It covers almost every important topics and the code provided by the author is really helpful. Easy to follow and great inspiration.

The book is about 500+ pages long. But the text is in smaller font than other books so basically the book can be 50% bigger if it was in a slightly larger font. In addition, Manu has left out all the code he has written from the book (but you’ll find it in the GitHub for the book and it’s free!). If he had published all the code that he uses and references in the book, the book can easily be 1500 pages long. That why I call it massive!

Why do I like this book so much?

This book is very detailed, very thorough when it comes to explaining the huge number of techniques that one uses in forecasting from ARIMA to exponential smoothing to ML to deep learning and now transformers. Manu does not leave anything behind.

In addition, Manu explains how to overcome hurdles with each technique by explaining their pitfalls and takes you along on a joyous ride from beginning to end.

I have gone through the most exciting chapters for me which begin from chapter 8 through 17. These address some of the most common techniques we use in forecasting in the real world. The book provides example notebooks for each technique from Linear regression to temporal fusion transformers (TFT). The interesting thing is Manu has coded almost all of the notebooks himself.

So I thought I will give everyone a heads-up about this fantastic book while I spend a month or so finishing it. Until then, if you can afford it and have the need to solve forecasting problems in your work, make sure you buy this book or get your company to pay for it. It’ll save you tons of time and money! Highly recommended!

#ml #deeplearning #forecasting #pytorch

Of several recent books on time series and forecasting in Python, this is arguably the best. The introductory chapters deal with several key issues in modern time series analysis. It is understood that real world time series exhibit properties such as serial correlation, cyclicality and nonstationarity, which are rarely addressed in older textbooks. The sections on classical forecasting methods include not only linear regression but exponential smoothing, ARIMA and spectral analysis. The sections on artificial intelligence include newer techniques such as recurrent neural nets. The book deals with forecasting over multiple intervals, a topic that is more complex in AI applications than might at first sight be apparent. Although some very advanced approaches are not dealt with in detail, this book provides an excellent introduction. A nice feature is the inclusion of sample code.

If you want to develop an understanding and expertise in time series, this post is for you!

The book provides a 360 degrees view of time series modeling. This book has everything from essential components of time series modeling to deep learning models.

Key Highlights:

1. A detailed explanation of different kinds of forecasting, such as multivariate forecasting, and in-sample/out-sample, is crucial to understanding picking the right kind of modeling approach.

2. Detailed explanation on how to perform data wrangling to prepare your data for a successful time series modeling model, like treating missing values.

3. Proper explanation of Brain functioning and Linear Algebra before diving deep into the Neural Network Model. The author has provided accurate intuition to set you up for success, even if you are reading about Deep Learning for the first time.

Some new concepts that you might learn:

1. Temporal embeddings and time delay embeddings and the benefits of creating global models over the local model

2. What is data leakage and ways to protect it

3. Multi-task learning, like Global Forecasting Models

4. Novel ways to validate your forecast beyond traditional metrics

The book is perfect for:

1. Data Science Practitioners looking to upgrade their knowledge of time series modeling

2. Aspiring Data Scientists looking to initiate their knowledge on the subject

3. You are an expert in ML aspects of Time Series but want to gain expertise using Deep Learning for forecasting and vice versa.

As a bonus, this book has python codes and practitioner tips in most chapters. The book reflects Manu Joseph's expertise and grasp of theoretical concepts.

I found this book to be a priceless resource as a student who has three years of expertise in data science and machine learning and is currently doing a master's in AI. I've been practically applying the concepts of this book to realtime industrial capstone projects. The book offers a rich explanation of time series forecasting that is written in an easy to digest manner. It offers a thorough review of all the crucial facets of time series modeling and forecasting for people who are just beginning their careers in machine learning. The book also offers machine learning experts a road map for using the most recent deep-learning methods to improve their time series models. The PyTorch Forecasting library is covered throughout the book in a self-contained and simple-to-follow manner, especially for people who are new to PyTorch, which makes it a very helpful resource. I've been testing these methods in my regular work and getting fantastic outcomes.

The book's layout is carefully thought out, with four main sections that progress from a fundamental introduction to time series to the usage of deep learning architectures. The material of each chapter is made simple yet insightful to follow through the use of mathematical illustrations, graphs, flow diagrams, tables, photos, and code snippets.

I would like to extend my heartfelt congratulations to Manu for sharing his expertise and insights on a broad range of concepts that are relevant to solving the challenges faced by industry professionals today.

This book has been an engaging read and I can see myself referencing it for years to come. As a data scientist with more than 5 years of experience in time series forecasting & supply chain analytics, I can say Manu has covered the majority of concepts industry professionals will find useful for today's problems: everything from the nuts and bolts of building ML models to the real-world issues like data leakage. The section on Global Forecasting Models was exceptionally well written and pulled together a variety of important concepts into a compelling and actionable summary.

I'm most excited by the elaborate Deep Learning section and can't wait to try out some of these cutting-edge techniques on my datasets. I can speak specifically to the Temporal Fusion Transformer section, as I have experience using that architecture before reading this book. The way the concepts were presented gave me a sense of clarity that I struggled to get from the research papers on the topic. I'm a strong believer that transformer models will take the forecasting industry by storm this decade. We have the data quantity and software packages available to us, but practitioners have been lacking detailed technical instructions for how to successfully train and deploy such models. Manu has provided a valuable contribution to our industry and I'm excited to see what potential it unlocks for forecasting professionals.

Modern Time Series Forecasting with Python is a comprehensive guide to using Python for time series forecasting. The book is well-written and easy to follow, making it accessible to readers with a range of backgrounds and experience levels. One of the best things about this book is the wide range of techniques and algorithms it covers, from traditional methods like ARIMA to more modern techniques such as deep learning. The author does a great job of explaining the theory behind each technique and providing practical examples of how to implement them in Python. Another great aspect of the book is the focus on real-world examples and case studies, which helps readers understand how the techniques can be applied in the context of real problems. Additionally, the book has a lot of code examples that are very helpful in understanding the concepts. Overall, Modern Time Series Forecasting with Python is an excellent resource for anyone interested in time series forecasting and machine learning.

Author gives sparse examples for neural networks. Deep learning and neural networks were the reason for buying the book. Yet the author doesn't adequately explain how to prepare datasets for use in PyTorch. The few examples he provides use different notation, making the the learning process unnecessarily convoluted. It seems as though he had cobbled together examples from others instead of creating his own examples. The short snipped from another book 'Python for Finance' provides a much better example of Machine Learning and Neural Net code - making it easier to understand how to implement in my own projects. Quite the let down, as this book was supposed to be an expansion into Neural Networks and Deep learning.

Easy to follow, and the code that accompanies the code really helps to assimilate the material.

I like the book. It contains lots of information that you should experiment with.

The book takes the reader on a journey from the most basic of TS concepts and models, all the way to the advanced level.

The best part about the book for me is that, the author has not just explained the topics but has also tried to equip the reader with the tools and explanations, that can help one understand the concepts intuitively. Have come across many books that leave you with the formulas, but not many that bring them home. This book bridges that gap to a very large extent.

10/10 a must read!

I feel that any teaching material and course books should be particularly reliable in terms of content quality. No one is immune from the occasional typo but trying to explain a very simple mathematical formula which is wrong is just negligence mixed with incompetence and ignorance. Who edited this book ? Some journalistic or philosophy major ?

Even without knowing the definition of SSE (Sum of Squared Errors), a slightly competent person would see how the kaboudan metric equation cannot approach 0 when a signal is predictable and 1 when not.

Title says it all