Configure asset deployments using JSON configuration files
Import a JSON file to create and configure all deployments of your asset for model evaluations. Export the configuration file to configure other assets and their deployments.
You can also import and export configurations for model evaluations by using the import subscription and export subscription API methods.
Load the JSON file content as a Python dictionary
For this example, the file sagemaker_native_multiclass_breast-cancer_all_monitors_sub_configuration.json
defines configuration data for a model that predicts cancer type.
To load the file in Python, run the following command:
configuration_file_path = 'sagemaker_native_multiclass_breast-cancer_all_monitors_sub_configuration.json'
with open(configuration_file_path, 'r') as fp:
subscription_configuration = json.load(fp)
The file contains configuration data. Refer to the following example. See the notebook for a complete example of the configuration content.
{'asset': {'asset_id': '0530ab0cd4f4dd5486b19c08df8b6914',
'asset_type': 'model',
'created_at': '2018-10-10T14:31:44.348Z',
'name': 'DEMO-multi-classification-2018-10-10-14-26-26',
'url': 's3://sagemaker-us-east-1-014862798213/sagemaker/DEMO-breast-cancer-prediction/DEMO-multi-classification-2018-10-10-14-26-26/output/model.tar.gz'},
'asset_properties': {'categorical_fields': [],
'feature_fields': ['radius_mean',
'texture_mean',
. . .
'input_data_schema': {'fields': [{'metadata': {'modeling_role': 'feature'},
'name': 'radius_mean',
'nullable': True,
'type': 'double'},
{'metadata': {'modeling_role': 'feature'},
'name': 'texture_mean',
'nullable': True,
'type': 'double'},
. . .
'input_data_type': 'structured',
'label_column': 'diagnosis',
'output_data_schema': {'fields': [{'metadata': {'modeling_role': 'feature'},
'name': 'radius_mean',
'nullable': True,
'type': 'double'},
{'metadata': {'modeling_role': 'feature'},
'name': 'texture_mean',
'nullable': True,
'type': 'double'},
. . .
'prediction_field': 'predicted_label',
'prediction_probability_field': 'score',
'problem_type': 'multiclass',
'training_data_schema': {'fields': [{'metadata': {'modeling_role': 'feature'},
'name': 'radius_mean',
'nullable': True,
'type': 'double'},
{'metadata': {'modeling_role': 'feature'},
'name': 'texture_mean',
'nullable': True,
'type': 'double'},
. . .
'configurations': {'explainability': {'training_statistics': {'base_values': {'0': 13.37,
'1': 18.84,
'10': 0.3242,
. . .
'fairness_monitoring': {'class_label': 'predicted_label',
'distributions': [{'attribute': 'radius_mean',
'class_labels': [{'counts': [{'class_value': 'B', 'count': 1}],
'label': '[6.8, 7.2]'},
{'counts': [{'class_value': 'B', 'count': 3}], 'label': '[7.6, 8.0]'},
{'counts': [{'class_value': 'B', 'count': 2}], 'label': '[8.0, 8.4]'},
. . .
'favourable_class': ['M'],
'features': [{'feature': 'radius_mean',
'majority': [[0.0, 10.0], [19.0, 20.0]],
'minority': [[15.0, 16.0]],
'threshold': 0.8,
'type': 'float'}],
'min_records': 5,
'perform_debias': True,
'run_status': 'INITIATED',
'training_data_class_label': None,
'unfavourable_class': ['B']},
'payload_logging': {'dynamic_schema_update': True,
'output_data_schema': {'fields': [{'metadata': {'modeling_role': 'feature'},
'name': 'radius_mean',
'nullable': True,
'type': 'double'},
{'metadata': {'modeling_role': 'feature'},
'name': 'texture_mean',
'nullable': True,
'type': 'double'},
. . .
'performance_monitoring': {},
'quality_monitoring': {'evaluation_definition': {'method': 'multiclass',
'threshold': 0.8},
'min_feedback_data_size': 5,
'scheduleId': '63c7f400-aa29-4539-91ad-8a4b9d2b9a51'}},
'deployments': [{'created_at': '2018-10-10T14:39:21.421Z',
'deployment_id': '37a83f399e6dc3b9d08d7d01fe690665',
'deployment_rn': 'arn:aws:sagemaker:us-east-1:014862798213:endpoint/demo-multi-classification-endpoint-201810101439',
'deployment_type': 'online',
'name': 'DEMO-multi-classification-endpoint-201810101439',
'scoring_endpoint': {'request_headers': {'Content-Type': 'application/json'},
'url': 'DEMO-multi-classification-endpoint-201810101439'},
'url': 'DEMO-multi-classification-endpoint-201810101439'}],
'export_info': {'api_version': 'v1',
'origin': '/v1/data_marts/b73545e6-0a6e-466c-8cd0-c47c044c5702/service_bindings/bf44cc7f-990d-4942-bfc6-cbcf71a1b78c/subscriptions/0530ab0cd4f4dd5486b19c08df8b6914',
'timestamp': '2019-02-11T11:41:01.613Z'}}
Import from the configuration file
Now, run the call to add and configure the asset deployment for the sample breast cancer prediction model deployment.
subscription = client.data_mart.subscriptions.import_configuration(binding_uid=binding_uid, configuration_data=subscription_configuration)
The binding_uid
parameter is optional if only one ML engine is bound.
Exporting to the configuration file
You can also export the configuration file as JSON:
exported_configuration = client.data_mart.subscriptions.export_configuration(binding_uid=binding_uid, subscription_uid=subscription.uid)
Results
The asset deployment is created and ready to use for model evaluations.
Parent topic: Preparing to evaluate a model