Time Series Forecasting using Deep Learning: Combining PyTorch, RNN, TCN, and Deep Neural Network Models to Provide Production-Ready Prediction Solutions

This books goes deep inside deep learningLots of useful and real world examples with great model designs!A must read for time series analysis

Claro y preciso, sin complicaciones matemáticas. Ilustra muy bien los conceptos clave. Considero que su didáctica es muy adecuada.Felicitaciones al autor.

Very good concise starting point for time series. It’s light on theory but the author makes this very clear in the introduction. This is a non-nonsense practical book, and assumes competence in python. The writing style is a bit terse in places but understandably so (given the author’s background), even so - the text is direct and clear. Liked that the author goes back to basics in code so that concepts are not just assumed or obscured by package calls (e.g., tensors, trend removal, alternate models). The more involved model descriptions (e.g., RNN/LSTN/TCN), can be a quite brief, you need to read between the lines, look at other sources, and work through the code.There is less coverage on more advanced topics, confidence intervals, regular and irregular timestamps, complex correlations, optimisations. The author could have cited deeper material and summarised gaps between where this code stops and real-world solutions start, That said, this is a solid starting point for time series. Another 100 pages would have really pushed this up a notch and been equally enjoyable to read. Nice one.

I do work with time series data and deep learning models in practice.Don't expect a super in depth treatment of the subject. That is not what the book tries to doWhat would I suggest is missing for beginners? I think even for beginners a word on uncertainty estimates would have been very important. Point predictions are borderline worthless for real world applications.. The part on financial data is not great. This would have been a great way to show ways how to "break" your own model and do more tests. I didn't look deeper into it, but I am prtty sure the model has no predictive power.But the hack of applying a fiter (Christano Fitzgerald , Baxter King..) to raw input, CNNs for time series and the emphasis of hybrid models show the right spririt: presenting things that can be really valuable in practice.The reader needs to be familiar with python. But the source code is not "production" code, but almost like pseudo code. Easy to read and good for explaining the basic ideas.Like the authors other book on genetic algorithms: for the money paid, it is a really good and hands on starter on the topic. There are many time series deep learning framworks around (GluonTS, PYtorch forecasting,....)

Concise explanations of time series networks and useful examples to know how we can implement the theories

Es un libro bueno a nivel introductorio, ya que no muestra ejemplos para ejecutar en GPU. Podría decir que solo es código comentado para varios ejemplos (buenos).