Mastering NLP: In-Depth Python Coding for Deep Learning Models | by Eligijus Bujokas | Oct, 2023

This article came to fruition after reading numerous documentation resources and looking at videos on YouTube about textual data, classification, recurrent neural networks, and other hot subjects on how to develop a machine-learning project using text data. A lot of the information is not that user-friendly and some of the parts are obfuscated, thus, I want to save the reader a lot of time and shed light on the most important concepts in using textual data in any machine learning project.

The supporting code for the examples presented here can be found at: https://github.com/Eligijus112/NLP-python

The topics covered in this article will be:

• Converting text to sequences
• Converting sequence indexes to embedded vectors
• In-depth RNN explanation
• The loss function for classification
• Full NLP pipeline using Pytorch

NLP stands for Natural Language Processing¹. This is a huge topic about how to use both hardware and software in tasks like:

• Translating one language to another
• Text classification
• Text summarization
• Next token prediction
• Named entity recognition

And much much more. In this article, I want to cover the most popular techniques and familiarize the reader with the concepts by simple and coded examples.

A lot of tasks in NLP start by tokenizing the text².

Text tokenization is a process where we split the original text into smaller parts — tokens. The tokens can be either characters, subwords, words, or a mix of all three.

Consider the string:

“NLP in Python is fun and very well documented. Let’s get started!”

I will use word-level tokens because the same logic would apply to lower-level tokenization as well.