sagemaker_helpers.py

# get default bucket
bucket_name = sagemaker.Session().default_bucket()

# upload data to s3


# training data for sagemaker
s3_input_train = sagemaker.inputs.TrainingInput(s3_data='s3://{}/{}/data/train'.format(bucket_name, prefix), content_type='csv')
s3_input_validation = sagemaker.inputs.TrainingInput(s3_data='s3://{}/{}/data/val'.format(bucket_name, prefix), content_type='csv')

# generic estimator

xgb = sagemaker.estimator.Estimator(container,
                                    role, 
                                    instance_count=1, 
                                    instance_type='ml.m4.xlarge',
                                    output_path=f's3://{bucket}/{prefix}/output',
                                    sagemaker_session=sess,
                                    use_spot_instances=True,        # Use spot instance
                                    max_run=600,                    # Max training time
                                    max_wait=3600)                  # Max training time + spot waiting time

initial commit