How can I specify content_type in a training job of XGBoost from Sagemaker in Python?

问题

I am trying to train a model using the sagemaker library. So far, my code is the following:

container = get_image_uri(boto3.Session().region_name,
                      'xgboost', 
                      repo_version='0.90-1')

estimator = sagemaker.estimator.Estimator(container, 
                                          role = 'AmazonSageMaker-ExecutionRole-20190305TXXX',
                                          train_instance_count = 1,
                                          train_instance_type = 'ml.m4.2xlarge',
                                          output_path = 's3://antifraud/production/',
                                          hyperparameters = {'num_rounds':'400',
                                                             'objective':'binary:logistic',
                                                             'eval_metric':'error@0.1'})

train_config = training_config(estimator=estimator,
                               inputs = {'train':'s3://antifraud/production/train',
                                         'validation':'s3://-antifraud/production/validation'})

And I get an error parsing the hyperparameters. This commands gives me a configuration JSON output in the console. I have been able to run a training job using boto3 with the configuration as Json, so I have figured out that the thing I am missing in my json configuration generated by my code is the content_type parameter, which should be there as follow:

"InputDataConfig": [
    {
        "ChannelName": "train",
        "DataSource": {
            "S3DataSource": {
                "S3DataType": "S3Prefix",
                "S3Uri": "s3://antifraud/production/data/train",
                "S3DataDistributionType": "FullyReplicated" 
            }
        },
        "ContentType": "text/csv",
        "CompressionType": "None"
    },
    {
        "ChannelName": "validation",
        "DataSource": {
            "S3DataSource": {
                "S3DataType": "S3Prefix",
                "S3Uri": "s3://antifraud/production/validation",
                "S3DataDistributionType": "FullyReplicated"
            }
        },
        "ContentType": "text/csv",
        "CompressionType": "None"
    }
]

I have tried coding content_type = 'text/csv' in container, estimator and train_config as parameter and also inside inputs as another key of the dictionary, with no success. How could I make this work?

回答1:

I have solved it using s3_input objects:

s3_input_train = sagemaker.s3_input(s3_data='s3://antifraud/production/data/{domain}-{product}-{today}/train_data.csv',
content_type='text/csv')
s3_input_validation = sagemaker.s3_input(s3_data='s3://antifraud/production/data/{domain}-{product}-{today}/validation_data.csv',
content_type='text/csv')

train_config = training_config(estimator=estimator,
inputs = {'train':s3_input_train,
          'validation':s3_input_validation})

来源：https://stackoverflow.com/questions/57908395/how-can-i-specify-content-type-in-a-training-job-of-xgboost-from-sagemaker-in-py

标签

python

amazon-web-services

amazon-sagemaker