Chapter 2: Getting off the Ground with the First Deep Learning Model

Deep learning (DL) models have gained tremendous popularity in recent times and have caught the attention of data scientists in academia and industry alike. The reason behind their great success is their ability to solve the simplest yet oldest problem in computer science—computer vision. It had long been the dream of computer scientists to find an algorithm that would make machines see like humans do… or at least be able to recognize objects. DL models power not just object recognition but are used in everything, from predicting who is in an image, to natural language processing (NLP)—where they can be used for predicting and generating text and understanding speech—and even creating deepfakes, such as videos. At their core, all DL models are built using neural network (NN) algorithms; however, they are much more than just a NN. While NNs have been used since the 1950s, it's only...

Technical requirements

The code for this chapter has been developed and tested on macOS with Anaconda or in Google Colaboratory (Google Colab) with Python 3.7. If you are using another environment, please make appropriate changes to your env variables.

In this chapter, we will primarily be using the following Python modules, mentioned with their versions:

• PyTorch Lightning (version: 1.5.2)
• Seaborn (version: 0.11.2)
• NumPy (version: 1.21.5)
• Torch (version: 1.10.0)
• pandas (version: 1.3.5)

!pip install torch==1.10.0 torchvision==0.11.1 torchtext==0.11.0 torchaudio==0.10.0 --quiet
!pip install pytorch-lightning==1.5.2 --quiet

In order to make sure that these modules work together and not go out of sync, we have used the specific version of torch, torchvision, torchtext, torchaudio with PyTorch Lightning 1.5.2. You can also use the latest version of PyTorch Lightning and torch compatible with each other. More details can be found on the GitHub link: https...

Getting started with Neural Networks

In this section, we will begin our journey by understanding the basics of Neural Networks.

Why Neural Networks?

Before we go deep into NNs, it is important to answer a simple question: Why do we even need a new classification algorithm when there are so many existing classification algorithms, such as decision trees? The simple answer is that there are some classification problems that decision trees would never be able to solve. As you might be aware, decision trees work by finding a set of objects in one class and then creating splits in the set to continue to create a pure class. This works well when there is a clear distinction between different classes in the dataset, but it fails when they are mixed. One such very basic problem that decision trees cannot ever solve is the XOR problem.

About the XOR operator

The XOR gate/operator is also known as exclusive OR. It is a digital logic in electronics. An XOR gate is a digital logic...

Building a Hello World MLP model

Welcome to the world of PyTorch Lightning!

Finally, it's time for us to build our first model using PyTorch Lightning. In this section, we will build a simple MLP model to accomplish the XOR operator. This is like a Hello World introduction to the world of NNs as well as PyTorch Lightning. We will follow these steps to build our first XOR operator:

1. Importing libraries
2. Preparing the data
3. Configuring the model
4. Training the model
5. Loading the model
6. Making predictions

Importing libraries

We begin by first importing the necessary libraries and printing their package versions, as follows:

import pytorch_lightning as pl
import torch
from torch import nn, optim
from torch.autograd import Variable
import pytorch_lightning as pl
from pytorch_lightning.callbacks import ModelCheckpoint
from torch.utils.data import DataLoader
print("torch version:",torch.__version__)
print("pytorch ligthening version...

Building our first Deep Learning model

Now that we have built a basic NN, it's time to use our knowledge of creating an MLP to build a DL model. You will notice that the core framework will remain the same and is built upon the same foundation.

So, what makes it deep?

While the exact origins of who first used DL are often debated, a popular misconception is that DL just involves a really big NN model with hundreds or thousands of layers. While most DL models are big, it is important to understand that the real secret is a concept called backpropagation.

As we have seen, NNs such as MLPs have been around for a long time, and by themselves, they could solve previously unsolved classification problems such as XOR or give better predictions than traditional classifiers. However, they were still not accurate when dealing with large unstructured data such as images. In order to learn in high-dimensional spaces, a simple method called backpropagation is used, which gives feedback...

Building a CNN model for image recognition

PyTorch Lightning is a versatile framework, which makes training and scaling DL models easy by focusing on building models than writing complex programs. PyTorch Lightning is bundled with many useful features and options for building DL models. Since it is hard to cover all the topics in a single chapter, we will keep exploring different features of PyTorch Lightning in every chapter.

Here are the steps for building an image classifier using a CNN:

1. Importing the packages
2. Collecting the data
3. Preparing the data
4. Building the model
5. Training the model
6. Evaluating the accuracy of the model

Importing the packages

We will get started using the following steps:

1. First things first—install and load the necessary packages, as follows:

   !pip install torch==1.10.0 torchvision==0.11.1 torchtext==0.11.0 torchaudio==0.10.0 --quiet
   !pip install pytorch-lightning==1.5.2 --quiet
   !pip install opendatasets -...

Summary

We got a taste of basic MLPs and CNNs in this chapter, which are the building blocks of DL. We learned that by using the PyTorch Lightning framework, we can easily build our models. While MLPs and CNNs may sound like basic models, they are quite advanced in terms of business applications, and many companies are just warming up to their industrial use. Neural Networks are used very widely as classifiers on structured data for predicting users' likes or propensity to respond to an offer or for marketing campaign optimization, among many other things. CNNs are also widely used in many industrial applications, such as counting the number of objects in an image, recognizing car dents for insurance claims, facial recognition to identify criminals, and so on.

In this chapter, we saw how to build the simplest yet most important XOR operator using an MLP model. We further extended the concept of MLPs to build our first CNN DL model to recognize images. Using PyTorch Lightning...