When we talk about pipelines in machine learning, we are usually talking about the fact that data is only passed through a learning algorithm raw, but also through a variety of preprocessing steps and even multiple learning algorithms before the final output is interpreted. Because it is so common to have several steps and transformation and prediction within a single machine learning pipeline, scikit-learn has a built-in module for building these pipelines.
Pipelines are especially important because it is actually improper to not use a pipeline when imputing values using the Imputer class. This is because the goal of the learning algorithm is to generalize the patterns in the training set in order to apply those ...
