Making real-time predictions

With batch predictions, you submit all the samples you want the model to predict at once to Amazon ML by creating a datasource. With real-time predictions, also called streaming or online predictions, the idea is to send one sample at a time to an API endpoint, a URL, via HTTP queries, and receive back predictions and information for each one of the samples.

Setting up real-time predictions on a model consists of knowing the prediction API endpoint URL and writing a script that can read your data, send each new sample to that API URL, and retrieve the predicted class or value. We will present a Python-based example in the following section.

Amazon ML also offers a way to make predictions on data you create on the fly on the prediction page. We can input the profile of a would-be passenger on the Titanic and see whether that profile would have survived or not. It is a great way to explore the influence of the dataset variables on the outcome.

Before setting up API...