You're reading from Python Machine Learning By Example Unlock machine learning best practices with real-world use cases

Product type Paperback

Published in Jul 2024

Publisher Packt

ISBN-13 9781835085622

Length 518 pages

Edition 4th Edition

Languages

Python

Tools

PyTorch

Concepts

Computer Vision

Author (1):

Yuxi (Hayden) Liu

View More author details

Table of Contents (18) Chapters

Preface

1. Getting Started with Machine Learning and Python

2. Building a Movie Recommendation Engine with Naïve Bayes FREE CHAPTER

3. Predicting Online Ad Click-Through with Tree-Based Algorithms

4. Predicting Online Ad Click-Through with Logistic Regression

5. Predicting Stock Prices with Regression Algorithms

6. Predicting Stock Prices with Artificial Neural Networks

7. Mining the 20 Newsgroups Dataset with Text Analysis Techniques

8. Discovering Underlying Topics in the Newsgroups Dataset with Clustering and Topic Modeling

9. Recognizing Faces with Support Vector Machine

10. Machine Learning Best Practices

11. Categorizing Images of Clothing with Convolutional Neural Networks

12. Making Predictions with Sequences Using Recurrent Neural Networks

13. Advancing Language Understanding and Generation with the Transformer Models

14. Building an Image Search Engine Using CLIP: a Multimodal Approach

15. Making Decisions in Complex Environments with Reinforcement Learning

16. Other Books You May Enjoy

17. Index

Overcoming long-term dependencies with LSTM

Let’s start with the vanishing gradient issue in vanilla RNNs. Where does it come from? Recall that during backpropagation, the gradient decays along with each time step in the RNN (that is, s_t=a(Ux_t+Ws_t-1); early elements in a long input sequence will have little contribution to the computation of the current gradient. This means that vanilla RNNs can only capture the temporal dependencies within a short time window. However, dependencies between time steps that are far away are sometimes critical signals to the prediction. RNN variants, including LSTM and gated recurrent units (GRUs), are specifically designed to solve problems that require learning long-term dependencies.

We will be focusing on LSTM in this book as it is a lot more popular than GRU. LSTM was introduced a decade earlier and is more mature than GRU. If you are interested in learning more about GRU and its applications, feel free to check out Hands-On Deep...