Building Machine Learning Pipelines

Let's go a step further and say we don't even want to select the columns in advance. Instead, we need the pipeline to do the selection for us. Now, we introduce FeatureUnion- It concatenates the results of multiple transformations happening in parallel! Since pipeline wants classes as selectors that have fit, transform methods, we will extend the base classes BaseEstimator & TransformerMixin. We could have applied grid-search, random search, and cross-validations in all of them!