Running a training job

AI Platform Training provides model training as an asynchronous (batch) service. This page describes how to configure and submit a training job by running gcloud ai-platform jobs submit training from the command line or by sending a request to the API at projects.jobs.create.

Before you begin

Before you can submit a training job, you must package your application and upload it and any unusual dependencies to a Cloud Storage bucket. Note: If you use the Google Cloud CLI to submit your job, you can package the application and submit the job in the same step.

Configuring the job

You pass your parameters to the training service by setting the members of the Job resource, which includes the items in the TrainingInput resource.

If you use the Google Cloud CLI to submit your training jobs, you can:

  • Specify the most common training parameters as flags of the gcloud ai-platform jobs submit training command.
  • Pass the remaining parameters in a YAML configuration file, named config.yaml by convention. The configuration file mirrors the structure of the JSON representation of the Job resource. You pass the path of your configuration file in the --config flag of the gcloud ai-platform jobs submit training command. So, if the path to your configuration file is config.yaml, you must set --config=config.yaml.

Gathering the job configuration data

The following properties are used to define your job.

Job name (jobId)
A name to use for the job (mixed-case letters, numbers, and underscores only, starting with a letter).
Cluster configuration (scaleTier)
A scale tier specifying the type of processing cluster to run your job on. This can be the CUSTOM scale tier, in which case you also explicitly specify the number and type of machines to use.
Disk configuration (diskConfig)
Configuration of the boot disk for each training VM. This field is optional; by default, each VM runs with a 100 GB pd-ssd boot disk. Specifying this field might incur extra disk charges.
Training application package (packageUris)
A packaged training application that is staged in a Cloud Storage location. If you are using the Google Cloud CLI, the application packaging step is largely automated. See the details in the guide to packaging your application.
Module name (pythonModule)
The name of the main module in your package. The main module is the Python file you call to start the application. If you use the gcloud command to submit your job, specify the main module name in the --module-name flag. See the guide to packaging your application.
Region (region)
The Compute Engine region where you want your job to run. You should run your training job in the same region as the Cloud Storage bucket that stores your training data. See the available regions for AI Platform Training services.
Job directory (jobDir)
The path to a Cloud Storage location to use for job output. Most training applications save checkpoints during training and save the trained model to a file at the end of the job. You need a Cloud Storage location to save them to. Your Google Cloud project must have write access to this bucket. The training service automatically passes the path you set for the job directory to your training application as a command-line argument named job_dir. You can parse it along with your application's other arguments and use it in your code. The advantage to using the job directory is that the training service validates the directory before starting your application.
Runtime version (runtimeVersion)
The AI Platform Training runtime version to use for the job.
Python version (pythonVersion)
The Python version to use for the job. Python 3.5 is available in runtime versions 1.13 through 1.14. Python 3.7 is available in runtime versions 1.15 and later.
Maximum wait time (scheduling.maxWaitTime)
A maximum waiting duration in seconds with the suffix s (for example, 3600s) determining how long you allow your job to remain in the QUEUED and PREPARING states. AI Platform Training does not always start running your job immediately due to resource constraints; specify this field if you are not willing to wait longer than a certain duration for the job to run. The limited duration starts when you create the job. If the job has not yet entered the RUNNING state by the end of this period, AI Platform Training cancels the job. This field is optional and it defaults to no limit. If you specify this field, you must set the value to at least 1800s (30 minutes).
Maximum running time (scheduling.maxRunningTime)
A maximum running duration in seconds with the suffix s (for example, 7200s) for your training job. The limited duration starts when the job enters the RUNNING state. If the job is still running after this amount of time, AI Platform Training cancels the job. This field is optional and it defaults to seven days (604800s).
Service account (serviceAccount)
The email address of a service account for AI Platform Training to use when it runs your training application. This can provide your training application access to Google Cloud resources without granting direct access to your project's AI Platform Google-managed service account. This field is optional. Learn more about the requirements for custom service accounts.

Formatting your configuration parameters

How you specify your configuration details depends on how you are starting your training job:

gcloud

Provide the job configuration details to the gcloud ai-platform jobs submit training command. You can do this in two ways:

  • With command-line flags.
  • In a YAML file representing the Job resource. You can name this file whatever you want. By convention the name is config.yaml.

Even if you use a YAML file, certain details must be supplied as command-line flags. For example, you must provide the --module-name flag and at least one of --package-path or --packages. If you use --package-path, you must also include --job-dir or --staging-bucket. Additionally, you must either provide the --region flag or set a default region for your gcloud client. These options—and any others you provide as command line flags—will override values for those options in your configuration file.

Example 1: In this example, you choose a preconfigured machine cluster and supply all the required details as command-line flags when submitting the job. No configuration file is necessary. See the guide to submitting the job in the next section.

Example 2: The following example shows the contents of the configuration file for a job with a custom processing cluster. The configuration file includes some but not all of the configuration details, assuming that you supply the other required details as command-line flags when submitting the job.

trainingInput:
  scaleTier: CUSTOM
  masterType: complex_model_m
  workerType: complex_model_m
  parameterServerType: large_model
  workerCount: 9
  parameterServerCount: 3
  runtimeVersion: '2.11'
  pythonVersion: '3.7'
  scheduling:
    maxWaitTime: 3600s
    maxRunningTime: 7200s

The preceding example specifies Python version 3.7, which is available when you use AI Platform Training runtime version 1.15 or later. It also configures worker and parameter server virtual machines; only configure these machines if your performs distributed training using TensorFlow or custom containers. Read more about machine types.

Python

When you submit a training job using the Google API Client Library for Python, set your configuration in a dictionary with the same structure as the Job resource. This takes the form of a dictionary with two keys: jobId and trainingInput, with their respective data being the name for the job and a second dictionary with keys for the objects in the TrainingInput resource.

The following example shows how to build a Job representation for a job with a custom processing cluster.

training_inputs = {
    'scaleTier': 'CUSTOM',
    'masterType': 'complex_model_m',
    'workerType': 'complex_model_m',
    'parameterServerType': 'large_model',
    'workerCount': 9,
    'parameterServerCount': 3,
    'packageUris': ['gs://my/trainer/path/package-0.0.0.tar.gz'],
    'pythonModule': 'trainer.task',
    'args': ['--arg1', 'value1', '--arg2', 'value2'],
    'region': 'us-central1',
    'jobDir': 'gs://my/training/job/directory',
    'runtimeVersion': '2.11',
    'pythonVersion': '3.7',
    'scheduling': {'maxWaitTime': '3600s', 'maxRunningTime': '7200s'},
}

job_spec = {'jobId': 'my_job_name', 'trainingInput': training_inputs}

Note that training_inputs and job_spec are arbitrary identifiers: you can name these dictionaries whatever you want. However, the dictionary keys must be named exactly as shown, to match the names in the Job and TrainingInput resources.

The preceding example specifies Python version 3.7, which is available when you use AI Platform Training runtime version 1.15 or later. It also configures worker and parameter server virtual machines; only configure these machines if your performs distributed training using TensorFlow or custom containers. Read more about machine types.

Submitting the job

When submitting a training job, you specify two sets of flags:

  • Job configuration parameters. AI Platform Training needs these values to set up resources in the cloud and deploy your application on each node in the processing cluster.
  • User arguments, or application parameters. AI Platform Training passes the value of these flags through to your application.

Create your job:

gcloud

Submit a training job using the gcloud ai-platform jobs submit training command.

First, it's useful to define some environment variables containing your configuration details. To create a job name, the following code appends the date and time to the model name:

PACKAGE_PATH="/path/to/your/application/sources"
now=$(date +"%Y%m%d_%H%M%S")
JOB_NAME="your_name_$now"
MODULE_NAME="trainer.task"
JOB_DIR="gs://your/chosen/job/output/path"
REGION="us-east1"
RUNTIME_VERSION="2.11"

The following job submission corresponds to configuration example 1 above, where you choose a preconfigured scale tier (basic) and you decide to supply all the configuration details via command-line flags. There is no need for a config.yaml file:

gcloud ai-platform jobs submit training $JOB_NAME \
        --scale-tier basic \
        --package-path $PACKAGE_PATH \
        --module-name $MODULE_NAME \
        --job-dir $JOB_DIR \
        --region $REGION \
        -- \
        --user_first_arg=first_arg_value \
        --user_second_arg=second_arg_value

The following job submission corresponds to configuration example 2 above, where some of the configuration is in the file and you supply the other details via command-line flags:

gcloud ai-platform jobs submit training $JOB_NAME \
        --package-path $PACKAGE_PATH \
        --module-name $MODULE_NAME \
        --job-dir $JOB_DIR \
        --region $REGION \
        --config config.yaml \
        -- \
        --user_first_arg=first_arg_value \
        --user_second_arg=second_arg_value

Notes:

  • If you specify an option both in your configuration file (config.yaml) and as a command-line flag, the value on the command line overrides the value in the configuration file.
  • The empty -- flag marks the end of the gcloud specific flags and the start of the USER_ARGS that you want to pass to your application.
  • Flags specific to AI Platform Training, such as --module-name, --runtime-version, and --job-dir, must come before the empty -- flag. The AI Platform Training service interprets these flags.
  • The --job-dir flag, if specified, must come before the empty -- flag, because AI Platform Training uses the --job-dir to validate the path.
  • Your application must handle the --job-dir flag too, if specified. Even though the flag comes before the empty --, the --job-dir is also passed to your application as a command-line flag.
  • You can define as many USER_ARGS as you need. AI Platform Training passes --user_first_arg, --user_second_arg, and so on, through to your application.

Python

You can use the Google API Client Library for Python to call the AI Platform Training and Prediction API without manually constructing HTTP requests. Before you run the following code sample, you must set up authentication.

  1. Save your project ID in the format the APIs need ('projects/_projectname'):

    project_name = 'my_project_name'
    project_id = 'projects/{}'.format(project_name)
    
  2. Get a Python representation of the AI Platform Training services:

    cloudml = discovery.build('ml', 'v1')
    
  3. Form your request and send it. Note that job_spec was created in the previous step where you formatted the configuration parameters

    request = cloudml.projects().jobs().create(body=job_spec,
                  parent=project_id)
    response = request.execute()
    
  4. Catch any HTTP errors. The simplest way is to put the previous command in a try block:

    try:
        response = request.execute()
        # You can put your code for handling success (if any) here.
    
    except errors.HttpError, err:
        # Do whatever error response is appropriate for your application.
        # For this example, just send some text to the logs.
        # You need to import logging for this to work.
        logging.error('There was an error creating the training job.'
                      ' Check the details:')
        logging.error(err._get_reason())
    

What's next