11 Machine learning in Pulsar

 

This chapter covers

  • Exploring how Pulsar Functions can be used to provide near real-time machine learning
  • Developing and maintaining the collection of inputs required by the machine learning model to provide a prediction
  • Executing any PMML-supported model inside a Pulsar function
  • Executing non-PMML models inside a Pulsar function

One of the primary goals of machine learning is the extraction of actionable insights from raw data. Having insights that are actionable means you can use them to make strategic, data-driven decisions that result in a positive outcome for your business and customers. For instance, every time a customer places an order on the GottaEat application, we want to be able to provide the customer with an accurate estimated delivery time. To accomplish this, we would need to develop a machine learning model that predicts the delivery time of any given order based on a number of factors that allow us to make a more accurate decision.

11.1 Deploying ML models

11.1.1 Batch processing

11.1.2 Near real-time

11.2 Near real-time model deployment

11.3 Feature vectors

11.3.1 Feature stores

11.3.2 Feature calculation

11.4 Delivery time estimation

11.4.1 ML model export

11.4.2 Feature vector mapping

11.4.3 Model deployment

11.5 Neural nets

11.5.1 Neural net training

11.5.2 Neural net deployment in Java

Summary