Chapter 1, Foreseeing Variable Problems in Building ML Models, covers how to identify the different problems that variables may present and that challenge machine learning algorithm performance. We'll learn how to identify missing data in variables, quantify the cardinality of the variable, and much more besides.
Chapter 2, Imputing Missing Data, explains how to engineer variables that show missing information for some observations. In a typical dataset, variables will display values for certain observations, while values will be missing for other observations. We'll introduce various techniques to fill those missing values with some additional values, and the code to execute the techniques.
Chapter 3, Encoding Categorical Variables, introduces various classical and widely used techniques to transform categorical variables into numerical variables and also demonstrates a technique for reducing the dimension of highly cardinal variables as well as how to tackle infrequent values. This chapter also includes more complex techniques for encoding categorical variables, as described and used in the 2009 KDD competition.
Chapter 4, Transforming Numerical Variables, uses various recipes to transform numerical variables, typically non-Gaussian, into variables that follow a more Gaussian-like distribution by applying multiple mathematical functions.
Chapter 5, Performing Variable Discretization, covers how to create bins and distribute the values of the variables across them. The aim of this technique is to improve the spread of values across a range. It also includes well established and frequently used techniques like equal width and equal frequency discretization and more complex processes like discretization with decision trees and many more.
Chapter 6, Working with Outliers, teaches a few mainstream techniques to remove outliers from the variables in the dataset. We'll also learn how to cap outliers at a given arbitrary minimum/maximum value.
Chapter 7, Deriving Features from Dates and Time Variables, describes how to create features from dates and time variables. Date variables can't be used as such to build machine learning models for multiple reasons. We'll learn how to combine information from multiple time variables, like calculating time elapsed between variables and also, importantly, working with variables in different time zones.
Chapter 8, Performing Feature Scaling, covers the methods that we can use to put the variables within the same scale. We'll also learn how to standardize variables, how to scale to minimum and maximum value, how to do mean normalization or scale to vector norm, among other techniques.
Chapter 9, Applying Mathematical Computations to Features, explains how to create new variables from existing ones by utilizing different mathematical computations. We'll learn how to create new features through the addition/difference/multiplication/division of existing variables and more. We will also learn how to expand the feature space with polynomial expansion and how to combine features using decision trees.
Chapter 10, Creating Features with Transactional and Time Series Data, covers how to create static features from transactional information, so that we obtain a static view of a customer, or client, at any point in time. We'll learn how to combine features using math operations, across transactions, in specific time windows and capture time between transactions. We'll also discuss how to determine time between special events. We'll briefly dive into signal processing and learn how to determine and quantify local maxima and local minima.
Chapter 11, Extracting Features from Text Variables, explains how to derive features from text variables. We'll learn to create new features through the addition of existing variables. We will learn how to capture the complexity of the text by capturing the number of characters, words, sentences, the vocabulary and the lexical variety. We will also learn how to create Bag of Words and how to implement TF-IDF with and without n-grams
