Monitoring models using advanced tools

So far, you have built and evaluated the CNN TFBS prediction model (Chapter 9, Building and Tuning Deep Learning Models), interpreted it (Chapter 10, Model Interpretability in Genomics), and deployed it successfully (this chapter). You have even ensured that the model is working smoothly and correctly in a production environment. So, you might be thinking you are done, right? Not even close – but this is the beginning of a new journey. Just imagine what could go wrong after model deployment. Models can start to degrade post-deployment and consistently not perform the way they are expected to. Even though you have done everything right from model building to model deployment, things can go wrong after the model goes live in the production environment. Even after you have troubleshooted and tested a model thoroughly, things can still go wrong after model deployment.

Why monitor models?

Theoretically, once a model has been deployed,...