This chapter will cover the following topics:
- Getting sample data from external sources
- Creating sample data for toy analysis
- Confirming the characteristics of created data
- Scaling data to the standard normal
- Creating binary features through thresholding
- Working with categorical variables
- Binarizing label features
- Imputing missing values through various strategies
- Using Pipelines for multiple preprocessing steps
- Reducing dimensionality with PCA
- Using factor analytics for decomposition
- Kernel PCA for nonlinear dimensionality reduction
- Using truncated SVD to reduce dimensionality
- Decomposition to classify with DictionaryLearning
- Putting it all together with Pipelines
- Using Gaussian processes for regression
- Defining the Gaussian process object directly
- Using stochastic gradient descent for regression
This chapter discusses setting data, preparing data, and premodel dimensionality reduction. These are not the attractive parts of machine learning (ML), but they often turn out to be what determines if a model will work or not.
There are three main parts to the chapter. Firstly, we'll create fake data; this might seem trivial, but creating fake data and fitting models to fake data is an important step in model testing. It's more useful in situations where we implement an algorithm from scratch, but I'll cover it here for completeness, and in the event you don't have data of your own, you can just create it. Secondly, we'll look at broadly handling data transformations as a preprocessing step, which includes data imputation, categorical variable encoding, and so on. Thirdly, we'll look at situations where we have a large number of features relative to the number of observations we have.
This chapter, especially the first half, will set the stage for the later chapters. In order to use scikit-learn, data is required. The first two sections will discuss acquiring the data; the rest of the first half will discuss preparing this data for use.