Submit a training job in the Azure Machine Learning studio

There are many ways to create a training job with Azure Machine Learning. You can train models by using the Azure CLI, the REST API, or you can use the studio UI to directly create a training job.

In this article, you learn how to use your own data and code to train a machine learning model with a guided experience for submitting training jobs in the Azure Machine Learning studio.

Prerequisites

• An Azure subscription. If you don't have an Azure subscription, create a free Azure account before you begin.

• An Azure Machine Learning workspace. To learn how, see Create workspace resources.

• An understanding of what a training job is in Azure Machine Learning. To learn more, see how to train models.

Get started

1. Sign in to the Azure Machine Learning studio, then select your subscription and workspace.

2. Enter the job creation UI from the homepage. Select + New and choose Training job.

   

3. Select your method of training, then choose Start configuring job to open the submission form.

   

   In the next section, we walk through the form with the steps for running a custom training script (command job).

Configure basic settings

Configure the basic information about your training job. You can proceed to the next page if you're satisfied with the defaults we chose for you, or make changes based on your desired preference.

The following fields are available:

Upload training script

On the next page, upload your source code. Configure any inputs or outputs required to execute the training job, and specify the command to execute your training script.

You can use a code file or a folder from your local machine or workspace's default blob storage. Azure shows the files to be uploaded after you make the selection.

Code

The command is run from the root directory of the uploaded code folder. After you select your code file or folder, you can see the files to be uploaded. Copy the relative path to the code containing your entry point and paste it into the box labeled Enter the command to start the job.

If the code is in the root directory, you can directly refer to it in the command. For instance, python main.py.

If the code isn't in the root directory, you should use the relative path. For example, the structure of the word language model is:

.
├── job.yml
├── data
└── src
    └── main.py

Here, the source code is in the src subdirectory. The command would be python ./src/main.py (plus other command-line arguments).

Inputs

When you use an input in the command, you need to specify the input name. To indicate an input variable, use the form ${{inputs.input_name}}. For instance, ${{inputs.wiki}}. You can then refer to it in the command, for instance, --data ${{inputs.wiki}}.

Select compute resources

On the next page, select the compute target on which you'd like your job to run. The job creation UI supports several compute types:

1. Select a compute type.

2. Select an existing compute resource. The dropdown shows the node information and SKU type to help your choice.

3. For a compute cluster or a Kubernetes cluster, you can also specify how many nodes you want for the job in Instance count. The default number of instances is 1.

4. When you're satisfied with your choices, choose Next.

If you're using Azure Machine Learning for the first time, you see an empty list and a link to create a new compute. For more information on creating the various types, see:

Specify the necessary environment

After selecting a compute target, you need to specify the runtime environment for your job. The job creation UI supports three types of environment:

• Curated environments
• Custom environments
• Container registry image

Curated environments

Curated environments are Azure-defined collections of Python packages used in common ML workloads. Curated environments are available in your workspace by default. These environments are backed by cached Docker images, which reduce the job preparation overhead. The cards displayed in the Curated environments page show details of each environment. To learn more, see Azure Machine Learning Curated Environments.

Custom environments

Custom environments are environments that you specify. You can specify an environment or reuse an environment that you already created. To learn more, see Manage software environments in Azure Machine Learning studio.

Container registry image

If you don't want to use the Azure Machine Learning curated environments or specify your own custom environment, you can use a docker image from a public container registry such as Docker Hub.

Review and create

After you configure the job, choose Next to go to the Review page. To modify a setting, choose the pencil icon and make the change.

To launch the job, choose Submit training job. After the job is created, Azure shows you the job details page, where you can monitor and manage your training job.

How to configure emails in the studio

To start receiving emails when your job, online endpoint, or batch endpoint is complete or if there's an issue (failed, canceled), use the following steps:

1. In Azure ML studio, go to settings by selecting the gear icon.
2. Select the Email notifications tab.
3. Toggle to enable or disable email notifications for a specific event.

Related content

• Deploy and score a machine learning model by using an online endpoint
• Train models with Azure Machine Learning CLI, SDK, and REST API