Amazon SageMaker
Developer Guide

Step 2.3.2: Create a Training Job

To train a model, Amazon SageMaker provides the CreateTrainingJob API. You provide the following information when making this API call:

  • The training algorithm—Specify the registry path of the Docker image that contains the training code. For the registry paths for the algorithms provided by Amazon SageMaker, see Common Parameters for Built-In Algorithms . In the following examples, when using the high-level Python library, you don't need to explicitly specify this path. The object knows the path.

  • Algorithm-specific hyperparameters—Specify algorithm-specific hyperparameters to influence the final quality of the model. For information, see K-Means Hyperparameters.

  • The input and output configuration—Provide the S3 bucket where training data is stored and where Amazon SageMaker saves the results of model training (the model artifacts).

The low-level AWS SDK for Python provides the corresponding create_training_job method and the high-level Python library provide the fit method.

To train the model, choose one of the following options.

  • Use the high-level Python library provided by Amazon SageMaker.

    This Python library provides the KMeans estimator, which is a class in the module. To start model training, call the fit method.

    1. Create an instance of the class.

      from sagemaker import KMeans data_location = 's3://{}/kmeans_highlevel_example/data'.format(bucket) output_location = 's3://{}/kmeans_highlevel_example/output'.format(bucket) print('training data will be uploaded to: {}'.format(data_location)) print('training artifacts will be uploaded to: {}'.format(output_location)) kmeans = KMeans(role=role, train_instance_count=2, train_instance_type='ml.c4.8xlarge', output_path=output_location, k=10, data_location=data_location)

      In the constructor, you specify the following parameters:

      • role— The IAM role that Amazon SageMaker can assume to perform tasks on your behalf (for example, reading training results, called model artifacts, from the S3 bucket and writing training results to Amazon S3).

      • output_path—The Amazon S3 location where Amazon SageMaker stores the training results.

      • train_instance_count and train_instance_type—The type and number of ML compute instances to use for model training.

      • k—The number of clusters to create. For more information, see K-Means Hyperparameters.

      • data_location—The Amazon S3 location where the high-level library uploads the transformed training data.

    2. To start model training, call the KMeans estimator's fit method.


      This is a synchronous operation. The method displays progress logs and waits until training completes before returning. For more information about model training, see Train a Model with Amazon SageMaker .

      The model training in this example takes about 15 minutes.

  • Use the SDK for Python.

    The low-level SDK for Python provides the create_training_job method, which maps to the CreateTrainingJob Amazon SageMaker API.

    %%time import boto3 from time import gmtime, strftime job_name = 'kmeans-lowlevel-' + strftime("%Y-%m-%d-%H-%M-%S", gmtime()) print("Training job", job_name) from import get_image_uri image = get_image_uri(boto3.Session().region_name, 'kmeans') output_location = 's3://{}/kmeans_lowlevel_example/output'.format(bucket) print('training artifacts will be uploaded to: {}'.format(output_location)) create_training_params = \ { "AlgorithmSpecification": { "TrainingImage": image, "TrainingInputMode": "File" }, "RoleArn": role, "OutputDataConfig": { "S3OutputPath": output_location }, "ResourceConfig": { "InstanceCount": 2, "InstanceType": "ml.c4.8xlarge", "VolumeSizeInGB": 50 }, "TrainingJobName": job_name, "HyperParameters": { "k": "10", "feature_dim": "784", "mini_batch_size": "500" }, "StoppingCondition": { "MaxRuntimeInSeconds": 60 * 60 }, "InputDataConfig": [ { "ChannelName": "train", "DataSource": { "S3DataSource": { "S3DataType": "S3Prefix", "S3Uri": data_location, "S3DataDistributionType": "FullyReplicated" } }, "CompressionType": "None", "RecordWrapperType": "None" } ] } sagemaker = boto3.client('sagemaker') sagemaker.create_training_job(**create_training_params) status = sagemaker.describe_training_job(TrainingJobName=job_name)['TrainingJobStatus'] print(status) try: sagemaker.get_waiter('training_job_completed_or_stopped').wait(TrainingJobName=job_name) finally: status = sagemaker.describe_training_job(TrainingJobName=job_name)['TrainingJobStatus'] print("Training job ended with status: " + status) if status == 'Failed': message = sagemaker.describe_training_job(TrainingJobName=job_name)['FailureReason'] print('Training failed with the following error: {}'.format(message)) raise Exception('Training job failed')

    The code uses a Waiter to wait until training is complete before returning.

You now have trained a model. The resulting artifacts are stored in your Amazon S3 bucket.

Next Step

Step 2.4: Deploy the Model to Amazon SageMaker