Python Feature Engineering Cookbook

Getting ready

Before jumping into the recipe, let's get familiar with some of the parameters of a BoW that we can adjust to make the BoW more or less comprehensive. When creating a BoW over several pieces of text, a new feature is created for each unique word that appears at least once in any of the text pieces we are analyzing. If the word appears only in one piece of text, it will show a value of 1 for that particular text and 0 for all of the others.

Therefore, BoWs tend to be sparse matrices, where most of the values are zeros. Also, the number of columns, that is, the number of words, can be quite large if we work with huge text corpora, and even bigger if we also include n-grams. To limit the number of columns created and the sparsity ...