Monitoring and Updating Deployed Models

Once your model is live, its performance will degrade over time due to changes in input data distributions (drift) or external conditions. This is basically a given in any scenario where an ML model is used and is not unexpected. This is where the CT in CI/CD/CT comes into play. You may be familiar with the term, Continuous Integration/Continuous Deployment or Deployment (CI/CD) which is a software engineering paradigm for building and releasing software. However, in an ML deployment, we must also consider Continuous Training for monitoring and managing models as they degrade. This recipe guides you through strategies to monitor deployed models, detect drift, and update (retrain or adjust) models to maintain reliability.

Getting ready

You’ll simulate incoming batches of data and require incremental-capable models to experiment with.

1. Load libraries:

   import numpy as np
   import matplotlib.pyplot as plt
   from sklearn.linear_model import SGDClassifier...