dougansonaustintx / pelion-sagemaker-controller Goto Github PK

# Required import
from pelion_sagemaker_controller import pelion_sagemaker_controller

#
# Invoke constructor with Pelion API Key, Pelion GW Device ID
# You can also optionally specify the Pelion API endpoint you want to use
#
api = pelion_sagemaker_controller.ControllerAPI(
		api_key='<ak_xxxx>', 
		device_id='<pelion_gw_device_id>', 
		api_endpoint='api.us-east-1.mbedcloud.com'
		)

api.pelion_get_config()

This call returns a JSON with the current Edge Device representing the 
Sagemaker service's configuration

api.pelion_set_config({'foo':'bar'})

This call updates or adds key/values to the current Edge Device's configuration

api.pelion_list_models()

This call returns a JSON outlining all of the loaded models

api.pelion_load_model('model-name','compiled-model-x.y.tar.gz')

This call loads up the requested Sagemaker-compiled model whose compiled 
contents are located within the S3 bucket defined in the configuration
and utilized by the Sagemaker service

api.pelion_unload_model('model-name')

This call unloads the loaded model referenced by the name 'model-name'

api.pelion_reload_model('model-name','compiled-model-x.y.tar.gz')

This call is a convenience method for simply performing an "unload" followed by
a "load" of a given model using the methods above. 

api.pelion_predict(
          'model-name',
          's3:///input.data', 
          's3:///prediction_result.data'
          )

This call invokes the model prediction using the specified input.data file that is
configured to be pulled from the Sagemaker S3 bucket (per configuration). The output
result from the prediction will be stored in a file that will be saved to the same
directory in the S3 bucket. 

In addition to S3 bucket support, you can locally reference input/output requirements
using the "file:///" protocol - in this case the Sagemaker Edge Agent working directory
on the Pelion Edge Gateway will contain the specified files. 

api.pelion_last_cmd_result()

This call returns the last invocation/call results. In cases where predictions take
a long time to complete, this call may be used in a polling situation to determine
when the prediction operation has completed.