Getting Started with TensorFlow.js

So far, we have gently introduced ourselves to the wonderful world of deep learning and we have got a fair sense of what deep learning has to offer in terms of making today's web applications more intelligent. In Chapter 1, Demystifying Artificial Intelligence and Fundamentals of Machine Learning, we saw a detailed overview of the web applications before and after AI breakout. In Chapter 3, Creating Your First Deep Learning Web Application, we built ourselves a simple image classifier-based web application using a simple neural network.

Web applications are all around us and they have easily become inseparable parts of our day-to-day lives. When it comes to building web applications, the use of JavaScript is too hard to ignore. So, what if we built an intelligent web application using JavaScript and no other scripting language? In this chapter...

Technical requirements

You can access the code used in this chapter at https://github.com/PacktPublishing/Hands-On-Python-Deep-Learning-for-Web/tree/master/Chapter4.

To work on this chapter, you'll need the following software: 

• TF.js 0.15.1+
• The @tensorflow/tfjs-node 0.3.0+ package from the NPM repository

The fundamentals of TF.js

In this section, we are going to briefly review some of the fundamental concepts of TF.js. We will start off by introducing TensorFlow and then we will proceed to study different components of TF.js. 

What is TensorFlow?

Before we can begin discussing TF.js, we must understand what TensorFlow is. TensorFlow is an open source library that is developed and maintained by Google. It is built on a data structure called tensors. Tensors are the generalized form of scalar and vector. TensorFlow provides a lot of efficient utilities for high-performance numerical computing across a wide range of scientific domains. TensorFlow also provides a very flexible suite of utilities for carrying out machine learning...

Your first deep learning web application with TF.js

In this section, we are going to develop a web application with the help of TF.js. This application will include the steps for a standard, full stack, deep learning-enabled web project. We will begin by preparing the data, we will then study the project architecture briefly, and then, we will proceed toward building the required components as we go.

Preparing the dataset

The Iris flower dataset, in its original form, is a CSV file containing the data of 150 rows split into 5 columns in a comma-separated format, with each entry separated by a new line.

However, we will be using a JSON format of the data for easier operability with JavaScript. The dataset in JSON format can...

Advantages and limitations of TF.js

Let's now summarize some of the advantages TF.js brings over TensorFlow, besides the ones we have already talked about in this chapter:

• Automatic GPU support: You don't need to install CUDA or GPU drivers separately with TF.js to benefit from the GPUs present on the system. This is because the browser itself implements GPU support. 
• Integration: It is fairly simple to integrate TF.js into a web development project using Node.js and then import pretrained models to the project and run them in the browser.

However, it also has several disadvantages that have to be kept in mind whenever developing for production. Some of these are as follows:

• Speed: TF.js is suitable for small datasets. On large-scale datasets, the computation speed suffers heavily and is nearly 10x slower.
• Lack of a tensor board: This great tool, which enables...

Summary

In this chapter, we learned how easy it is to create models with TF.js. You not only get the whole JavaScript ecosystem to work with, but you also get all the pretrained TensorFlow models within TF.js. We developed a simple web application using the Iris dataset and, along the way, we learned about several components that TF.js has to offer. By now, we have already built two simple end-to-end deep learning-based web applications. 

Our progress is indeed apparent. In the upcoming chapters, we will be building our own deep learning APIs and using them to create intelligent web applications. But before that, let's make ourselves familiar with the whole concept of APIs in the next chapter.