In this chapter, we have learned how to reuse data preprocessing logic for training and inference and how to run online as opposed to offline inferences. We started by understanding the architecture of the machine learning inference pipeline. Then, we used the ABC News Headlines dataset to illustrate big data processing through AWS Glue and SparkML. Then, we discovered topics from the news headlines by fitting the NTM algorithm to processed headlines. Finally, we walked through real-time as opposed to batch inferences by utilizing the same data preprocessing logic for inference. Through the inference pipeline, data scientists and machine learning engineers can increase speed with which ML solutions are marketed.
In the next chapter, we'll do a deep dive into Neural Topic Models (NTMs).
