Class CustomContainerTrainingJob (1.13.0)

CustomContainerTrainingJob(
    display_name: str,
    container_uri: str,
    command: Optional[Sequence[str]] = None,
    model_serving_container_image_uri: Optional[str] = None,
    model_serving_container_predict_route: Optional[str] = None,
    model_serving_container_health_route: Optional[str] = None,
    model_serving_container_command: Optional[Sequence[str]] = None,
    model_serving_container_args: Optional[Sequence[str]] = None,
    model_serving_container_environment_variables: Optional[Dict[str, str]] = None,
    model_serving_container_ports: Optional[Sequence[int]] = None,
    model_description: Optional[str] = None,
    model_instance_schema_uri: Optional[str] = None,
    model_parameters_schema_uri: Optional[str] = None,
    model_prediction_schema_uri: Optional[str] = None,
    project: Optional[str] = None,
    location: Optional[str] = None,
    credentials: Optional[google.auth.credentials.Credentials] = None,
    labels: Optional[Dict[str, str]] = None,
    training_encryption_spec_key_name: Optional[str] = None,
    model_encryption_spec_key_name: Optional[str] = None,
    staging_bucket: Optional[str] = None,
)

Class to launch a Custom Training Job in Vertex AI using a Container.

Inheritance

builtins.object > google.cloud.aiplatform.base.VertexAiResourceNoun > builtins.object > google.cloud.aiplatform.base.FutureManager > google.cloud.aiplatform.base.VertexAiResourceNounWithFutureManager > builtins.object > abc.ABC > google.cloud.aiplatform.base.DoneMixin > google.cloud.aiplatform.base.StatefulResource > google.cloud.aiplatform.base.VertexAiStatefulResource > google.cloud.aiplatform.training_jobs._TrainingJob > google.cloud.aiplatform.training_jobs._CustomTrainingJob > CustomContainerTrainingJob

Methods

CustomContainerTrainingJob

CustomContainerTrainingJob(
    display_name: str,
    container_uri: str,
    command: Optional[Sequence[str]] = None,
    model_serving_container_image_uri: Optional[str] = None,
    model_serving_container_predict_route: Optional[str] = None,
    model_serving_container_health_route: Optional[str] = None,
    model_serving_container_command: Optional[Sequence[str]] = None,
    model_serving_container_args: Optional[Sequence[str]] = None,
    model_serving_container_environment_variables: Optional[Dict[str, str]] = None,
    model_serving_container_ports: Optional[Sequence[int]] = None,
    model_description: Optional[str] = None,
    model_instance_schema_uri: Optional[str] = None,
    model_parameters_schema_uri: Optional[str] = None,
    model_prediction_schema_uri: Optional[str] = None,
    project: Optional[str] = None,
    location: Optional[str] = None,
    credentials: Optional[google.auth.credentials.Credentials] = None,
    labels: Optional[Dict[str, str]] = None,
    training_encryption_spec_key_name: Optional[str] = None,
    model_encryption_spec_key_name: Optional[str] = None,
    staging_bucket: Optional[str] = None,
)

Constructs a Custom Container Training Job.

job = aiplatform.CustomContainerTrainingJob( display_name='test-train', container_uri='gcr.io/my_project_id/my_image_name:tag', command=['python3', 'run_script.py'] model_serving_container_image_uri='gcr.io/my-trainer/serving:1', model_serving_container_predict_route='predict', model_serving_container_health_route='metadata, labels={'key': 'value'}, )

Usage with Dataset:

ds = aiplatform.TabularDataset( 'projects/my-project/locations/us-central1/datasets/12345')

job.run( ds, replica_count=1, model_display_name='my-trained-model', model_labels={'key': 'value'}, )

Usage without Dataset:

job.run(replica_count=1, model_display_name='my-trained-model)

TODO(b/169782082) add documentation about traning utilities To ensure your model gets saved in Vertex AI, write your saved model to os.environ["AIP_MODEL_DIR"] in your provided training script.

Parameters
NameDescription
display_name str

Required. The user-defined name of this TrainingPipeline.

container_uri str

Required: Uri of the training container image in the GCR.

command Sequence[str]

The command to be invoked when the container is started. It overrides the entrypoint instruction in Dockerfile when provided

model_serving_container_image_uri str

If the training produces a managed Vertex AI Model, the URI of the Model serving container suitable for serving the model produced by the training script.

model_serving_container_predict_route str

If the training produces a managed Vertex AI Model, An HTTP path to send prediction requests to the container, and which must be supported by it. If not specified a default HTTP path will be used by Vertex AI.

model_serving_container_health_route str

If the training produces a managed Vertex AI Model, an HTTP path to send health check requests to the container, and which must be supported by it. If not specified a standard HTTP path will be used by AI Platform.

model_serving_container_command Sequence[str]

The command with which the container is run. Not executed within a shell. The Docker image's ENTRYPOINT is used if this is not provided. Variable references $(VAR_NAME) are expanded using the container's environment. If a variable cannot be resolved, the reference in the input string will be unchanged. The $(VAR_NAME) syntax can be escaped with a double $$, ie: $$(VAR_NAME). Escaped references will never be expanded, regardless of whether the variable exists or not.

model_serving_container_args Sequence[str]

The arguments to the command. The Docker image's CMD is used if this is not provided. Variable references $(VAR_NAME) are expanded using the container's environment. If a variable cannot be resolved, the reference in the input string will be unchanged. The $(VAR_NAME) syntax can be escaped with a double $$, ie: $$(VAR_NAME). Escaped references will never be expanded, regardless of whether the variable exists or not.

model_serving_container_environment_variables Dict[str, str]

The environment variables that are to be present in the container. Should be a dictionary where keys are environment variable names and values are environment variable values for those names.

model_serving_container_ports Sequence[int]

Declaration of ports that are exposed by the container. This field is primarily informational, it gives Vertex AI information about the network connections the container uses. Listing or not a port here has no impact on whether the port is actually exposed, any port listening on the default "0.0.0.0" address inside a container will be accessible from the network.

model_description str

The description of the Model.

model_instance_schema_uri str

Optional. Points to a YAML file stored on Google Cloud Storage describing the format of a single instance, which are used in PredictRequest.instances, ExplainRequest.instances and BatchPredictionJob.input_config. The schema is defined as an OpenAPI 3.0.2 Schema Object <https://tinyurl.com/y538mdwt#schema-object>__. AutoML Models always have this field populated by AI Platform. Note: The URI given on output will be immutable and probably different, including the URI scheme, than the one given on input. The output URI will point to a location where the user only has a read access.

model_parameters_schema_uri str

Optional. Points to a YAML file stored on Google Cloud Storage describing the parameters of prediction and explanation via PredictRequest.parameters, ExplainRequest.parameters and BatchPredictionJob.model_parameters. The schema is defined as an OpenAPI 3.0.2 Schema Object <https://tinyurl.com/y538mdwt#schema-object>__. AutoML Models always have this field populated by AI Platform, if no parameters are supported it is set to an empty string. Note: The URI given on output will be immutable and probably different, including the URI scheme, than the one given on input. The output URI will point to a location where the user only has a read access.

model_prediction_schema_uri str

Optional. Points to a YAML file stored on Google Cloud Storage describing the format of a single prediction produced by this Model, which are returned via PredictResponse.predictions, ExplainResponse.explanations, and BatchPredictionJob.output_config. The schema is defined as an OpenAPI 3.0.2 Schema Object <https://tinyurl.com/y538mdwt#schema-object>__. AutoML Models always have this field populated by AI Platform. Note: The URI given on output will be immutable and probably different, including the URI scheme, than the one given on input. The output URI will point to a location where the user only has a read access.

project str

Project to run training in. Overrides project set in aiplatform.init.

location str

Location to run training in. Overrides location set in aiplatform.init.

credentials auth_credentials.Credentials

Custom credentials to use to run call training service. Overrides credentials set in aiplatform.init.

labels Dict[str, str]

Optional. The labels with user-defined metadata to organize TrainingPipelines. Label keys and values can be no longer than 64 characters (Unicode codepoints), can only contain lowercase letters, numeric characters, underscores and dashes. International characters are allowed. See https://goo.gl/xmQnxf for more information and examples of labels.

training_encryption_spec_key_name Optional[str]

Optional. The Cloud KMS resource identifier of the customer managed encryption key used to protect the training pipeline. Has the form: projects/my-project/locations/my-region/keyRings/my-kr/cryptoKeys/my-key. The key needs to be in the same region as where the compute resource is created. If set, this TrainingPipeline will be secured by this key. Note: Model trained by this TrainingPipeline is also secured by this key if model_to_upload is not set separately. Overrides encryption_spec_key_name set in aiplatform.init.

model_encryption_spec_key_name Optional[str]

Optional. The Cloud KMS resource identifier of the customer managed encryption key used to protect the model. Has the form: projects/my-project/locations/my-region/keyRings/my-kr/cryptoKeys/my-key. The key needs to be in the same region as where the compute resource is created. If set, the trained Model will be secured by this key. Overrides encryption_spec_key_name set in aiplatform.init.

staging_bucket str

Bucket used to stage source and training artifacts. Overrides staging_bucket set in aiplatform.init.

run

run(
    dataset: Optional[
        Union[
            google.cloud.aiplatform.datasets.image_dataset.ImageDataset,
            google.cloud.aiplatform.datasets.tabular_dataset.TabularDataset,
            google.cloud.aiplatform.datasets.text_dataset.TextDataset,
            google.cloud.aiplatform.datasets.video_dataset.VideoDataset,
        ]
    ] = None,
    annotation_schema_uri: Optional[str] = None,
    model_display_name: Optional[str] = None,
    model_labels: Optional[Dict[str, str]] = None,
    base_output_dir: Optional[str] = None,
    service_account: Optional[str] = None,
    network: Optional[str] = None,
    bigquery_destination: Optional[str] = None,
    args: Optional[List[Union[float, int, str]]] = None,
    environment_variables: Optional[Dict[str, str]] = None,
    replica_count: int = 1,
    machine_type: str = "n1-standard-4",
    accelerator_type: str = "ACCELERATOR_TYPE_UNSPECIFIED",
    accelerator_count: int = 0,
    boot_disk_type: str = "pd-ssd",
    boot_disk_size_gb: int = 100,
    reduction_server_replica_count: int = 0,
    reduction_server_machine_type: Optional[str] = None,
    reduction_server_container_uri: Optional[str] = None,
    training_fraction_split: Optional[float] = None,
    validation_fraction_split: Optional[float] = None,
    test_fraction_split: Optional[float] = None,
    training_filter_split: Optional[str] = None,
    validation_filter_split: Optional[str] = None,
    test_filter_split: Optional[str] = None,
    predefined_split_column_name: Optional[str] = None,
    timestamp_split_column_name: Optional[str] = None,
    timeout: Optional[int] = None,
    restart_job_on_worker_restart: bool = False,
    enable_web_access: bool = False,
    tensorboard: Optional[str] = None,
    sync=True,
    create_request_timeout: Optional[float] = None,
)

Runs the custom training job.

Distributed Training Support: If replica count = 1 then one chief replica will be provisioned. If replica_count > 1 the remainder will be provisioned as a worker replica pool. ie: replica_count = 10 will result in 1 chief and 9 workers All replicas have same machine_type, accelerator_type, and accelerator_count

If training on a Vertex AI dataset, you can use one of the following split configurations: Data fraction splits: Any of training_fraction_split, validation_fraction_split and test_fraction_split may optionally be provided, they must sum to up to 1. If the provided ones sum to less than 1, the remainder is assigned to sets as decided by Vertex AI. If none of the fractions are set, by default roughly 80% of data will be used for training, 10% for validation, and 10% for test.

Data filter splits:
Assigns input data to training, validation, and test sets
based on the given filters, data pieces not matched by any
filter are ignored. Currently only supported for Datasets
containing DataItems.
If any of the filters in this message are to match nothing, then
they can be set as '-' (the minus sign).
If using filter splits, all of ``training_filter_split``, ``validation_filter_split`` and
``test_filter_split`` must be provided.
Supported only for unstructured Datasets.

Predefined splits:
Assigns input data to training, validation, and test sets based on the value of a provided key.
If using predefined splits, ``predefined_split_column_name`` must be provided.
Supported only for tabular Datasets.

Timestamp splits:
Assigns input data to training, validation, and test sets
based on a provided timestamps. The youngest data pieces are
assigned to training set, next to validation set, and the oldest
to the test set.
Supported only for tabular Datasets.
Parameters
NameDescription
dataset Union[datasets.ImageDataset,datasets.TabularDataset,datasets.TextDataset,datasets.VideoDataset]

Vertex AI to fit this training against. Custom training script should retrieve datasets through passed in environment variables uris: os.environ["AIP_TRAINING_DATA_URI"] os.environ["AIP_VALIDATION_DATA_URI"] os.environ["AIP_TEST_DATA_URI"] Additionally the dataset format is passed in as: os.environ["AIP_DATA_FORMAT"]

annotation_schema_uri str

Google Cloud Storage URI points to a YAML file describing annotation schema. The schema is defined as an OpenAPI 3.0.2 Schema Object The schema files that can be used here are found in gs://google-cloud-aiplatform/schema/dataset/annotation/, note that the chosen schema must be consistent with metadata of the Dataset specified by dataset_id. Only Annotations that both match this schema and belong to DataItems not ignored by the split method are used in respectively training, validation or test role, depending on the role of the DataItem they are on. When used in conjunction with annotations_filter, the Annotations used for training are filtered by both annotations_filter and annotation_schema_uri.

model_display_name str

If the script produces a managed Vertex AI Model. The display name of the Model. The name can be up to 128 characters long and can be consist of any UTF-8 characters. If not provided upon creation, the job's display_name is used.

model_labels Dict[str, str]

Optional. The labels with user-defined metadata to organize your Models. Label keys and values can be no longer than 64 characters (Unicode codepoints), can only contain lowercase letters, numeric characters, underscores and dashes. International characters are allowed. See https://goo.gl/xmQnxf for more information and examples of labels.

base_output_dir str

GCS output directory of job. If not provided a timestamped directory in the staging directory will be used. Vertex AI sets the following environment variables when it runs your training code: - AIP_MODEL_DIR: a Cloud Storage URI of a directory intended for saving model artifacts, i.e. <base_output_dir>/model/ - AIP_CHECKPOINT_DIR: a Cloud Storage URI of a directory intended for saving checkpoints, i.e. <base_output_dir>/checkpoints/ - AIP_TENSORBOARD_LOG_DIR: a Cloud Storage URI of a directory intended for saving TensorBoard logs, i.e. <base_output_dir>/logs/

service_account str

Specifies the service account for workload run-as account. Users submitting jobs must have act-as permission on this run-as account.

network str

The full name of the Compute Engine network to which the job should be peered. For example, projects/12345/global/networks/myVPC. Private services access must already be configured for the network. If left unspecified, the job is not peered with any network.

bigquery_destination str

Provide this field if dataset is a BiqQuery dataset. The BigQuery project location where the training data is to be written to. In the given project a new dataset is created with name dataset_<dataset-id>_<annotation-type>_<timestamp-of-training-call> where timestamp is in YYYY_MM_DDThh_mm_ss_sssZ format. All training input data will be written into that dataset. In the dataset three tables will be created, training, validation and test. - AIP_DATA_FORMAT = "bigquery". - AIP_TRAINING_DATA_URI ="bigquery_destination.dataset_.training" - AIP_VALIDATION_DATA_URI = "bigquery_destination.dataset_.validation" - AIP_TEST_DATA_URI = "bigquery_destination.dataset_*.test"

args List[Unions[str, int, float]]

Command line arguments to be passed to the Python script.

environment_variables Dict[str, str]

Environment variables to be passed to the container. Should be a dictionary where keys are environment variable names and values are environment variable values for those names. At most 10 environment variables can be specified. The Name of the environment variable must be unique. environment_variables = { 'MY_KEY': 'MY_VALUE' }

replica_count int

The number of worker replicas. If replica count = 1 then one chief replica will be provisioned. If replica_count > 1 the remainder will be provisioned as a worker replica pool.

machine_type str

The type of machine to use for training.

accelerator_type str

Hardware accelerator type. One of ACCELERATOR_TYPE_UNSPECIFIED, NVIDIA_TESLA_K80, NVIDIA_TESLA_P100, NVIDIA_TESLA_V100, NVIDIA_TESLA_P4, NVIDIA_TESLA_T4

accelerator_count int

The number of accelerators to attach to a worker replica.

boot_disk_type str

Type of the boot disk, default is pd-ssd. Valid values: pd-ssd (Persistent Disk Solid State Drive) or pd-standard (Persistent Disk Hard Disk Drive).

boot_disk_size_gb int

Size in GB of the boot disk, default is 100GB. boot disk size must be within the range of [100, 64000].

reduction_server_replica_count int

The number of reduction server replicas, default is 0.

reduction_server_machine_type str

Optional. The type of machine to use for reduction server.

reduction_server_container_uri str

Optional. The Uri of the reduction server container image. See details: https://cloud.google.com/vertex-ai/docs/training/distributed-training#reduce_training_time_with_reduction_server

training_fraction_split float

Optional. The fraction of the input data that is to be used to train the Model. This is ignored if Dataset is not provided.

validation_fraction_split float

Optional. The fraction of the input data that is to be used to validate the Model. This is ignored if Dataset is not provided.

test_fraction_split float

Optional. The fraction of the input data that is to be used to evaluate the Model. This is ignored if Dataset is not provided.

training_filter_split str

Optional. A filter on DataItems of the Dataset. DataItems that match this filter are used to train the Model. A filter with same syntax as the one used in DatasetService.ListDataItems may be used. If a single DataItem is matched by more than one of the FilterSplit filters, then it is assigned to the first set that applies to it in the training, validation, test order. This is ignored if Dataset is not provided.

validation_filter_split str

Optional. A filter on DataItems of the Dataset. DataItems that match this filter are used to validate the Model. A filter with same syntax as the one used in DatasetService.ListDataItems may be used. If a single DataItem is matched by more than one of the FilterSplit filters, then it is assigned to the first set that applies to it in the training, validation, test order. This is ignored if Dataset is not provided.

test_filter_split str

Optional. A filter on DataItems of the Dataset. DataItems that match this filter are used to test the Model. A filter with same syntax as the one used in DatasetService.ListDataItems may be used. If a single DataItem is matched by more than one of the FilterSplit filters, then it is assigned to the first set that applies to it in the training, validation, test order. This is ignored if Dataset is not provided.

predefined_split_column_name str

Optional. The key is a name of one of the Dataset's data columns. The value of the key (either the label's value or value in the column) must be one of {training, validation, test}, and it defines to which set the given piece of data is assigned. If for a piece of data the key is not present or has an invalid value, that piece is ignored by the pipeline. Supported only for tabular and time series Datasets.

timestamp_split_column_name str

Optional. The key is a name of one of the Dataset's data columns. The value of the key values of the key (the values in the column) must be in RFC 3339 date-time format, where time-offset = "Z" (e.g. 1985-04-12T23:20:50.52Z). If for a piece of data the key is not present or has an invalid value, that piece is ignored by the pipeline. Supported only for tabular and time series Datasets.

timeout int

The maximum job running time in seconds. The default is 7 days.

restart_job_on_worker_restart bool

Restarts the entire CustomJob if a worker gets restarted. This feature can be used by distributed training jobs that are not resilient to workers leaving and joining a job.

enable_web_access bool

Whether you want Vertex AI to enable interactive shell access to training containers. https://cloud.google.com/vertex-ai/docs/training/monitor-debug-interactive-shell

tensorboard str

Optional. The name of a Vertex AI Tensorboard resource to which this CustomJob will upload Tensorboard logs. Format: projects/{project}/locations/{location}/tensorboards/{tensorboard} The training script should write Tensorboard to following Vertex AI environment variable: AIP_TENSORBOARD_LOG_DIR service_account is required with provided tensorboard. For more information on configuring your service account please visit: https://cloud.google.com/vertex-ai/docs/experiments/tensorboard-training

create_request_timeout float

Optional. The timeout for the create request in seconds.

sync bool

Whether to execute this method synchronously. If False, this method will be executed in concurrent Future and any downstream object will be immediately returned and synced when the Future has completed.

Exceptions
TypeDescription
RuntimeErrorIf Training job has already been run, staging_bucket has not been set, or model_display_name was provided but required arguments were not provided in constructor.
Returns
TypeDescription
modelThe trained Vertex AI Model resource or None if training did not produce a Vertex AI Model.