Deploying batch transformers

Some use cases don't benefit from a real-time endpoint. For example, you may want to predict 10 GB of data once a week in one go, get the results, and feed them to a downstream application. Batch transformers are a very simple way to get this done.

In this example, we will use the Scikit-Learn script that we trained on the Boston Housing dataset in Chapter 7, Extending Machine Learning Services with Built-in Frameworks.Let's get started:

1. Configure the estimator as usual:

   from sagemaker.sklearn import SKLearn
   sk = SKLearn(entry_point='sklearn-boston-housing.py',   role=sagemaker.get_execution_role(),   instance_count=1,   instance_type='ml.m5.large',   output_path=output,   hyperparameters={'normalize': True, 'test-size': 0.1})
   sk.fit({'training':training})

2. Let's predict the training set in batch mode....