AWS SageMaker 入門: 機械学習モデルの構築とデプロイ

import sagemaker
from sagemaker.tensorflow import TensorFlow

sagemaker_session = sagemaker.Session()
bucket = sagemaker_session.default_bucket()
prefix = 'sagemaker/my-training'

estimator = TensorFlow(entry_point='train.py',
                       role='arn:aws:iam::1234567890:role/MySageMakerRole',
                       framework_version='2.1.0',
                       train_instance_count=1,
                       train_instance_type='ml.c5.xlarge',
                       output_path='s3://{}/{}/output'.format(bucket, prefix),
                       base_job_name='my-training-job')

inputs = 's3://{}/{}/data'.format(bucket, prefix)

estimator.fit(inputs)

import boto3
import sagemaker
import os
from sagemaker.pytorch import PyTorch

sagemaker_session = sagemaker.Session()

# S3 bucket name for storing training data
bucket = sagemaker_session.default_bucket()
prefix = 'sagemaker/pytorch-mnist-example'

# S3 location to store training data
input_data = 's3://{}/{}/'.format(bucket, prefix)

# S3 location for model artifacts
output_path = 's3://{}/{}/output'.format(bucket, prefix)

# IAM role for executing the training job
role = 'arn:aws:iam::1234567890:role/MySageMakerRole'

# Get the PyTorch estimator
estimator = PyTorch(entry_point='train.py',
                    source_dir='source',
                    role=role,
                    framework_version='1.8.0',
                    train_instance_count=1,
                    train_instance_type='ml.c5.2xlarge',
                    output_path=output_path,
                    sagemaker_session=sagemaker_session,
                    hyperparameters={
                        'epochs': 10,
                        'batch-size': 128,
                        'learning-rate': 0.01,
                        'backend': 'gloo',
                        'rank': 0,
                        'world-size': 1
                    })

# Launch the training job
estimator.fit(inputs=input_data)

Resources:
  SageMakerExecutionRole:
    Type: AWS::IAM::Role
    Properties:
      AssumeRolePolicyDocument:
        Version: '2012-10-17'
        Statement:
          - Effect: Allow
            Principal:
              Service: sagemaker.amazonaws.com
            Action: sts:AssumeRole
      Path: /
      Policies:
        - PolicyName: SageMakerExecutionPolicy
          PolicyDocument:
            Version: '2012-10-17'
            Statement:
              - Effect: Allow
                Action:
                  - s3:ListBucket
                  - s3:GetObject
                  - s3:PutObject
                  - sagemaker:CreateTrainingJob
                  - sagemaker:DescribeTrainingJob
                  - sagemaker:CreateModel
                  - sagemaker:CreateEndpointConfig
                  - sagemaker:CreateEndpoint
                  - sagemaker:DeleteEndpoint
                  - sagemaker:InvokeEndpoint
                  - sagemaker:ListEndpointConfigs
                  - sagemaker:ListEndpoints
                Resource: '*'
  MySageMakerInstance:
    Type: AWS::SageMaker::NotebookInstance
    Properties:
      InstanceType: ml.t2.medium
      RoleArn: !GetAtt SageMakerExecutionRole.Arn
      NotebookInstanceName: my-notebook-instance