We use scikit-learn to train the models every few weeks when we get more labeled data. Once a model is trained we use joblib to save the entire pipeline (normal

We use scikit-learn to train the models every few weeks when we get more labeled data. Once a model is trained we use joblib to save the entire pipeline (normalization, feature processing etc). In production we have a thin Rest wrapper that loads the model pipeline to memory and serves prediction requests. We scale the number of these servers based on the load.