Machine Learning > AI EasyMaker > SDK User Guide

SDK Settings

Install AI EasyMaker Python SDK

python -m pip install easymaker

• AI EasyMaker is installed in the notebook by default.

Initialize AI EasyMaker SDK

You can find the AppKey in the URL & Appkey menu at the right top on the console. You can learn more about Access Tokens in API Call and Authentication. Enter the AppKey, AccessToken, and region information of enabled AI EasyMaker. Intialization code is required to use the AI EasyMaker SDK.

import easymaker

easymaker.init(
    appkey='EASYMAKER_APPKEY',
    region='kr1',
    access_token='EASYMAKER_ACCESS_TOKEN',
    experiment_id="EXPERIMENT_ID", # Optional
)

Experiment

Create Experiment

Before creating a training, you must create an experiment to sort trainings.

[Parameter]

Name	Type	Required	Default value	Valid range	Description
experiment_name	String	Required	None	Up to 50 characters	Experiment name
description	String	Optional	None	Up to 255 characters	Description for experiment
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return upon creation request

experiment = easymaker.Experiment().create(
    experiment_name='experiment_name',
    description='experiment_description',
    # wait=False,
)

List Experiments

experiment_list = easymaker.Experiment.get_list()
for experiment in experiment_list:
    experiment.print_info()

Delete Experiment

[Parameter]

Name	Type	Required	Default value	Valid range	Description
experiment_id	String	Required	None	Up to 36 characters	Experiment ID

easymaker.Experiment(experiment_id).delete()

Training

List Images

image_list = easymaker.Training.get_image_list()
for image in image_list:
    image.print_info()

List Instances

instance_type_list = easymaker.Training.get_instance_type_list()
for instance in instance_type_list:
    instance.print_info()

Create Training

[Parameter]

Name	Type	Required	Default value	Valid range	Description
experiment_id	String	Required if not entered in easymaker.init	None	Up to 36 characters	Experiment ID
training_name	String	Required	None	Up to 50 characters	Training name
description	String	Optional	None	Up to 255 characters	Description for training
image_name	String	Required	None	None	Image name to be used for training (Inquiry available with CLI)
instance_type_name	String	Required	None	None	Instance type name (Inquiry available with CLI)
distributed_node_count	Integer	Optional	1	1~10	Number of nodes to apply distributed training to
use_torchrun	Boolean	Optional	False	True, False	Whether torchrun is enabled, only available for Pytorch images
nproc_per_node	Integer	Required when use_torchrun is True	1	1 to (number of CPUs or number of GPUs)	Number of processes per node, value that must be set if use_torchrun is enabled
data_storage_size	Integer	Required when using Object Storage	None	300~10000	Storage size to download data for training (unit: GB), unnecessary when using NAS
algorithm_name	String	Required when using algorithms provided by NHN Cloud	None	Up to 64 characters	Algorithm name (Inquiry available with CLI)
source_dir_uri	String	Required when using own algorithm	None	Up to 255 characters	Path of files required for training (NHN Cloud Object Storage or NHN Cloud NAS)
entry_point	String	Required when using own algorithm	None	Up to 255 characters	Information of Python files to be executed initially in source_dir_uri
model_upload_uri	String	Required	None	Up to 255 characters	Path to upload the model completed with training (NHN Cloud Object Storage or NHN Cloud NAS)
check_point_input_uri	String	Optional	None	Up to 255 characters	Input checkpoint file path (NHN Cloud Object Storage or NHN Cloud NAS)
check_point_upload_uri	String	Optional	None	Up to 255 characters	The path where the checkpoint file will be uploaded (NHN Cloud Object Storage or NHN Cloud NAS)
timeout_hours	Integer	Optional	720	1~720	Max training time (unit: hour)
hyperparameter_list	easymaker.Parameter Array	Optional	None	Max 100	Information of hyperparameters (consists of parameterName/parameterValue)
hyperparameter_list[0].parameter_name	String	Optional	None	Up to 255 characters	hyperparameter key
hyperparameter_list[0].parameter_value	String	Optional	None	Up to 1000 characters	Parameter value
dataset_list	easymaker.Dataset Array	Optional	None	Max 10	Information of dataset to be used for training (consists of dataset_name/data_uri)
dataset_list[0].dataset_name	String	Optional	None	Up to 36 characters	Data name
dataset_list[0].data_uri	String	Optional	None	Up to 255 characters	Data path
use_log	Boolean	Optional	False	True, False	Whether to leave logs in the Log & Crash Search service
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return upon creation request

training = easymaker.Training().run(
    experiment_id=experiment.experiment_id, # Optional if already set in init
    training_name='training_name',
    description='training_description',
    image_name='Ubuntu 18.04 CPU TensorFlow Training',
    instance_type_name='m2.c4m8',
    distributed_node_count=1,
    data_storage_size=300,  # minimum size : 300GB
    source_dir_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{soucre_download_path}',
    entry_point='training_start.py',
    hyperparameter_list=[
        easymaker.Parameter(
            parameter_name= "epochs",
            parameter_value= "10",
        ),
        easymaker.Parameter(
            parameter_name= "batch-size",
            parameter_value= "30",
        ),
    ],
    timeout_hours=100,
    model_upload_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{model_upload_path}',
    check_point_input_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{checkpoint_input_path}',
    check_point_upload_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{checkpoint_upload_path}',
    dataset_list=[
     easymaker.Dataset(
            dataset_name= "train",
            data_uri= "obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{train_data_path}",
        ),
        easymaker.Dataset(
            dataset_name= "test",
            data_uri= "obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{test_data_path}",
        ),
    ],
    use_log=True,
    # wait=False,
)

List of Training

training_list = easymaker.Training.get_list()
for training in training_list:
    training.print_info()

Delete Training

[Parameter]

Name	Type	Required	Default value	Valid range	Description
training_id	String	Required	None	Up to 36 characters	Training ID

easymaker.Training(training_id).delete()

Hyperparameter Tuning

List Images

image_list = easymaker.HyperparameterTuning.get_image_list()
for image in image_list:
    image.print_info()

List Instances

instance_type_list = easymaker.HyperparameterTuning.get_instance_type_list()
for instance in instance_type_list:
    instance.print_info()

Create Hyperparameter Tuning

[Parameter]

Name	Type	Required	Default value	Valid range	Description
experiment_id	String	Required if not entered in easymaker.init	None	Up to 36 characters	Experiment ID
hyperparameter_tuning_name	String	Required	None	Up to 50 characters	Hyperparameter Tuning Name
description	String	Optional	None	Up to 255 characters	Description of hyperparameter tuning
image_name	String	Required	None	None	Image name to be used for hyperparameter tuning (Inquiry available with CLI)
instance_type_name	String	Required	None	None	Instance type name (Inquiry available with CLI)
distributed_node_count	Integer	Required	1	The product of distributed_node_count and parallel_trial_count is 10 or less.	Number of nodes to apply distributed learning per each training in hyperparameter tuning
parallel_trial_count	Integer	Required	1	The product of distributed_node_count and parallel_trial_count is 10 or less.	Number of trainings to run in parallel in hyperparameter tuning
use_torchrun	Boolean	Optioanl	False	True, False	Use torchrun or not, Only available in Pytorch images
nproc_per_node	Integer	Required when use_torchrun is True	1	1~(Number of CPUs or GPUs)	Number of processes per node, Required when use_torchrun is used
data_storage_size	Integer	Required when using Object Storage	None	300~10000	Size of storage space to download data required for hyperparameter tuning (unit: GB), not required when using NAS
algorithm_name	String	Required when using algorithms provided by NHN Cloud	None	Up to 64 characters	Algorithm name (Inquiry available with CLI)
source_dir_uri	String	Required when using own algorithm	None	Up to 255 characters	Path containing files required for hyperparameter tuning (NHN Cloud Object Storage or NHN Cloud NAS)
entry_point	String	Required when using own algorithm	None	Up to 255 characters	Information of Python files to be executed initially in source_dir_uri
model_upload_uri	String	Required	None	Up to 255 characters	The path where the trained model in hyperparameter tuning will be uploaded (NHN Cloud Object Storage or NHN Cloud NAS)
check_point_input_uri	String	Optional	None	Up to 255 characters	Input checkpoint file path (NHN Cloud Object Storage or NHN Cloud NAS)
check_point_upload_uri	String	Optional	None	Up to 255 characters	The path where the checkpoint file will be uploaded (NHN Cloud Object Storage or NHN Cloud NAS)
timeout_hours	Integer	Optional	720	1~720	Maximum hyperparameter tuning time (unit: hours)
hyperparameter_spec_list	easymaker.HyperparameterSpec Array	Optional	None	Up to 100	Hyperparameter specification information
hyperparameter_spec_list[0]. hyperparameter_name	String	Optional	None	Up to 255 characters	Hyperparameter name
hyperparameter_spec_list[0]. hyperparameter_type_code	easymaker.HYPERPARAMETER_TYPE_CODE	Optional	None	INT, DOUBLE, DISCRETE, CATEGORICAL	Hyperparameter Type
hyperparameter_spec_list[0]. hyperparameter_min_value	String	Required if hyperparameterTypeCode is INT, DOUBLE(Enter a number as a string type)	None	None	Hyperparameter minimum value
hyperparameter_spec_list[0]. hyperparameter_max_value	String	Required if hyperparameterTypeCode is INT, DOUBLE(Enter a number as a string type)	None	None	Hyperparameter maximum value
hyperparameter_spec_list[0]. hyperparameter_step	String	Required if hyperparameterTypeCode is INT, DOUBLE and GRID strategy	None	None	Magnitude of change in hyperparameter values when using the "Grid" tuning strategy
hyperparameter_spec_list[0]. hyperparameter_specified_values	String	Required if hyperparameterTypeCode is DISCRETE or CATEGORICAL	None	Up to 3000 characters	A list of defined hyperparameters (strings or numbers separated by ,)
dataset_list	easymaker.Dataset Array	Optional	None	Max 10	Dataset information to be used for hyperparameter tuning (configured as dataset_name/data_uri)
dataset_list[0].dataset_name	String	Optional	None	Up to 36 characters	Data name
dataset_list[0].dataset_uri	String	Optional	None	Up to 255 characters	Data path
metric_list	easymaker.Metric	Required when using own algorithm	None	Up to 10 (string list of indicator names)	Define which metrics to collect from logs output by the training code.
metric_list[0].name	String	Required when using own algorithm	None	None	Metric name
metric_regex	String	Select when using own algorithm	([\w\	-]+)\s=\s([+-]?\d*(.\d+)?([Ee][+-]?\d+)?)	Up to 255 characters
objective_metric_name	String	Required when using own algorithm	None	Up to 36 characters, one of metric_list	Choose which metrics you want to optimize for.
objective_type_code	easymaker.OBJECTIVE_TYPE_CODE	Required when using own algorithm	None	MINIMIZE, MAXIMIZE	Choose a target metric optimization type.
objective_goal	Double	Optional	None	None	The tuning job ends when the target metric reaches this value.
max_failed_trial_count	Integer	Optional	None	None	Define the maximum number of failed lessons. When the number of failed trainings reaches this value, tuning ends in failure.
max_trial_count	Integer	Optional	None	None	Defines the maximum number of lessons. Tuning runs until the number of auto-run training reaches this value.
tuning_strategy_name	easymaker.TUNING_STRATEGY	Required	None	None	Choose which strategy to use to find the optimal hyperparameters.
tuning_strategy_random_state	Integer	Optional	None	None	Determine random number generation. Specify a fixed value for reproducible results.
early_stopping_algorithm	easymaker.EARLY_STOPPING_ALGORITHM	Required	None	EARLY_STOPPING_ALGORITHM. MEDIAN	Stop training early if the model is no longer good even though training continues.
early_stopping_min_trial_count	Integer	Optional	3	None	Define how many trainings the target metric value will be taken from when calculating the median.
early_stopping_start_step	Integer	Optional	4	None	Set the training step from which to apply early stop.
use_log	Boolean	Optional	False	True, False	Whether to leave logs in the Log & Crash Search service
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return upon creation request

hyperparameter_tuning = easymaker.HyperparameterTuning().run(
    experiment_id=experiment.experiment_id, # Optional if already set in init
    hyperparameter_tuning_name='hyperparameter_tuning_name',
    description='hyperparameter_tuning_description',
    image_name='Ubuntu 18.04 CPU TensorFlow Training',
    instance_type_name='m2.c8m16',
    distributed_node_count=1,
    parallel_trial_count=1,
    data_storage_size=300,
    source_dir_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{soucre_download_path}',
    entry_point='training_start.py',
    hyperparameter_spec_list=[
        easymaker.HyperparameterSpec(
            hyperparameter_name="learning_rate",
            hyperparameter_type_code=easymaker.HYPERPARAMETER_TYPE_CODE.DOUBLE,
            hyperparameter_min_value="0.01",
            hyperparameter_max_value="0.05",
        ),
        easymaker.HyperparameterSpec(
            hyperparameter_name="epochs",
            hyperparameter_type_code=easymaker.HYPERPARAMETER_TYPE_CODE.INT,
            hyperparameter_min_value="100",
            hyperparameter_max_value="1000",
        )
    ],
    timeout_hours=10,
    model_upload_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{model_upload_path}',
    check_point_input_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{checkpoint_input_path}',
    check_point_upload_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{checkpoint_upload_path}',
    dataset_list=[
        easymaker.Dataset(
            dataset_name="train",
            data_uri= "obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{train_data_path}"
        ),
        easymaker.Dataset(
            dataset_name="test",
            data_uri="obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{test_data_path}"
        )
    ],
    metric_list=[
        easymaker.Metric(name="loss"),
        easymaker.Metric(name="accuracy"),
        easymaker.Metric(name="val_loss"),
    ],
    metric_regex='([\w|-]+)\s*:\s*([+-]?\d*(\.\d+)?([Ee][+-]?\d+)?)',
    objective_metric_name="val_loss",
    objective_type_code=easymaker.OBJECTIVE_TYPE_CODE.MINIMIZE,
    objective_goal=0.01,
    max_failed_trial_count=3,
    max_trial_count=10,
    tuning_strategy_name=easymaker.TUNING_STRATEGY.BAYESIAN_OPTIMIZATION,
    tuning_strategy_random_state=1,
    early_stopping_algorithm=easymaker.EARLY_STOPPING_ALGORITHM.MEDIAN,
    early_stopping_min_trial_count=3,
    early_stopping_start_step=4,
    use_log=True,
    # wait=False,
)

List Hyperparameter Tuning

hyperparameter_tuning_list = easymaker.HyperparameterTuning.get_list()
for hyperparameter_tuning in hyperparameter_tuning_list:
    hyperparameter_tuning.print_info()

Delete Hyperparameter Tuning

[Parameter]

Name	Type	Required	Default value	Valid range	Description
hyperparameter_tuning_id	String	Required	None	Up to 36 characters	Hyperparameter Tuning ID

easymaker.HyperparameterTuning(hyperparameter_tuning_id).delete()

Model

Create Model

Request to create a model with the training ID. The model is used when creating endpoints.

[Parameter]

Name	Type	Required	Default value	Valid range	Description
training_id	String	Required if hyperparameter_tuning_id does not exist	None	None	Training ID to create a model
hyperparameter_tuning_id	String	Required if training_id is not present	None	None	Hyperparameter tuning ID to be created by model (created by best learning)
model_name	String	Required	None	Up to 50 characters	Model name
description	String	Optional	None	Up to 255 characters	Description for model
parameter_list	Array	Optional	None	Up to 10	Parameter information (consist of parameterName/parameterValue)
parameter_list[0].parameterName	String	Optional	None	Up to 64 characters	Parameter name
parameter_list[0].parameterValue	String	Optional	None	Up to 255 characters	Parameter value

model = easymaker.Model().create(
    training_id=training.training_id,  # or hyperparameter_tuning_id=hyperparameter_tuning.hyperparameter_tuning_id,
    model_name='model_name',
    description='model_description',
)

Even if there is no training ID, you can create a model by entering the path information for the model and framework type.

[Parameter]

Name	Type	Required	Default value	Valid range	Description
model_type_code	Enum	Required	None	easymaker.TENSORFLOW, easymaker.PYTORCH, easymaker.SCIKIT_LEARN	Framework information used for training
model_upload_uri	String	Required	None	Up to 255 characters	Path for model file (NHN Cloud Object Storage or NHN Cloud NAS)
model_name	String	Required	None	Up to 50 characters	Model name
description	String	Optional	None	Up to 255 characters	Description for model
parameter_list	Array	Optional	None	Max 10	Information of parameters (consists of parameterName/parameterValue)
parameter_list[0].parameterName	String	Optional	None	Up to 64 characters	Parameter name
parameter_list[0].parameterValue	String	Optional	None	Up to 255 characters	Parameter value

# TensorFlow Model
model = easymaker.Model().create_by_model_upload_uri(
    model_type_code=easymaker.TENSORFLOW,
    model_upload_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{model_upload_path}',
    model_name='model_name',
    description='model_description',
)
# HuggingFace Model
model = easymaker.Model().create_hugging_face_model(
    model_name='model_name',
    description='model_description',
    parameter_list=[
        {
            'parameterName': 'model_id',
            'parameterValue': 'huggingface_model_id',
        }
    ],
)

List Models

model_list = easymaker.Model.get_list()
for model in model_list:
    model.print_info()

Delete Model

[Parameter]

Name	Type	Required	Default value	Valid range	Description
model_id	String	Required	None	Up to 36 characters	Model ID

easymaker.Model(model_id).delete()

Evaluate Models

List Instances

instance_type_list = easymaker.ModelEvaluation.get_instance_type_list()
for instance in instance_type_list:
    instance.print_info()

Create a model evaluation

Create a model evaluation that measures the performance metrics of the model. Batch inference is run with the selected model and the evaluation metrics are saved.

[Parameter]

Name	Type	Required	Default value	Valid range	Description
model_evaluation_name	String	Required	None	Up to 50 characters	Model evaluation name
description	String	Optional	None	Up to 255 characters	Description for model evaluation
model_id	String	Required	None	Up to 36 characters	Model to evaluate ID
objective_code	String	Required	None	"CLASSIFICATION", "REGRESSION"	Evaluation objectives
class_names	String	Optional	None	1~5000	List of possible classes resulting from the classification model (strings or numbers separated by ,)
instance_type_name	String	Required	None	None	Instance type name (Inquiry available with CLI)
input_data_uri	String	Required	None	Up to 255 characters	Input data file path (NHN Cloud Object Storage or NHN Cloud NAS)
input_data_type_code	String	Required	None	"CSV", "JSONL"	Input data type
target_field_name	String	Required	None	Up to 255 characters	Field name of the ground truth label
timeout_hours	Integer	Optional	720	1~720	Maximum model evaluation time (in hours)
batch_inference_instance_type_name	String	Required	None	None	Instance type name (Inquiry available with CLI)
batch_inference_instance_count	Integer	Required	None	1~10	Number of instances to use for batch inference
batch_inference_pod_count	Integer	Required	None	1~100	Number of pods to apply distributed inference to
batch_inference_output_upload_uri	String	Required	None	Up to 255 characters	Path where batch inference result files will be uploaded (NHN Cloud Object Storage or NHN Cloud NAS)
batch_inference_max_batch_size	Integer	Required	None	1~1000	Number of data samples processed simultaneously
batch_inference_inference_timeout_seconds	Integer	Required	None	1~1200	Maximum allowed time for a single inference request
use_log	Boolean	Optional	False	True, False	Whether to leave logs in the Log & Crash Search service
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return upon creation request

# Create Regression Model Evaluation
regression_model_evaluation  = easymaker.ModelEvaluation().create(
    model_evaluation_name="regression_model_evaluation",
    description="regression model evaluation sample",
    model_id=regression_model.model_id,
    objective_code="REGRESSION",
    instance_type_name="m2.c4m8",
    input_data_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{input_data_path}',
    input_data_type_code="CSV",
    target_field_name="target_field_name",
    timeout_hours=1,
    batch_inference_instance_type_name="m2.c4m8",
    batch_inference_instance_count=1,
    batch_inference_pod_count=1,
    batch_inference_output_upload_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{input_data_path}',
    batch_inference_max_batch_size=100,
    batch_inference_inference_timeout_seconds=1200,
    use_log=False,
    wait=True,
)
# Create Classification Model Evaluation
classification_model_evaluation  = easymaker.ModelEvaluation().create(
    model_evaluation_name="classification_model_evaluation",
    description="classification model evaluation sample",
    model_id=classification_model.model_id,
    objective_code="CLASSIFICATION",
    class_names="classA,classB,classC",
    instance_type_name="m2.c4m8",
    input_data_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{input_data_path}',
    input_data_type_code="CSV",
    target_field_name="target_field_name",
    timeout_hours=1,
    batch_inference_instance_type_name="m2.c4m8",
    batch_inference_instance_count=1,
    batch_inference_pod_count=1,
    batch_inference_output_upload_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{input_data_path}',
    batch_inference_max_batch_size=100,
    batch_inference_inference_timeout_seconds=1200,
    use_log=False,
    wait=True,
)

List Model Evaluations

model_evaluation_list = easymaker.ModelEvaluation.get_list()
for model_evaluation in model_evaluation_list:
    model_evaluation.print_info()

Delete a model evaluation

[Parameter]

Name	Type	Required	Default value	Valid range	Description
model_evaluation_id	String	Required	None	Up to 36 characters	Model evaluation ID

easymaker.ModelEvaluation(model_evaluation_id).delete()

Endpoint

List Instances

instance_type_list = easymaker.Endpoint.get_instance_type_list()
for instance in instance_type_list:
    instance.print_info()

Create Endpoint

When creating an endpoint, the default stage is created.

[Parameter]

Name	Type	Required	Default value	Valid range	Description
endpoint_name	String	Required	None	Up to 50 characters	Endpoint name
description	String	Optional	None	Up to 255 characters	Description for endpoint
instance_type_name	String	Required	None	None	Instance type name to be used for endpoint
instance_count	Integer	Optional	1	1~10	Instance count to be used for endpoint
endpoint_model_resource_list	easymaker.EndpointModelResource Array	Required	None	Up to 10	Resource information to be used on the stage
endpoint_model_resource_list[0].model_id	String	Required	None	None	Model ID to be created as a stage resource
endpoint_model_resource_list[0].resource_option_detail	easymaker.ResourceOptionDetail	Required	None		Details of stage resource
endpoint_model_resource_list[0].resource_option_detail.cpu	String	Required	None	0.0~	CPU to be used for stage resource
endpoint_model_resource_list[0].resource_option_detail.memory	String	Required	None	1Mi~	Memory to be used for stage resource
endpoint_model_resource_list[0].pod_auto_scale_enable	Boolean	Optional	False	True, False	Pod autoscaler to be used for stage resource
endpoint_model_resource_list[0].scale_metric_code	easymaker.SCALE_METRIC_CODE	Optional	None	CPU_UTILIZATION, MEMORY_UTILIZATION	Scaling unit to be used for stage resource
endpoint_model_resource_list[0].scale_metric_target	Integer	Optional	None	1~	Scaling threshold to be used for stage resource
endpoint_model_resource_list[0].description	String	Optional	None	최대 255자	Description of stage resource
use_log	Boolean	Optional	False	True, False	Whether to leave logs in the Log & Crash Search service
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return upon creation request

endpoint = easymaker.Endpoint().create(
    endpoint_name='endpoint_name',
    description='endpoint_description',
    instance_type_name='c2.c16m16',
    instance_count=1,
    endpoint_model_resource_list=[
        easymaker.EndpointModelResource(
            model_id=model.model_id,
            resource_option_detail=easymaker.ResourceOptionDetail(
                cpu="15",
                memory="15Gi",
            ),
            pod_auto_scale_enable=True,
            scale_metric_code=easymaker.SCALE_METRIC_CODE.CPU_UTILIZATION,
            scale_metric_target=50,
        )
    ],
    use_log=True,
    # wait=False,
)

Use the created endpoint

endpoint = easymaker.Endpoint(endpoint_id)

Add Stage

You can add a new stage to existing endpoints.

[Parameter]

Name	Type	Required	Default value	Valid range	Description
endpoint_id	String	Required	None	Up to 36 characters	Endpoint ID
stage_name	String	Required	None	Up to 50 characters	Stage name
description	String	Optional	None	Up to 255 characters	Description for stage
instance_type_name	String	Required	None	None	Instance type name to be used for endpoint
instance_count	Integer	Optional	1	1~10	Instance count to be used for endpoint
endpoint_model_resource_list	easymaker.EndpointModelResource Array	Required	None	up to 10	Resource information to be used on the stage
endpoint_model_resource_list[0].model_id	String	Required	None	None	Model ID to be created as a stage resource
endpoint_model_resource_list[0].resource_option_detail	easymaker.ResourceOptionDetail	Required	None		Details of stage resource
endpoint_model_resource_list[0].resource_option_detail.cpu	String	Required	None	0.0~	CPU to be used for stage resource
endpoint_model_resource_list[0].resource_option_detail.memory	String	Required	None	1Mi~	Memory to be used for stage resource
endpoint_model_resource_list[0].pod_auto_scale_enable	Boolean	Optional	False	True, False	Pod autoscaler to be used for stage resource
endpoint_model_resource_list[0].scale_metric_code	easymaker.SCALE_METRIC_CODE	Optional	None	CPU_UTILIZATION, MEMORY_UTILIZATION	Scaling unit to be used for stage resource
endpoint_model_resource_list[0].scale_metric_target	Integer	Optional	None	1~	Scaling threshold to be used for stage resource
endpoint_model_resource_list[0].description	String	Optional	None	Up to 255 characters	Description of stage resource
use_log	Boolean	Optional	False	True, False	Whether to leave logs in the Log & Crash Search service
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return upon creation request

endpoint_stage = easymaker.EndpointStage().create(
    endpoint_id=endpoint.endpoint_id,
    stage_name='stage01',  # lowercase/number within 30 characters
    description='test endpoint',
    instance_type_name='c2.c16m16',
    instance_count=1,
    endpoint_model_resource_list=[
        easymaker.EndpointModelResource(
            model_id=model.model_id,
            resource_option_detail=easymaker.ResourceOptionDetail(
                cpu="15",
                memory="15Gi",
            ),
            pod_auto_scale_enable=True,
            scale_metric_code=easymaker.SCALE_METRIC_CODE.CPU_UTILIZATION,
            scale_metric_target=50,
            description='stage_resource_description'
        )
    ],
    use_log=True,
    # wait=False,
)

Retrieve Stages

Retrieves endpoint stages.

endpoint_stage_list = easymaker.Endpoint(endpoint_id).get_stage_list()

Endpoint Inference

Inference to the default stage

input_data = [6.0, 3.4, 4.5, 1.6]
easymaker.Endpoint('endpoint_id').predict(
    model_id=model_id,
    json={'instances': [input_data]},
)

Inference by specifying a specific stage

input_data = [6.0, 3.4, 4.5, 1.6]
easymaker.EndpointStage('endpoint_stage_id').predict(
    model_id=model_id,
    json={'instances': [input_data]},
)

List Endpoints

endpoint_list = easymaker.Endpoint.get_list()
for endpoint in endpoint_list:
    endpoint.print_info()

Delete Endpoint

[Parameter]

Name	Type	Required	Default value	Valid range	Description
endpoint_id	String	Required	None	Up to 36 characters	Endpoint ID

easymaker.Endpoint(endpoint_id).delete()

Delete Endpoint Stage

[Parameter]

Name	Type	Required	Default value	Valid range	Description
stage_id	String	Required	None	Up to 36 characters	Stage ID

easymaker.EndpointStage(stage_id).delete()

Batch Inference

List Instances

instance_type_list = easymaker.BatchInference.get_instance_type_list()
for instance in instance_type_list:
    instance.print_info()

Create Batch Inference

[Parameter]

Name	Type	Required	Default value	Valid range	Description
batch_inference_name	String	Required	None	Up to 50 characters	Batch inference name
instance_count	Integer	Optional	1	1~10	Number of instances to use for batch inference
timeout_hours	Integer	Required	720	1~720	Maximum batch inference time (in hours)
instance_type_name	String	Required	None	None	Instance type name (Inquiry available with CLI)
model_id	String	Required	None	None	Model ID
pod_count	Integer	Optional	1	1~100	Number of pods to apply distributed inference to
batch_size	Integer	Required	None	1~1000	Number of data samples processed simultaneously
inference_timeout_seconds	Integer	Required	None	1~1200	Maximum allowable time for a single inference request
input_data_uri	String	Required	None	Up to 255	Path for input data file (NHN Cloud Object Storage or NHN Cloud NAS)
input_data_type	String	Required	None	"JSON", "JSONL"	Input data type
include_glob_pattern	String	Optional	None	Up to 255	Glob pattern to include a set of files in the input data
exclude_glob_pattern	String	Optional	None	Up to 255	Glob pattern to exclude a set of files in the input data
output_upload_uri	String	Required	None	Up to 255	The path where the batch inference result file will be uploaded (NHN Cloud Object Storage or NHN Cloud NAS)
data_storage_size	Integer	Required	None	300~10000	Storage size to download data for batch inference (unit: GB)
description	String	Optional	None	Up to 255	Explanation of batch inference
use_log	Boolean	Optional	False	True, False	Whether to leave logs with the Log & Crash Search service
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return upon creation request

batch_inference = easymaker.BatchInference().run(
    batch_inference_name='batch_inference_name',
    instance_count=1,
    timeout_hours=100,
    instance_type_name='m2.c4m8',
    model_id=model.model_id,
    pod_count=1,
    batch_size=32,
    inference_timeout_seconds=120,
    input_data_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{input_data_path}',
    input_data_type='JSONL',
    include_glob_pattern=None,
    exclude_glob_pattern=None,
    output_upload_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{output_upload_path}',
    data_storage_size=300, # minimum size : 300GB
    description='description',
    use_log=True,
    # wait=False,
)

List of Batch Inference

batch_inference_list = easymaker.BatchInference.get_list()
for batch_inference in batch_inference_list:
    batch_inference.print_info()

Delete Batch Inference

[Parameter]

Name	Type	Required	Default value	Valid range	Description
batch_inference_id	String	Required	None	Up to 36 characters	Batch Inference ID

easymaker.BatchInference(batch_inference_id).delete()

Pipeline

Create Pipeline

[Parameter]

Name	Type	Required	Default value	Valid range	Description
pipeline_name	String	Required	None	Max 50 characters	Pipeline name
pipeline_spec_manifest_path	String	Required	None	none	Pipeline file path to upload
description	String	Optional	None	Max 255 characters	Description for pipeline
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return immediately after creation request

pipeline = easymaker.Pipeline().upload(
    pipeline_name='pipeline_01',
    pipeline_spec_manifest_path='./sample-pipeline.yaml',
    description='test',
    # wait=False,
)

List of Pipeline

pipeline_list = easymaker.Pipeline.get_list()
for pipeline in pipeline_list:
    pipeline.print_info()

Delete Pipeline

[Parameter]

Name	Type	Required	Default value	Valid range	Description
pipeline_id	String	Required	None	Up to 36 characters	Pipeline ID

easymaker.Pipeline(pipeline_id).delete()

List Instances

instance_type_list = easymaker.PipelineRun.get_instance_type_list()
for instance in instance_type_list:
    instance.print_info()

Create Pipeline Run

[Parameter]

Name	Type	Required	Default value	Valid range	Description
pipeline_run_name	String	Required	None	Up to 50 Characters	Pipeline run name
pipeline_id	String	Required	None	Up to 36 Characters	Pipeline schedule name
experiment_id	String	Required if not entered in easymaker.init	None	Up to 36 Characters	Experiment ID
description	String	Optional	None	Up to 255 Characters	Description of pipeline execution
instance_type_name	String	Required	None	None	Instance type name (Inquiry available with CLI)
instance_count	Integer	Optional	1	1~10	Number of instances to use
boot_storage_size	Integer	Required	None	50~	The boot storage size (in GB) of the instance that will run the pipeline.
parameter_list	easymaker.Parameter Array	Optional	None	None	Parameter information to pass to the pipeline
parameter_list[0].parameter_name	String	Optional	None	Up to 255 Characters	Parameter key
parameter_list[0].parameter_value	String	Optional	None	Up to 1000 Characters	Parameter value
nas_list	easymaker.Nas Array	Optional	None	Up to 10	NAS information
nas_list[0].mount_dir_name	String	Optional	None	Up to 64 Characters	Directory name to be mounted on instances
nas_list[0].nas_uri	String	Optional	None	Up to 255 Characters	The path to the NAS in the format nas://{NAS ID}:/{path}
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return immediately after creation request

pipeline_run = easymaker.PipelineRun().create(
    pipeline_run_name='pipeline_run',
    description='test',
    pipeline_id=pipeline.pipeline_id,
    experiment_id=experiment.experiment_id, # Optional if already set in init
    instance_type_name='m2.c4m8',
    instance_count=1,
    boot_storage_size=50,
    parameter_list=[
        easymaker.Parameter(parameter_name="experiment_name", parameter_value="pipeline_experiment"),
    ],
    nas_list=[
        easymaker.Nas(mount_dir_name="user_nas", nas_uri="nas://{NAS ID}:/{path}"),
    ],
    # wait=False,
)

List of Pipeline Run

pipeline_run_list = easymaker.PipelineRun.get_list()
for pipeline_run in pipeline_run_list:
    pipeline_run.print_info()

Delete Pipeline Run

[Parameter]

Name	Type	Required	Default value	Valid range	Description
pipeline_run_id	String	Required	None	Up to 36 characters	Pipeline run ID

easymaker.PipelineRun(pipeline_run_id).delete()

Create Pipeline Schedule

[Parameter]

Name	Type	Required	Default value	Valid range	Description
pipeline_recurring_run_name	String	Required	None	Max 50 characters	Pipeline schedule name
pipeline_id	String	Required	None	Max 36 character	Pipeline schedule name
experiment_id	String	Required if not entered in easymaker.init	None	Max 36 character	Experiment ID
description	String	Optional	None	Max 255 character	Description of pipeline schedules
instance_type_name	String	Required	None	None	Instance type name (Inquiry available with CLI)
instance_count	Integer	Optional	1	1~10	Number of instances to use
boot_storage_size	Integer	Required	None	50~	The boot storage size (in GB) of the instance that will run the pipeline.
schedule_periodic_minutes	String	schedule_cron_expression 미입력시 Required	None	None	Set a time interval to run the pipeline repeatedly
schedule_cron_expression	String	schedule_periodic_minutes 미입력시 Required	None	None	Set up a Cron expression to run the pipeline repeatedly
max_concurrency_count	Integer	Optional	1	1~10	Limit the number of concurrent runs by specifying a maximum number of parallel runs
schedule_start_datetime	String	Optional	None	None	Set a start time for the pipeline schedule, which will run the pipeline at the set interval if not entered.
schedule_end_datetime	String	Optional	None	None	Set an end time for a pipeline schedule, creating a pipeline run until it stops if no input is received.
use_catchup	Boolean	Optional	None	None	Missed run catch-up: Whether to catch up when pipeline runs fall behind schedule.
parameter_list	easymaker.Parameter Array	Optional	None	None	Parameter information to pass to the pipeline
parameter_list[0].parameter_name	String	Optional	None	Up to 255 character	Parameter key
parameter_list[0].parameter_value	String	Optional	None	Up to 1000 character	Parameter value
nas_list	easymaker.Nas Array	Optional	None	Up to 10	NAS information
nas_list[0].mount_dir_name	String	Optional	None	Up to 64 character	Directory name to be mounted on instances
nas_list[0].nas_uri	String	Optional	None	Up to 255 character	The path to the NAS in the format nas://{NAS ID}:/{path}
wait	Boolean	Optional	True	True, False	True: return after creation is complete, False: return immediately after creation request

pipeline_recurring_run = easymaker.PipelineRecurringRun().create(
    pipeline_recurring_run_name='pipeline_recurring_run',
    description='test',
    pipeline_id=pipeline.pipeline_id,
    experiment_id=experiment.experiment_id, # Optional if already set in init
    instance_type_name='m2.c4m8',
    parameter_list=[
        easymaker.Parameter(parameter_name="experiment_name", parameter_value="pipeline_experiment"),
    ],
    nas_list=[
        easymaker.Nas(mount_dir_name="user_nas", nas_uri="nas://{NAS ID}:/{path}"),
    ],
    boot_storage_size=50,
    schedule_cron_expression='0 0 * * * ?',
    max_concurrency_count=1,
    schedule_start_datetime='2025-01-01T00:00:00+09:00'
    # wait=False,
)

Stop/Restart Pipeline Schedule

[Parameter]

Name	Type	Required	Default value	Valid range	Description
pipeline_recurring_run_id	String	Required	None	Max 36 characters	Pipeline Schedule ID

easymaker.PipelineRecurringRun(pipeline_recurring_run_id).stop()
easymaker.PipelineRecurringRun(pipeline_recurring_run_id).start()

List of Pipeline Schedule

pipeline_recurring_run_list = easymaker.PipelineRecurringRun.get_list()
for pipeline_recurring_run in pipeline_recurring_run_list:
    pipeline_recurring_run.print_info()

Delete Pipeline Schedule

[Parameter]

Name	Type	Required	Default value	Valid range	Description
pipeline_recurring_run_id	String	Required	None	Max 36 characters	Pipeline Schedule ID

easymaker.PipelineRecurringRun(pipeline_recurring_run_id).delete()

Other Features

NHN Cloud - Log & Crash Search Log Sending Feature

easymaker_logger = easymaker.logger(logncrash_appkey='log&crash_product_app_key')
easymaker_logger.send('test log meassage')  # Output to stdout & send log to log&crash product
easymaker_logger.send(log_message='log meassage',
                      log_level='ERROR',  # default: INFO
                      project_version='2.0.0',  # default: 1.0.0
                      parameters={'serviceType': 'EasyMakerSample'})  # Add custom parameters

NHN Cloud - Object Storage File Sending Feature

Provide a feature to upload and download files with Object Storage.

easymaker.upload(
    easymaker_obs_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_{tenant_id}/{container_name}/{upload_path}',
    local_path='./local_dir',
    username='userId@nhn.com',
    password='nhn_object_storage_api_password'
)

easymaker.download(
    easymaker_obs_uri='obs://kr1-api-object-storage.nhncloudservice.com/v1/AUTH_00000000000000000000000000000000/SDK/sample/source_dir',
    download_dir_path='./download_dir',
    username='userId@nhn.com',
    password='nhn_object_storage_api_password'
)