Data can make or break your machine learning (ML) algorithm. This chapter will lay the basic ground work for manipulating data, augmenting data, and balancing imbalanced datasets or data with massive outliers. Each recipe will provide a guide on how to use open source libraries to accelerate our Generative Adversarial Network (GAN) training.

Data is the lifeblood of ML algorithms. Your models will only be as good as the data you provide to them. After all, you are what you eat. We have to focus on developing a good, clean dataset for learning. This begins with getting an environment set up and preparing the data to be ingested into an algorithm. We do have a fundamental advantage within this process because GANs can take considerably smaller sets of data than other techniques. This advantage comes with the explicit caveat that we will need to ensure that the data we're using encompasses the entire trade space of possibilities for our application.

One of the deep dark secrets they don't teach you about this field...

What's a development environment? Everyone thinks setting up a development environment needs to be this incredibly arduous process. The installation process could be worse. It's actually quite simple and I intend to show you the basics in this recipe.

Let's lay out the requirements for the equipment you'll need to be successful in this book:

• GPU: 10 series CUDA-enabled Nidea GPU 
• Operating system: Ubuntu Linux 16.04+
• CPU/RAM: i5 or i7 with at least 8 GB of RAM

First and foremost, the GPU is a requirement for this type of book. Although these algorithms can technically train on a CPU, it could take days in some cases for a single model...

In computer science, data types will represent the way the data is stored in the program. For this section, we are going to discuss the structure of the MNIST data and demonstrate how to simply manipulate the data.

Focusing on data types is where we will begin the data processing journey. Each step in this process is crucial to understand. Data types refer to the structure in which the data is held in Python. Think of a dictionary, array of floats, and so on. These are the data types that we would like to understand and consider. The example that we are going to explore in this set of recipes is going to be a parsing example for images since the first few recipes in this book will revolve around...

Data preprocessing is the explicit process of ensuring that your data can be ingested into your algorithm simply. In this section, you will learn how to work with data for ML in future sections.

Why even worry about preprocessing? It's easy to overlook the easy steps. As we ingest data into our algorithms, we'll need to ensure that each of the data points is both useful and accurate. This means we need to ensure that both the X data and Y labels, in a supervised learning problem space, are correct prior to going to a learner. So, how do we ensure that each of the data points is correct? For large datasets, we can look at macro metrics such as a three sigma outlier. For smaller datasets...

Anomalous data is the risk that your data is not evenly distributed or easily separable. Datasets from the real world are going to contain outliers and data that needs to be adjusted. In this recipe, we will discuss a basic technique used in data analysis to work with anomalous data and distribute the results while maintaining the data distribution.

Outliers are a huge issue with datasets where you want to have a clean distribution of data. In terms of the generative model, we are interested in ensuring that the model can find the right representation of the distribution and model it appropriately. This recipe is going to focus on the tools you will use in these instances to solve problems with...

Balancing data and handling anomalous data are often thought of as the same process. In our case, data balancing involves understanding the techniques used to spread anomalous data without disrupting the underlying data distribution. In this recipe, we will discuss the core concepts in data balancing.

Generative modeling is attempting to build a model that represents the entire data distribution. In order to learn this underlying distribution, the data must represent that data in a verbose but compact form—that is, we want to ensure that each of the traits on features that we are attempting to learn, is represented in similar quantities the way in which they would be generated.

Data augmentation is the idea that one image can be altered or corrupted to encourage deep learning techniques to generalize for the objective, rather than focusing on single features. In this section, we'll show a simple script for applying different augmentations.

The imgaug library is commonly used in deep learning research and this figure demonstrates a subset of available augmentations in this free-to-use library:

Data augmentation is a cornerstone of deep learning data analysis. Each project needs to understand how data augmentation can improve their project. Why would you choose to include data augmentation in your project? In images, it's easy to understand. By augmenting...