Building Machine Learning Pipelines

Chapter 15. The Future of Pipelines and Next Steps

In the past 14 chapters, we have captured the current state of machine learning pipelines and given our recommendations on how to build them. Machine learning pipelines are a relatively new concept, and there’s much more to come in this space. In this chapter, we will discuss a few things that we feel are important but don’t fit well with current pipelines, and we also consider future steps for ML pipelines.

Model Experiment Tracking

Throughout this book, we have assumed that you’ve already experimented and the model architecture is basically settled. However, we would like to share some thoughts on how to track experiments and make experimentation a smooth process. Your experimental process may include exploring potential model architectures, hyperparameters, and feature sets. But whatever you explore, the key point we would like to make is that your experimental process should fit closely with your production process.

Whether you optimize your models manually or you tune the models automatically, capturing and sharing the results of the optimization process is essential. Team members can quickly evaluate the progress of the model updates. At the same time, the author of the models can receive automated records of the performed experiments. Good experiment tracking helps data science teams become more efficient.

Experiment tracking also adds to the audit trail of the model and may be a safeguard against potential litigations. ...