Working with MXNet and Visualizing Datasets – Gluon and DataLoader

In the previous chapter, we learned how to set up MXNet. We also verified how MXNet could leverage our hardware to provide maximum performance. Before applying deep learning (DL) to solve specific problems, we need to understand how to load, manage, and visualize the datasets we will be working with. In this chapter, we will start using MXNet to analyze some toy datasets in the domains of numerical regression, data classification, image classification, and text classification. To manage those tasks efficiently, we will see new MXNet libraries and functions such as Gluon (an API for DL) and DataLoader.

In this chapter, we will cover the following topics:

• Understanding regression datasets – loading, managing, and visualizing the House Sales dataset
• Understanding classification datasets – loading, managing, and visualizing the Iris dataset
• Understanding image datasets – loading...

Technical requirements

Apart from the technical requirements specified in the Preface, no other requirements apply to this chapter.

The code for this chapter can be found at the following GitHub URL: https://github.com/PacktPublishing/Deep-Learning-with-MXNet-Cookbook/tree/main/ch02

Furthermore, you can directly access each recipe from Google Colab; for example, for the first recipe of this chapter, visit https://colab.research.google.com/github/PacktPublishing/Deep-Learning-with-MXNet-Cookbook/blob/main/ch02/2_1_Toy_Dataset_for_Regression_Load_Manage_and_Visualize_House_Sales_Dataset.ipynb.

Understanding regression datasets – loading, managing, and visualizing the House Sales dataset

The training process of machine learning (ML) models can be divided into three main sub-groups:

• Supervised learning (SL): The expected outputs are known for at least some data
• Unsupervised learning (UL): The expected outputs are not known but the data has some features that could help with understanding its internal distribution
• Reinforcement learning (RL): An agent explores the environment and makes decisions based on the inputs acquired from the environment

There is also an approach that falls in between the first two sub-groups called weakly SL, where there are not enough known outputs to follow an SL approach for one of the following reasons:

• The outputs are inaccurate
• Only some of the output features are known (incomplete)
• They are not exactly the expected outputs but are connected/related to the task we intend to achieve (inexact)

...

Understanding classification datasets – loading, managing, and visualizing the Iris dataset

In the previous recipe, we studied one of the most common problem types in SL: regression. In this recipe, we will take a closer look at another of these problem types: classification.

In classification problems, we want to estimate a categorial output, a class, from a set of given classes, using a variable number of input features. In this recipe, we will analyze a toy classification dataset from Kaggle: the Iris dataset, one of the most renowned classification datasets.

The Iris dataset presents the problem of estimating the iris class of the flower of plants, from three classes (iris setosa, iris versicolor, and iris virginica) with the help of the following four features:

• Sepal length (in cm)
• Sepal width (in cm)
• Petal length (in cm)
• Petal width (in cm)

These data features are provided for 150 flowers, with 50 instances for each of the 3 classes (making...

Understanding image datasets – loading, managing, and visualizing the Fashion-MNIST dataset

One of the fields that has grown considerably in DL in the last years has been computer vision (CV). Since the AlexNet revolution in 2012, CV has expanded from lab research to surpassing human performance in real-world datasets (known as “in the wild”).

In this recipe, we will explore the simplest CV task: image classification. Given a set of images, our task is to correctly classify that image among a given set of labels (classes).

One of the most classic image classification datasets is the MNIST (which stands for the Modified National Institute of Standards and Technology) database. Similarly sized, but more suited for current CV analysis, is the Fashion-MNIST dataset. This dataset is a multi-label image classification dataset, with a training set of 60k examples and a test set of 10k examples, with each example belonging to 1 of these 10 categories (starting with...

Understanding text datasets – loading, managing, and visualizing the Enron Email dataset

Another field that has grown considerably in DL in recent years is natural language processing (NLP). Similarly to CV, this field aims to surpass human performance in real-world datasets.

In this recipe, we will explore one of the simplest NLP tasks: text classification. Given a set of sentences and paragraphs, our task is to correctly classify that text among a given set of labels (classes).

One of the most classic text classification tasks is to distinguish whether received email is spam or not (ham). These datasets are binary text classification datasets (only two labels to assign, 0 and 1, or ham and spam).

In our specific scenario, we will use a real-world email dataset. This set of emails was made public during the investigation of the Enron scandal in the early 2000s by the US Government. This dataset was first published in 2004 and is composed of emails from ~150 users,...