The key takeaways from this section include:

• Machine learning pipelines create a nice workflow to combine data manipulations, preprocessing, and modeling
• Machine learning pipelines can be used along with grid search to evaluate several parameter settings
  • Grid search can considerably blow up computation time when computing for several parameters along with cross-validation
  • Some models are very sensitive to hyperparameter changes, so they should be chosen with care, and even with big grids a good outcome isn't always guaranteed
• Machine learning pipelines can also be pickled so that they can be used in the future without re-training
• Model deployment can be something as simple as pickling a model, or a more complex approach like a cloud function that exposes model predictions through an HTTP API