the destination keys of one step's outputs become the source keys to the inputs of the next step.
pipeline:
    name: SamplePipelineFromYAML
    description: Sample multistep YAML pipeline
    data_references:
        TitanicDS:
            dataset_name: 'titanic_ds'
            bind_mode: download
    default_compute: cpu-cluster
    steps:
        Dataprep:
            type: "PythonScriptStep"
            name: "DataPrep Step"
            compute: cpu-cluster
            runconfig: ".\\default_runconfig.yml"
            script_name: "prep.py"
            arguments:
            - '--train_path'
            - output:train_path
            - '--test_path'
            - output:test_path
            allow_reuse: True
            inputs:
                titanic_ds:
                    source: TitanicDS
                    bind_mode: download
            outputs:
                train_path:
                    destination: train_csv
                    datastore: workspaceblobstore
                test_path:
                    destination: test_csv
        Training:
            type: "PythonScriptStep"
            name: "Training Step"
            compute: cpu-cluster
            runconfig: ".\\default_runconfig.yml"
            script_name: "train.py"
            arguments:
            - "--train_path"
            - input:train_path
            - "--test_path"
            - input:test_path
            inputs:
                train_path:
                    source: train_csv
                    bind_mode: download
                test_path:
                    source: test_csv
                    bind_mode: download

计划Schedules

为管道定义计划时,该计划可以由数据存储触发,或根据某个时间间隔重复运行。When defining the schedule for a pipeline, it can be either datastore-triggered or recurring based on a time interval. 下面是用于定义计划的键:The following are the keys used to define a schedule:

YAML 键YAML key 说明Description
description 计划的说明。A description of the schedule.
recurrence 包含重复设置(如果计划是重复性的)。Contains recurrence settings, if the schedule is recurring.
pipeline_parameters 管道所需的任何参数。Any parameters that are required by the pipeline.
wait_for_provisioning 是否等待计划预配完成。Whether to wait for provisioning of the schedule to complete.
wait_timeout 超时之前等待的秒数。The number of seconds to wait before timing out.
datastore_name 要在其中监视已修改/已添加的 Blob 的数据存储。The datastore to monitor for modified/added blobs.
polling_interval 轮询已修改/已添加的 Blob 的间隔时间(分钟)。How long, in minutes, between polling for modified/added blobs. 默认值:5 分钟。Default value: 5 minutes. 仅支持数据存储计划。Only supported for datastore schedules.
data_path_parameter_name 要使用更改的 Blob 路径设置的数据路径管道参数的名称。The name of the data path pipeline parameter to set with the changed blob path. 仅支持数据存储计划。Only supported for datastore schedules.
continue_on_step_failure 当某个步骤失败时,是否继续执行提交的 PipelineRun 中的其他步骤。Whether to continue execution of other steps in the submitted PipelineRun if a step fails. 如果提供此键,将替代管道的 continue_on_step_failure 设置。If provided, will override the continue_on_step_failure setting of the pipeline.
path_on_datastore 可选。Optional. 要在其中监视已修改/已添加的 Blob 的数据存储上的路径。The path on the datastore to monitor for modified/added blobs. 该路径位于数据存储的容器下,因此,计划监视的实际路径是 container/path_on_datastoreThe path is under the container for the datastore, so the actual path the schedule monitors is container/path_on_datastore. 如果没有此路径,将监视数据存储容器。If none, the datastore container is monitored. 不会监视在 path_on_datastore 的子文件夹中进行的添加/修改。Additions/modifications made in a subfolder of the path_on_datastore are not monitored. 仅支持数据存储计划。Only supported for datastore schedules.

以下示例包含数据存储触发的计划的定义:The following example contains the definition for a datastore-triggered schedule:

Schedule: 
      description: "Test create with datastore" 
      recurrence: ~ 
      pipeline_parameters: {} 
      wait_for_provisioning: True 
      wait_timeout: 3600 
      datastore_name: "workspaceblobstore" 
      polling_interval: 5 
      data_path_parameter_name: "input_data" 
      continue_on_step_failure: None 
      path_on_datastore: "file/path" 

定义重复性计划时,请在 recurrence 下使用以下键:When defining a recurring schedule, use the following keys under recurrence:

YAML 键YAML key 说明Description
frequency 计划的重复频率。How often the schedule recurs. 有效值为 "Minute""Hour""Day""Week""Month"Valid values are "Minute", "Hour", "Day", "Week", or "Month".
interval 计划的激发频率。How often the schedule fires. 整数值为再次激发计划之前要等待的时间单位数。The integer value is the number of time units to wait until the schedule fires again.
start_time 计划的开始时间。The start time for the schedule. 该值的字符串格式为 YYYY-MM-DDThh:mm:ssThe string format of the value is YYYY-MM-DDThh:mm:ss. 如果未提供开始时间,则第一个工作负荷会即时运行,以后的工作负荷将按计划运行。If no start time is provided, the first workload is run instantly and future workloads are run based on the schedule. 如果开始时间是过去的时间,则第一个工作负荷将在下一个计算的运行时间运行。If the start time is in the past, the first workload is run at the next calculated run time.
time_zone 开始时间所用的时区。The time zone for the start time. 如果未提供时区,将使用 UTC。If no time zone is provided, UTC is used.
hours 如果 frequency"Day""Week",可以指定从 0 到 23 的一个或多个整数(用逗号分隔),表示一天中应运行管道的小时时间。If frequency is "Day" or "Week", you can specify one or more integers from 0 to 23, separated by commas, as the hours of the day when the pipeline should run. 只能使用 time_of_dayhoursminutesOnly time_of_day or hours and minutes can be used.
minutes 如果 frequency"Day""Week",可以指定从 0 到 59 的一个或多个整数(用逗号分隔),表示指定的小时中应运行管道的分钟时间。If frequency is "Day" or "Week", you can specify one or more integers from 0 to 59, separated by commas, as the minutes of the hour when the pipeline should run. 只能使用 time_of_dayhoursminutesOnly time_of_day or hours and minutes can be used.
time_of_day 如果 frequency"Day""Week",可以指定一天中要运行计划的时间。If frequency is "Day" or "Week", you can specify a time of day for the schedule to run. 该值的字符串格式为 hh:mmThe string format of the value is hh:mm. 只能使用 time_of_dayhoursminutesOnly time_of_day or hours and minutes can be used.
week_days 如果 frequency"Week",可以指定应运行计划的一个或多个星期日期(用逗号分隔)。If frequency is "Week", you can specify one or more days, separated by commas, when the schedule should run. 有效值为 "Monday""Tuesday""Wednesday""Thursday""Friday""Saturday""Sunday"Valid values are "Monday", "Tuesday", "Wednesday", "Thursday", "Friday", "Saturday", and "Sunday".

以下示例包含重复性计划的定义:The following example contains the definition for a recurring schedule:

Schedule: 
    description: "Test create with recurrence" 
    recurrence: 
        frequency: Week # Can be "Minute", "Hour", "Day", "Week", or "Month". 
        interval: 1 # how often fires 
        start_time: 2019-06-07T10:50:00 
        time_zone: UTC 
        hours: 
        - 1 
        minutes: 
        - 0 
        time_of_day: null 
        week_days: 
        - Friday 
    pipeline_parameters: 
        'a': 1 
    wait_for_provisioning: True 
    wait_timeout: 3600 
    datastore_name: ~ 
    polling_interval: ~ 
    data_path_parameter_name: ~ 
    continue_on_step_failure: None 
    path_on_datastore: ~ 

后续步骤Next steps

了解如何使用 Azure 机器学习的 CLI 扩展Learn how to use the CLI extension for Azure Machine Learning.