Model versioning

A model that goes to production is the eventual result of a series of experimentation and model modifications with different versions of training and test data, and different machine learning methods and their corresponding hyperparameters. Model versioning helps us ensure that changes that are made to models are traceable, helping to establish reproducibility in our machine learning projects. It ensures that every version of a model can be easily reproduced by providing a complete snapshot of the model’s parameters, hyperparameters, and training data at a given point in time. It allows us to easily roll back to a previous version in case of issues with a newly deployed model or to recover an older version that may have been unintentionally modified or deleted.

Let’s go through a very simple example to better understand the need for model versioning. Figure 10.1 shows the performance of a random forest model with five estimators, or decision trees...