14. Feature Engineering for Domains: Domain-Specific Learning
# setup from mlwpy import * %matplotlib inline import cv2
In a perfect world, our standard classification algorithms would be provided with data that is relevant, plentiful, tabular (formatted in a table), and naturally discriminitive when we look at it as points in space.
In reality, we may be stuck with data that is
Only an approximation of our target task
Limited in quantity or covers only some of many possibilities
Misaligned with the prediction method(s) we are trying to apply to it
Written as text or images which are decidedly not in an example-feature table
Issues 1 and 2 are specific to a learning problem you are focused on. We discussed issue 3 in Chapters 10 and ...