def test_hyperparameter_optimization_happy_case():
with patch(
'boto3.Session'
):
with patch(
'sagemaker.Session'
) as mocked_sagemaker_session:
sagemaker_session_instance = mocked_sagemaker_session.return_value
with patch(
'sagemaker.get_execution_role',
return_value='arn_role'
):
with patch(
'sagemaker.estimator.Estimator'
) as mocked_sagemaker_estimator:
with patch(
'sagify.sagemaker.sagemaker.SageMakerClient._construct_image_location',
return_value='image-full-name'
):
with patch(
'sagemaker.tuner.HyperparameterTuner'
) as mocked_sagemaker_tuner:
sage_maker_client = sagemaker.SageMakerClient('sagemaker', 'us-east-1')
sage_maker_client.hyperparameter_optimization(
image_name='image',
input_s3_data_location='s3://bucket/input',
instance_count=1,
instance_type='m1.xlarge',
volume_size=30,
max_run=60,
max_jobs=3,
max_parallel_jobs=2,
output_path='s3://bucket/output',
objective_type='Maximize',
objective_metric_name='Precision',
hyperparams_ranges_dict={
'lr': ContinuousParameter(0.001, 0.1),
'batch-size': CategoricalParameter([32, 64, 128, 256, 512])
},
base_job_name="Some-job-name-prefix",
job_name="some job name"
)
mocked_sagemaker_estimator.assert_called_with(
image_name='image-full-name',
role='arn_role',
train_instance_count=1,
train_instance_type='m1.xlarge',
train_volume_size=30,
train_max_run=60,
input_mode='File',
output_path='s3://bucket/output',
sagemaker_session=sagemaker_session_instance
)
mocked_sagemaker_tuner_instance = mocked_sagemaker_tuner.return_value
assert mocked_sagemaker_tuner_instance.fit.call_count == 1
mocked_sagemaker_tuner_instance.fit.assert_called_with(
's3://bucket/input', job_name='some job name'
)
def test_deploy_xgboost_happy_case():
with patch('boto3.Session'):
with patch('sagemaker.Session') as mocked_sagemaker_session:
sagemaker_session_instance = mocked_sagemaker_session.return_value
with patch('sagemaker.get_execution_role',
return_value='arn_role'):
with patch('sagemaker.xgboost.model.XGBoostModel'
) as mocked_sagemaker_xgboost_model:
sage_maker_client = sagemaker.SageMakerClient(
'sagemaker', 'us-east-1')
sage_maker_client.deploy_xgboost(
s3_model_location=
's3://bucket/model_input/model.tar.gz',
instance_count=1,
instance_type='m1.xlarge',
framework_version='0.90-2')
mocked_sagemaker_xgboost_model.assert_called_with(
role='arn_role',
model_data='s3://bucket/model_input/model.tar.gz',
framework_version='0.90-2',
py_version='py3',
entry_point='xgboost_inference.py',
source_dir=os.path.join(sagemaker._FILE_DIR_PATH,
'xgboost_code'),
model_server_workers=None,
sagemaker_session=sagemaker_session_instance)
sagemaker_model_instance = mocked_sagemaker_xgboost_model.return_value
assert sagemaker_model_instance.deploy.call_count == 1
sagemaker_model_instance.deploy.assert_called_with(
initial_instance_count=1,
instance_type='m1.xlarge',
tags=None,
endpoint_name=None)
def train(dir, input_s3_dir, output_s3_dir, hyperparams_file, ec2_type,
volume_size, time_out):
"""
Command to train ML model(s) on SageMaker
"""
logger.info(ASCII_LOGO)
logger.info("Started training on SageMaker...\n")
config = _read_config(dir)
hyperparams_dict = _read_hyperparams_config(
hyperparams_file) if hyperparams_file else None
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile,
config.aws_region)
s3_model_location = sage_maker_client.train(
image_name=config.image_name,
input_s3_data_location=input_s3_dir,
train_instance_count=1,
train_instance_type=ec2_type,
train_volume_size=volume_size,
train_max_run=time_out,
output_path=output_s3_dir,
hyperparameters=hyperparams_dict)
logger.info("Training on SageMaker succeeded")
logger.info("Model S3 location: {}".format(s3_model_location))
def test_deploy_happy_case():
with patch('boto3.Session'):
with patch('sagemaker.Session') as mocked_sagemaker_session:
sagemaker_session_instance = mocked_sagemaker_session.return_value
with patch('sagemaker.get_execution_role',
return_value='arn_role'):
with patch('sagemaker.Model') as mocked_sagemaker_model:
with patch(
'sagify.sagemaker.sagemaker.SageMakerClient._construct_image_location',
return_value='image-full-name'):
sage_maker_client = sagemaker.SageMakerClient(
'sagemaker', 'us-east-1')
sage_maker_client.deploy(
image_name='image',
s3_model_location=
's3://bucket/model_input/model.tar.gz',
train_instance_count=1,
train_instance_type='m1.xlarge')
mocked_sagemaker_model.assert_called_with(
model_data='s3://bucket/model_input/model.tar.gz',
image='image-full-name',
role='arn_role',
sagemaker_session=sagemaker_session_instance)
sagemaker_model_instance = mocked_sagemaker_model.return_value
assert sagemaker_model_instance.deploy.call_count == 1
sagemaker_model_instance.deploy.assert_called_with(
initial_instance_count=1,
instance_type='m1.xlarge',
tags=None,
endpoint_name=None,
update_endpoint=True)
def test_batch_transform_with_tags():
with patch('boto3.Session'):
with patch('sagemaker.Session') as mocked_sagemaker_session:
sagemaker_session_instance = mocked_sagemaker_session.return_value
with patch('sagemaker.get_execution_role',
return_value='arn_role'):
with patch('sagemaker.Model') as mocked_sagemaker_model:
with patch(
'sagify.sagemaker.sagemaker.SageMakerClient._construct_image_location',
return_value='image-full-name'):
sage_maker_client = sagemaker.SageMakerClient(
'sagemaker', 'us-east-1')
tags = [
{
'Key': 'key_name_1',
'Value': 1,
},
{
'Key': 'key_name_2',
'Value': '2',
},
]
sage_maker_client.batch_transform(
image_name='image',
s3_model_location=
's3://bucket/model_input/model.tar.gz',
s3_input_location='s3://bucket/input_data',
s3_output_location='s3://bucket/output_data',
transform_instance_count=1,
transform_instance_type='m1.xlarge',
tags=tags)
mocked_sagemaker_model.assert_called_with(
model_data='s3://bucket/model_input/model.tar.gz',
image='image-full-name',
role='arn_role',
sagemaker_session=sagemaker_session_instance)
sagemaker_model_instance = mocked_sagemaker_model.return_value
assert sagemaker_model_instance.transformer.call_count == 1
sagemaker_model_instance.transformer.assert_called_with(
instance_type='m1.xlarge',
instance_count=1,
assemble_with='Line',
output_path='s3://bucket/output_data',
tags=tags,
accept='application/json',
strategy="SingleRecord")
transformer = sagemaker_model_instance.transformer.return_value
assert transformer.transform.call_count == 1
transformer.transform.assert_called_with(
data='s3://bucket/input_data',
split_type='Line',
content_type='application/json',
job_name=None)
def train(dir,
job_name,
input_s3_dir,
output_s3_dir,
hyperparams_file,
ec2_type,
volume_size,
time_out,
docker_tag,
tags=None):
"""
Trains ML model(s) on SageMaker
:param dir: [str], source root directory
:param job_name [str], training job name
:param input_s3_dir: [str], S3 location to input data
:param output_s3_dir: [str], S3 location to save output (models, etc)
:param hyperparams_file: [str], path to hyperparams json file
:param ec2_type: [str], ec2 instance type. Refere to:
https://aws.amazon.com/sagemaker/pricing/instance-types/
:param volume_size: [int], size in GB of the EBS volume
:param time_out: [int], time-out in seconds
:param tags: [optional[list[dict]], default: None], List of tags for labeling a training
job. For more, see https://docs.aws.amazon.com/sagemaker/latest/dg/API_Tag.html. Example:
[
{
'Key': 'key_name_1',
'Value': key_value_1,
},
{
'Key': 'key_name_2',
'Value': key_value_2,
},
...
]
:return: [str], S3 model location
"""
config = _read_config(dir)
hyperparams_dict = _read_hyperparams_config(
hyperparams_file) if hyperparams_file else None
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile,
config.aws_region)
image_name = config.image_name + ':' + docker_tag
return sage_maker_client.train(image_name=image_name,
job_name=job_name,
input_s3_data_location=input_s3_dir,
train_instance_count=1,
train_instance_type=ec2_type,
train_volume_size=volume_size,
train_max_run=time_out,
output_path=output_s3_dir,
hyperparameters=hyperparams_dict,
tags=tags)
def test_deploy_with_tags():
with patch(
'boto3.Session'
):
with patch(
'sagemaker.Session'
) as mocked_sagemaker_session:
sagemaker_session_instance = mocked_sagemaker_session.return_value
with patch(
'sagemaker.get_execution_role',
return_value='arn_role'
):
with patch(
'sagemaker.Model'
) as mocked_sagemaker_model:
with patch(
'sagify.sagemaker.sagemaker.SageMakerClient._construct_image_location',
return_value='image-full-name'
):
sage_maker_client = sagemaker.SageMakerClient('sagemaker', 'us-east-1')
tags = [
{
'Key': 'key_name_1',
'Value': 1,
},
{
'Key': 'key_name_2',
'Value': '2',
},
]
sage_maker_client.deploy(
image_name='image',
s3_model_location='s3://bucket/model_input/model.tar.gz',
train_instance_count=1,
train_instance_type='m1.xlarge',
tags=tags
)
mocked_sagemaker_model.assert_called_with(
model_data='s3://bucket/model_input/model.tar.gz',
image='image-full-name',
role='arn_role',
sagemaker_session=sagemaker_session_instance
)
sagemaker_model_instance = mocked_sagemaker_model.return_value
assert sagemaker_model_instance.deploy.call_count == 1
sagemaker_model_instance.deploy.assert_called_with(
initial_instance_count=1,
instance_type='m1.xlarge',
tags=tags
)
def deploy(
dir,
s3_model_location,
num_instances,
ec2_type,
docker_tag,
aws_role=None,
external_id=None,
tags=None,
endpoint_name=None
):
"""
Deploys ML model(s) on SageMaker
:param dir: [str], source root directory
:param s3_model_location: [str], S3 model location
:param num_instances: [int], number of ec2 instances
:param ec2_type: [str], ec2 instance type. Refer to:
https://aws.amazon.com/sagemaker/pricing/instance-types/
:param docker_tag: [str], the Docker tag for the image
:param aws_role: [str], the AWS role assumed by SageMaker while deploying
:param external_id: [str], Optional external id used when using an IAM role
:param tags: [optional[list[dict]], default: None], List of tags for labeling a training
job. For more, see https://docs.aws.amazon.com/sagemaker/latest/dg/API_Tag.html. Example:
[
{
'Key': 'key_name_1',
'Value': key_value_1,
},
{
'Key': 'key_name_2',
'Value': key_value_2,
},
...
]
:param endpoint_name: [optional[str]], Optional name for the SageMaker endpoint
:return: [str], endpoint name
"""
config = _read_config(dir)
image_name = config.image_name+':'+docker_tag
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile, config.aws_region, aws_role, external_id)
return sage_maker_client.deploy(
image_name=image_name,
s3_model_location=s3_model_location,
train_instance_count=num_instances,
train_instance_type=ec2_type,
tags=tags,
endpoint_name=endpoint_name
)
def batch_transform(dir,
s3_model_location,
s3_input_location,
s3_output_location,
num_instances,
ec2_type,
docker_tag,
aws_role=None,
external_id=None,
tags=None):
"""
Executes a batch transform job given a trained ML model on SageMaker
:param dir: [str], source root directory
:param s3_model_location: [str], S3 model location
:param s3_input_location: [str], S3 input data location
:param s3_output_location: [str], S3 location to save predictions
:param num_instances: [int], number of ec2 instances
:param ec2_type: [str], ec2 instance type. Refer to:
https://aws.amazon.com/sagemaker/pricing/instance-types/
:param docker_tag: [str], the Docker tag for the image
:param aws_role: [str], the AWS role assumed by SageMaker while deploying
:param external_id: [str], Optional external id used when using an IAM role
:param tags: [optional[list[dict]], default: None], List of tags for labeling a training
job. For more, see https://docs.aws.amazon.com/sagemaker/latest/dg/API_Tag.html. Example:
[
{
'Key': 'key_name_1',
'Value': key_value_1,
},
{
'Key': 'key_name_2',
'Value': key_value_2,
},
...
]
"""
config = _read_config(dir)
image_name = config.image_name + ':' + docker_tag
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile,
config.aws_region, aws_role,
external_id)
sage_maker_client.batch_transform(image_name=image_name,
s3_model_location=s3_model_location,
s3_input_location=s3_input_location,
s3_output_location=s3_output_location,
transform_instance_count=num_instances,
transform_instance_type=ec2_type,
tags=tags)
def upload_data(dir, input_dir, s3_dir):
"""
Uploads data to S3
:param dir: [str], source root directory
:param input_dir: [str], path to local data input directory
:param s3_dir: [str], S3 location to upload data
:return: [str], S3 location to upload data
"""
config = _read_config(dir)
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile, config.aws_region)
return sage_maker_client.upload_data(input_dir, s3_dir)
def test_train_happy_case():
with patch(
'boto3.Session'
):
with patch(
'sagemaker.Session'
) as mocked_sagemaker_session:
sagemaker_session_instance = mocked_sagemaker_session.return_value
with patch(
'sagemaker.get_execution_role',
return_value='arn_role'
):
with patch(
'sagemaker.estimator.Estimator'
) as mocked_sagemaker_estimator:
with patch(
'sagify.sagemaker.sagemaker.SageMakerClient._construct_image_location',
return_value='image-full-name'
):
sage_maker_client = sagemaker.SageMakerClient('sagemaker', 'us-east-1')
sage_maker_client.train(
image_name='image',
input_s3_data_location='s3://bucket/input',
train_instance_count=1,
train_instance_type='m1.xlarge',
train_volume_size=30,
train_max_run=60,
output_path='s3://bucket/output',
hyperparameters={'n_estimator': 3},
base_job_name="Some-job-name-prefix",
job_name="some job name"
)
mocked_sagemaker_estimator.assert_called_with(
image_name='image-full-name',
role='arn_role',
train_instance_count=1,
train_instance_type='m1.xlarge',
train_volume_size=30,
train_max_run=60,
input_mode='File',
base_job_name="Some-job-name-prefix",
output_path='s3://bucket/output',
hyperparameters={'n_estimator': 3},
sagemaker_session=sagemaker_session_instance,
metric_definitions=None
)
sagemaker_estimator_instance = mocked_sagemaker_estimator.return_value
assert sagemaker_estimator_instance.fit.call_count == 1
sagemaker_estimator_instance.fit.assert_called_with('s3://bucket/input', job_name='some job name')
def upload_data(dir, input_dir, s3_dir):
"""
Command to upload data to S3
"""
logger.info(ASCII_LOGO)
logger.info("Started uploading data to S3...\n")
config = _read_config(dir)
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile,
config.aws_region)
s3_path = sage_maker_client.upload_data(input_dir, s3_dir)
logger.info("Data uploaded to {} successfully".format(s3_path))
def test_upload_data_with_s3_path_that_contains_only_bucket_name():
with patch('boto3.Session'):
with patch('sagemaker.Session') as mocked_sagemaker_session:
sagemaker_session_instance = mocked_sagemaker_session.return_value
with patch('sagemaker.get_execution_role',
return_value='arn_role'):
sage_maker_client = sagemaker.SageMakerClient(
'sagemaker', 'us-east-1')
sage_maker_client.upload_data(input_dir='/input/data',
s3_dir='s3://bucket/')
assert sagemaker_session_instance.upload_data.call_count == 1
sagemaker_session_instance.upload_data.assert_called_with(
path='/input/data', bucket='bucket', key_prefix='data')
def deploy(dir,
s3_model_location,
model_name,
vpc_configs,
num_instances,
ec2_type,
docker_tag,
tags=None):
"""
Deploys ML model(s) on SageMaker
:param dir: [str], source root directory
:param s3_model_location: [str], S3 model location
:param model_name: [str], SageMaker model name
:param num_instances: [int], number of ec2 instances
:param ec2_type: [str], ec2 instance type. Refere to:
https://aws.amazon.com/sagemaker/pricing/instance-types/
:param tags: [optional[list[dict]], default: None], List of tags for labeling a training
job. For more, see https://docs.aws.amazon.com/sagemaker/latest/dg/API_Tag.html. Example:
[
{
'Key': 'key_name_1',
'Value': key_value_1,
},
{
'Key': 'key_name_2',
'Value': key_value_2,
},
...
]
:return: [str], endpoint name
"""
config = _read_config(dir)
image_name = config.image_name
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile,
config.aws_region,
vpc_configs)
return sage_maker_client.deploy(image_name=image_name,
s3_model_location=s3_model_location,
model_name=model_name,
train_instance_count=num_instances,
train_instance_type=ec2_type,
tags=tags)
def deploy(dir, s3_model_location, num_instances, ec2_type):
"""
Command to deploy ML model(s) on SageMaker
"""
logger.info(ASCII_LOGO)
logger.info("Started deployment on SageMaker ...\n")
config = _read_config(dir)
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile,
config.aws_region)
endpoint_name = sage_maker_client.deploy(
image_name=config.image_name,
s3_model_location=s3_model_location,
train_instance_count=num_instances,
train_instance_type=ec2_type)
logger.info("Model deployed to SageMaker successfully")
logger.info("Endpoint name: {}".format(endpoint_name))
def test_deploy_hugging_face_with_hub():
with patch('boto3.Session'):
with patch('sagemaker.Session') as mocked_sagemaker_session:
sagemaker_session_instance = mocked_sagemaker_session.return_value
with patch('sagemaker.get_execution_role',
return_value='arn_role'):
with patch('sagemaker.huggingface.HuggingFaceModel'
) as mocked_sagemaker_hf_model:
sage_maker_client = sagemaker.SageMakerClient(
'sagemaker', 'us-east-1')
hub = {'HF_MODEL_ID': 'gpt2', 'HF_TASK': 'text-generation'}
sage_maker_client.deploy_hugging_face(
instance_count=1,
instance_type='m1.xlarge',
transformers_version='4.6.1',
pytorch_version='1.7.1',
hub=hub)
mocked_sagemaker_hf_model.assert_called_with(
role='arn_role',
model_data=None,
transformers_version='4.6.1',
pytorch_version='1.7.1',
tensorflow_version=None,
model_server_workers=None,
py_version='py36',
env=hub,
sagemaker_session=sagemaker_session_instance)
sagemaker_model_instance = mocked_sagemaker_hf_model.return_value
assert sagemaker_model_instance.deploy.call_count == 1
sagemaker_model_instance.deploy.assert_called_with(
initial_instance_count=1,
instance_type='m1.xlarge',
tags=None,
endpoint_name=None)
def test_deploy_hugging_face_happy_case():
with patch('boto3.Session'):
with patch('sagemaker.Session') as mocked_sagemaker_session:
sagemaker_session_instance = mocked_sagemaker_session.return_value
with patch('sagemaker.get_execution_role',
return_value='arn_role'):
with patch('sagemaker.huggingface.HuggingFaceModel'
) as mocked_sagemaker_hf_model:
sage_maker_client = sagemaker.SageMakerClient(
'sagemaker', 'us-east-1')
sage_maker_client.deploy_hugging_face(
s3_model_location=
's3://bucket/model_input/model.tar.gz',
instance_count=1,
instance_type='m1.xlarge',
transformers_version='4.6.1',
pytorch_version='1.7.1')
mocked_sagemaker_hf_model.assert_called_with(
role='arn_role',
model_data='s3://bucket/model_input/model.tar.gz',
transformers_version='4.6.1',
pytorch_version='1.7.1',
tensorflow_version=None,
model_server_workers=None,
py_version='py36',
env=None,
sagemaker_session=sagemaker_session_instance)
sagemaker_model_instance = mocked_sagemaker_hf_model.return_value
assert sagemaker_model_instance.deploy.call_count == 1
sagemaker_model_instance.deploy.assert_called_with(
initial_instance_count=1,
instance_type='m1.xlarge',
tags=None,
endpoint_name=None)
def lightning_deploy(framework,
num_instances,
ec2_type,
aws_region,
s3_model_location=None,
model_server_workers=None,
aws_profile=None,
aws_role=None,
external_id=None,
tags=None,
endpoint_name=None,
extra_config_file=None):
"""
Deploys ML model(s) on SageMaker without code
:param framework: [str], The name of the ML framework.
Valid values: sklearn, huggingface, xgboost
:param num_instances: [int], number of ec2 instances
:param ec2_type: [str], ec2 instance type. Refer to:
https://aws.amazon.com/sagemaker/pricing/instance-types/
:param aws_region: [str], the AWS region
:param s3_model_location: [str], S3 model location
:param model_server_workers: [int], Optional number of worker processes used by the inference
server. If None, server will use one worker per vCPU.
:param aws_profile: [optional[str]], Optional AWS profile
:param aws_role: [optional[str]], Optional AWS role assumed by SageMaker while deploying
:param external_id: [optional[str]], Optional external id used when using an IAM role
:param tags: [optional[list[dict]], default: None], List of tags for labeling a training
job. For more, see https://docs.aws.amazon.com/sagemaker/latest/dg/API_Tag.html. Example:
[
{
'Key': 'key_name_1',
'Value': key_value_1,
},
{
'Key': 'key_name_2',
'Value': key_value_2,
},
...
]
:param endpoint_name: [optional[str]], Optional name for the SageMaker endpoint
:param extra_config_file: [optional[str]], Optional Json file with ML framework specific arguments
:return: [str], endpoint name
"""
sage_maker_client = sagemaker.SageMakerClient(aws_profile, aws_region,
aws_role, external_id)
if not os.path.isfile(extra_config_file):
raise ValueError("The given extra config file {} doesn't exist".format(
extra_config_file))
with open(extra_config_file) as _in_file:
extra_config_dict = json.load(_in_file)
if framework == 'sklearn':
return sage_maker_client.deploy_sklearn(
s3_model_location=s3_model_location,
instance_count=num_instances,
instance_type=ec2_type,
model_server_workers=model_server_workers,
tags=tags,
endpoint_name=endpoint_name,
**extra_config_dict)
elif framework == 'huggingface':
return sage_maker_client.deploy_hugging_face(
s3_model_location=s3_model_location,
instance_count=num_instances,
instance_type=ec2_type,
model_server_workers=model_server_workers,
tags=tags,
endpoint_name=endpoint_name,
**extra_config_dict)
elif framework == 'xgboost':
return sage_maker_client.deploy_xgboost(
s3_model_location=s3_model_location,
instance_count=num_instances,
instance_type=ec2_type,
model_server_workers=model_server_workers,
tags=tags,
endpoint_name=endpoint_name,
**extra_config_dict)
raise ValueError("Invalid framework value")
def hyperparameter_optimization(
dir,
input_s3_dir,
output_s3_dir,
hyperparams_config_file,
ec2_type,
max_jobs,
max_parallel_jobs,
volume_size,
time_out,
docker_tag,
aws_role,
external_id,
base_job_name,
job_name,
wait,
tags=None
):
"""
Hyperparameter Optimization on SageMaker
:param dir: [str], source root directory
:param input_s3_dir: [str], S3 location to input data
:param output_s3_dir: [str], S3 location to save the multiple trained models
:param hyperparams_config_file: [str], path to hyperparameters config json file
:param ec2_type: [str], ec2 instance type. Refer to:
https://aws.amazon.com/sagemaker/pricing/instance-types/
:param max_jobs: [int], Maximum total number of training jobs to start for the hyperparameter tuning job
:param max_parallel_jobs: [int], Maximum number of parallel training jobs to start
:param volume_size: [int], size in GB of the EBS volume
:param time_out: [int], time-out in seconds
:param docker_tag: [str], the Docker tag for the image
:param aws_role: [str], the AWS role assumed by SageMaker while training
:param external_id: [str], Optional external id used when using an IAM role
:param base_job_name: [str], Optional prefix for the SageMaker training job
:param job_name: [str], Optional name for the SageMaker tuning job. Overrides `base_job_name`
:param wait: [bool, default=False], Wait until hyperparameter tuning is done
:param tags: [optional[list[dict]], default: None], List of tags for labeling a training
job. For more, see https://docs.aws.amazon.com/sagemaker/latest/dg/API_Tag.html. Example:
[
{
'Key': 'key_name_1',
'Value': key_value_1,
},
{
'Key': 'key_name_2',
'Value': key_value_2,
},
...
]
:return: [str], S3 model location
"""
config = _read_config(dir)
objective_metric_name, objective_type, hyperparams_ranges_dict = _read_hyperparams_ranges_config(
hyperparams_config_file
)
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile, config.aws_region, aws_role, external_id)
image_name = config.image_name+':'+docker_tag
return sage_maker_client.hyperparameter_optimization(
image_name=image_name,
input_s3_data_location=input_s3_dir,
instance_count=1,
instance_type=ec2_type,
volume_size=volume_size,
objective_type=objective_type,
objective_metric_name=objective_metric_name,
max_jobs=max_jobs,
max_parallel_jobs=max_parallel_jobs,
max_run=time_out,
output_path=output_s3_dir,
hyperparams_ranges_dict=hyperparams_ranges_dict,
base_job_name=base_job_name,
job_name=job_name,
tags=tags,
wait=wait
)
def train(
dir,
input_s3_dir,
output_s3_dir,
hyperparams_file,
ec2_type,
volume_size,
time_out,
docker_tag,
aws_role,
external_id,
base_job_name,
job_name,
metric_names=None,
tags=None
):
"""
Trains ML model(s) on SageMaker
:param dir: [str], source root directory
:param input_s3_dir: [str], S3 location to input data
:param output_s3_dir: [str], S3 location to save output (models, etc)
:param hyperparams_file: [str], path to hyperparams json file
:param ec2_type: [str], ec2 instance type. Refer to:
https://aws.amazon.com/sagemaker/pricing/instance-types/
:param volume_size: [int], size in GB of the EBS volume
:param time_out: [int], time-out in seconds
:param docker_tag: [str], the Docker tag for the image
:param aws_role: [str], the AWS role assumed by SageMaker while training
:param external_id: [str], Optional external id used when using an IAM role
:param base_job_name: [str], Optional prefix for the SageMaker training job
:param job_name: [str], Optional name for the SageMaker training job. Overrides `base_job_name`
:param metric_names: [list[str], default=None], Optional list of string metric names
:param tags: [optional[list[dict]], default: None], List of tags for labeling a training
job. For more, see https://docs.aws.amazon.com/sagemaker/latest/dg/API_Tag.html. Example:
[
{
'Key': 'key_name_1',
'Value': key_value_1,
},
{
'Key': 'key_name_2',
'Value': key_value_2,
},
...
]
:return: [str], S3 model location
"""
config = _read_config(dir)
hyperparams_dict = _read_hyperparams_config(hyperparams_file) if hyperparams_file else None
sage_maker_client = sagemaker.SageMakerClient(config.aws_profile, config.aws_region, aws_role, external_id)
image_name = config.image_name+':'+docker_tag
return sage_maker_client.train(
image_name=image_name,
input_s3_data_location=input_s3_dir,
train_instance_count=1,
train_instance_type=ec2_type,
train_volume_size=volume_size,
train_max_run=time_out,
output_path=output_s3_dir,
hyperparameters=hyperparams_dict,
base_job_name=base_job_name,
job_name=job_name,
tags=tags,
metric_names=metric_names
)
