In this third part, you learn to design and build production pipelines for model training, deployment, and serving. We start by introducing you to how hyperparameter tuning works under the surface and then show you both a do-it-yourself (DIY) method and automatic hyperparameter tuning using KerasTuner. In both cases, effective hyperparameter tuning requires good judgement in choosing the search space, so we discuss these best practices.
Next, we turn to transfer learning. In transfer learning, you reuse the weights from another trained model and fine-tune a new model with less data and less training time. We cover several variations of transfer learning, one for when the domain of the new dataset is very similar to the trained model (for example, vegetables versus fruits), and another when the domain is very different. Finally, we cover domain transfer techniques for initializing models when doing full training.