def main(argv=None):
parser = create_parser()
args = parser.parse_args(argv)
logging.getLogger().setLevel(logging.INFO)
client = _utils.get_sagemaker_client(args.region, args.endpoint_url)
logging.info('Submitting Training Job to SageMaker...')
job_name = _utils.create_training_job(client, vars(args))
logging.info('Job request submitted. Waiting for completion...')
_utils.wait_for_training_job(client, job_name)
image = _utils.get_image_from_job(client, job_name)
model_artifact_url = _utils.get_model_artifacts_from_job(client, job_name)
logging.info('Get model artifacts %s from training job %s.',
model_artifact_url, job_name)
with open('/tmp/model_artifact_url.txt', 'w') as f:
f.write(model_artifact_url)
with open('/tmp/job_name.txt', 'w') as f:
f.write(job_name)
with open('/tmp/training_image.txt', 'w') as f:
f.write(image)
logging.info('Job completed.')
def test_create_training_job(self):
mock_client = MagicMock()
mock_args = self.parser.parse_args(required_args + ['--job_name', 'test-job'])
response = _utils.create_training_job(mock_client, vars(mock_args))
mock_client.create_training_job.assert_called_once_with(
AlgorithmSpecification={'TrainingImage': 'test-image', 'TrainingInputMode': 'File'},
EnableInterContainerTrafficEncryption=False,
EnableManagedSpotTraining=False,
EnableNetworkIsolation=True,
HyperParameters={},
InputDataConfig=[{'ChannelName': 'train',
'DataSource': {'S3DataSource': {'S3Uri': 's3://fake-bucket/data', 'S3DataType': 'S3Prefix', 'S3DataDistributionType': 'FullyReplicated'}},
'ContentType': '',
'CompressionType': 'None',
'RecordWrapperType': 'None',
'InputMode': 'File'
}],
OutputDataConfig={'KmsKeyId': '', 'S3OutputPath': 'test-path'},
ResourceConfig={'InstanceType': 'ml.m4.xlarge', 'InstanceCount': 1, 'VolumeSizeInGB': 50, 'VolumeKmsKeyId': ''},
RoleArn='arn:aws:iam::123456789012:user/Development/product_1234/*',
StoppingCondition={'MaxRuntimeInSeconds': 3600},
Tags=[],
TrainingJobName='test-job'
)
self.assertEqual(response, 'test-job')
def main(argv=None):
parser = create_parser()
args = parser.parse_args(argv)
logging.getLogger().setLevel(logging.INFO)
client = _utils.get_sagemaker_client(args.region, args.endpoint_url)
logging.info('Submitting Training Job to SageMaker...')
job_name = _utils.create_training_job(client, vars(args))
def signal_term_handler(signalNumber, frame):
_utils.stop_training_job(client, job_name)
logging.info(f"Training Job: {job_name} request submitted to Stop")
signal.signal(signal.SIGTERM, signal_term_handler)
logging.info('Job request submitted. Waiting for completion...')
try:
_utils.wait_for_training_job(client, job_name)
_utils.wait_for_debug_rules(client, job_name)
except:
raise
finally:
cw_client = _utils.get_cloudwatch_client(args.region)
_utils.print_logs_for_job(cw_client, '/aws/sagemaker/TrainingJobs', job_name)
image = _utils.get_image_from_job(client, job_name)
model_artifact_url = _utils.get_model_artifacts_from_job(client, job_name)
logging.info('Get model artifacts %s from training job %s.', model_artifact_url, job_name)
_utils.write_output(args.model_artifact_url_output_path, model_artifact_url)
_utils.write_output(args.job_name_output_path, job_name)
_utils.write_output(args.training_image_output_path, image)
logging.info('Job completed.')
def test_sagemaker_exception_in_create_training_job(self):
mock_client = MagicMock()
mock_exception = ClientError({"Error": {"Message": "SageMaker broke"}}, "create_training_job")
mock_client.create_training_job.side_effect = mock_exception
mock_args = self.parser.parse_args(required_args)
with self.assertRaises(Exception):
response = _utils.create_training_job(mock_client, vars(mock_args))
def main(argv=None):
parser = argparse.ArgumentParser(description='SageMaker Training Job')
parser.add_argument('--region',
type=str,
help='The region where the training job launches.')
parser.add_argument(
'--image',
type=str,
help=
'The registry path of the Docker image that contains the training algorithm.'
)
parser.add_argument('--instance_type',
type=str,
help='The ML compute instance type.')
parser.add_argument(
'--instance_count',
type=int,
help=
'The registry path of the Docker image that contains the training algorithm.'
)
parser.add_argument(
'--volume_size',
type=int,
help='The size of the ML storage volume that you want to provision.')
parser.add_argument(
'--dataset_path',
type=str,
help=
'The S3 location of the data source that is associated with a channel.'
)
parser.add_argument(
'--model_artifact_path',
type=str,
help=
'Identifies the S3 path where you want Amazon SageMaker to store the model artifacts.'
)
parser.add_argument(
'--role',
type=str,
help=
'The Amazon Resource Name (ARN) that Amazon SageMaker assumes to perform tasks on your behalf.'
)
args = parser.parse_args()
logging.getLogger().setLevel(logging.INFO)
client = _utils.get_client(args.region)
logging.info('Submitting Training Job to SageMaker...')
job_name = _utils.create_training_job(client, args.image,
args.instance_type,
args.instance_count,
args.volume_size, args.dataset_path,
args.model_artifact_path, args.role)
logging.info('Job request submitted. Waiting for completion...')
_utils.wait_for_training_job(client, job_name)
model_artifact_url = _utils.get_model_artifacts_from_job(client, job_name)
logging.info('Get model artifacts %s from training job %s.',
model_artifact_url, job_name)
with open('/tmp/model_artifact_url.txt', 'w') as f:
f.write(model_artifact_url)
with open('/tmp/job_name.txt', 'w') as f:
f.write(job_name)
logging.info('Job completed.')
def main(argv=None):
parser = argparse.ArgumentParser(description='SageMaker Training Job')
parser.add_argument('--region',
type=str.strip,
required=True,
help='The region where the training job launches.')
parser.add_argument('--job_name',
type=str.strip,
required=False,
help='The name of the training job.',
default='')
parser.add_argument(
'--role',
type=str.strip,
required=True,
help=
'The Amazon Resource Name (ARN) that Amazon SageMaker assumes to perform tasks on your behalf.'
)
parser.add_argument(
'--image',
type=str.strip,
required=True,
help=
'The registry path of the Docker image that contains the training algorithm.',
default='')
parser.add_argument(
'--algorithm_name',
type=str.strip,
required=False,
help='The name of the resource algorithm to use for the training job.',
default='')
parser.add_argument(
'--metric_definitions',
type=_utils.str_to_json_dict,
required=False,
help=
'The dictionary of name-regex pairs specify the metrics that the algorithm emits.',
default='{}')
parser.add_argument(
'--training_input_mode',
choices=['File', 'Pipe'],
type=str.strip,
help='The input mode that the algorithm supports. File or Pipe.',
default='File')
parser.add_argument(
'--hyperparameters',
type=_utils.str_to_json_dict,
help='Dictionary of hyperparameters for the the algorithm.',
default='{}')
parser.add_argument(
'--channels',
type=_utils.str_to_json_list,
required=True,
help=
'A list of dicts specifying the input channels. Must have at least one.'
)
parser.add_argument(
'--instance_type',
required=True,
choices=[
'ml.m4.xlarge', 'ml.m4.2xlarge', 'ml.m4.4xlarge', 'ml.m4.10xlarge',
'ml.m4.16xlarge', 'ml.m5.large', 'ml.m5.xlarge', 'ml.m5.2xlarge',
'ml.m5.4xlarge', 'ml.m5.12xlarge', 'ml.m5.24xlarge',
'ml.c4.xlarge', 'ml.c4.2xlarge', 'ml.c4.4xlarge', 'ml.c4.8xlarge',
'ml.p2.xlarge', 'ml.p2.8xlarge', 'ml.p2.16xlarge', 'ml.p3.2xlarge',
'ml.p3.8xlarge', 'ml.p3.16xlarge', 'ml.c5.xlarge', 'ml.c5.2xlarge',
'ml.c5.4xlarge', 'ml.c5.9xlarge', 'ml.c5.18xlarge'
],
type=str.strip,
help='The ML compute instance type.',
default='ml.m4.xlarge')
parser.add_argument(
'--instance_count',
required=True,
type=_utils.str_to_int,
help=
'The registry path of the Docker image that contains the training algorithm.',
default=1)
parser.add_argument(
'--volume_size',
type=_utils.str_to_int,
required=True,
help='The size of the ML storage volume that you want to provision.',
default=1)
parser.add_argument(
'--resource_encryption_key',
type=str.strip,
required=False,
help=
'The AWS KMS key that Amazon SageMaker uses to encrypt data on the storage volume attached to the ML compute instance(s).',
default='')
parser.add_argument(
'--max_run_time',
type=_utils.str_to_int,
required=True,
help='The maximum run time in seconds for the training job.',
default=86400)
parser.add_argument(
'--model_artifact_path',
type=str.strip,
required=True,
help=
'Identifies the S3 path where you want Amazon SageMaker to store the model artifacts.'
)
parser.add_argument(
'--output_encryption_key',
type=str.strip,
required=False,
help=
'The AWS KMS key that Amazon SageMaker uses to encrypt the model artifacts.',
default='')
parser.add_argument(
'--vpc_security_group_ids',
type=str.strip,
required=False,
help='The VPC security group IDs, in the form sg-xxxxxxxx.')
parser.add_argument(
'--vpc_subnets',
type=str.strip,
required=False,
help=
'The ID of the subnets in the VPC to which you want to connect your hpo job.'
)
parser.add_argument('--network_isolation',
type=_utils.str_to_bool,
required=False,
help='Isolates the training container.',
default=True)
parser.add_argument(
'--traffic_encryption',
type=_utils.str_to_bool,
required=False,
help=
'Encrypts all communications between ML compute instances in distributed training.',
default=False)
parser.add_argument(
'--tags',
type=_utils.str_to_json_dict,
required=False,
help='An array of key-value pairs, to categorize AWS resources.',
default='{}')
args = parser.parse_args()
logging.getLogger().setLevel(logging.INFO)
client = _utils.get_client(args.region)
logging.info('Submitting Training Job to SageMaker...')
job_name = _utils.create_training_job(client, vars(args))
logging.info('Job request submitted. Waiting for completion...')
_utils.wait_for_training_job(client, job_name)
image = _utils.get_image_from_job(client, job_name)
model_artifact_url = _utils.get_model_artifacts_from_job(client, job_name)
logging.info('Get model artifacts %s from training job %s.',
model_artifact_url, job_name)
with open('/tmp/model_artifact_url.txt', 'w') as f:
f.write(model_artifact_url)
with open('/tmp/job_name.txt', 'w') as f:
f.write(job_name)
with open('/tmp/training_image.txt', 'w') as f:
f.write(image)
logging.info('Job completed.')
