如何使用 AWS SageMaker Autopilot 创建的模型生成批量预测？

Question

I created a complete model using Amazon Web Services (AWS) SageMaker Autopilot. I would like to see what forecasts the model makes on my training data. I'm running this in a SageMaker Studio notebook. Here's my code.

import sagemaker

image = sagemaker.image_uris.retrieve("xgboost", sagemaker.session.Session().boto_region_name, version="latest")

model = sagemaker.model.Model(
    image_uri = image,
    model_data = "s3://sagemaker-us-east-.../batch-prediction/sagemaker-xgboost-2021-.../output/model.tar.gz"
)

transformer = model.transformer(
    instance_count = 1,
    instance_type = "ml.c4.xlarge"
)

Here's the full error stack.

---------------------------------------------------------------------------
TypeError                                 Traceback (most recent call last)
<ipython-input-14-23386d1fc99a> in <module>
      1 transformer = my_model.transformer(
      2     instance_count = 1,
----> 3     instance_type = "ml.c4.xlarge"
      4 )

/opt/conda/lib/python3.7/site-packages/sagemaker/model.py in transformer(self, instance_count, instance_type, strategy, assemble_with, output_path, output_kms_key, accept, env, max_concurrent_transforms, max_payload, tags, volume_kms_key)
    772         self._init_sagemaker_session_if_does_not_exist(instance_type)
    773 
--> 774         self._create_sagemaker_model(instance_type, tags=tags)
    775         if self.enable_network_isolation():
    776             env = None

/opt/conda/lib/python3.7/site-packages/sagemaker/model.py in _create_sagemaker_model(self, instance_type, accelerator_type, tags)
    259             vpc_config=self.vpc_config,
    260             enable_network_isolation=enable_network_isolation,
--> 261             tags=tags,
    262         )
    263 

/opt/conda/lib/python3.7/site-packages/sagemaker/session.py in create_model(self, name, role, container_defs, vpc_config, enable_network_isolation, primary_container, tags)
   2596             enable_network_isolation=enable_network_isolation,
   2597             primary_container=primary_container,
-> 2598             tags=tags,
   2599         )
   2600         LOGGER.info("Creating model with name: %s", name)

/opt/conda/lib/python3.7/site-packages/sagemaker/session.py in _create_model_request(self, name, role, container_defs, vpc_config, enable_network_isolation, primary_container, tags)
   2512             container_defs = primary_container
   2513 
-> 2514         role = self.expand_role(role)
   2515 
   2516         if isinstance(container_defs, list):

/opt/conda/lib/python3.7/site-packages/sagemaker/session.py in expand_role(self, role)
   3466             str: The corresponding AWS IAM role ARN.
   3467         """
-> 3468         if "/" in role:
   3469             return role
   3470         return self.boto_session.resource("iam").Role(role).arn

TypeError: argument of type 'NoneType' is not iterable

Answer 1

You also need to specify a SageMaker role (either name or full ARN) to give permissions to Amazon SageMaker training jobs to access training data and model artifacts. Here is the link to the documentation .

So after your import sagemaker you need to add something like

from sagemaker import get_execution_role

sagemaker_session = sagemaker.Session() 
role = get_execution_role()

and later have

model = sagemaker.model.Model(
    image_uri = image,
    model_data = "s3://sagemaker-us-east-.../batch-prediction/sagemaker-xgboost-2021-.../output/model.tar.gz",
role=role
...