The future of ML is automated
Training an ML model is a complex iterative process that includes data preparation, feature engineering, model selection, optimization, and deployment. Above all, an enterprise-grade end-to-end ML pipeline needs to be reproducible, interpretable, secure, and automated, which poses an additional challenge for most companies in terms of know-how, costs, and infrastructure requirements.
In previous chapters, we learned the ins and outs of this process, and hence we can confirm that there is nothing simple or easy about it. Tuning a feature engineering approach will affect model training; the missing value strategy during data cleansing will influence the optimization process.
On top of all this, the information captured by your model is rarely constant and therefore most ML models require frequent retraining and deployments. This leads to a whole new requirement for MLOps: a DevOps pipeline for ML to ensure continuous integration and continuous deployment...
