Modern Time Series Forecasting with Python: Industry-ready machine learning and deep learning time series analysis with PyTorch and pandas 2nd ed. Edition

I have some financial clients that prefer I use python for their projects. This book came in handy when they needed me to crunch some numbers in their tick data. I like how it presented the information in layered approach that made it simple for me to understand and actually deliver something to my clients.Highly recommend to pick this up if you need to work with time series data with Python.

This book is amazing. I recommend it.

I couldn't have made it as far as I have now without this book. Invaluable!

Time series forecasting was got me interested in AI. Yes, LLMs are cool and all, but I see the ability to make predictions as much cooler. And I entered into a few machine learning contests for fun and ended up making some money.But I never had a book like this. I probably would have done a lot better with it. So whether you need it for your job, you're just curious, or maybe you want to enter some Kaggle contests, this book will be a big help. Time series forecasting is used in weather forecasting, sales forecasting, stock price forecasting, fraud detection, and more. And this book not only introduces you to the fundamental time series concepts and the latest techniques to make your own predictions, but also real-world examples of time series forecasting.

One major, organizational AI integration goal seeks better vision and understanding by finding patterns, and forecasting trends humans may miss. “Modern Time Series Forecasting with Python, 2nd edition” (Packt, 2025) by Manu Josepth and Jeff Tackes provides a step-by-step reference to building efficient forecasting. This exhaustive tome covers foundational math, algorithm development, and effective coding to achieve useful results. Learning is divided in four sections: a basic time series approach, machine learning fundamentals, deep learning principles and the mechanics of forecasting. Recommend this as a desk reference for anyone working with these challenging solutions.Beginning with time series studies mirrors basic data approaches in many areas. The first step begins with understanding the overall goal by determining which data will be studied and how one intends to forecast. Once the strategy is set, one needs to acquire and process data and determine visualization methods. After all, if I can not see the AI conclusions, it will not do any good. Finally, one needs to recognize baselines. Baselines determine variability, even with forecasting, to help determine when changes occurs and if those changes matter.After presenting the basics, the next step exhaustively dives into establishing different interactions with ML tools for time series outputs. One key element here examines stationarity within a time series, the ability to determine whether statistical references remain consistent over time. These measures allow determining if characteristics such as mean, variance, and other aspects connect rationally. For example, if studying temperatures, and the average mean usually only moves a couple of degrees over a day, a much wider swing could suggest either issues in the data, or an underlying condition. The section also discusses comparing multiple time series, increasing variables by comparing elements not initially connected, such as using local temperature, power usage, and then comparing to employment statistics within an area.The deep learning section presents solution for the true expert and may be a little thick if your own experience is lacking. Deep learning has become possible by compute and data increases in hardware and software over the recent years. These solutions require multiple models involving recurrent neural networks and establishing weights for measured factors. The latter chapters in this section provide some excellent strategies as well as a comprehensive review of common deep learning architectures. Comparing the options early, and matching to a strategy can be highly beneficial if you launch your own applications.The last section ties everything together with a look at how the comprehensive models from early can be applied in a forecasting environment. The authors cover multistep forecasting when one model provides answers to another model for a subsequent forecast, extremely valuable in blending multiple local solutions into wider, global conclusions. Not content to leave the work at just forecasting, the book wraps up with discussing validating ML models, and metrics demonstrating common errors. I always love metrics, and enjoyed reviewing the intrinsic and extrinsic metrics common to ML frameworks.I enjoyed the entire book; if I had one complaint, it ran a little long at almost 700 pages. The authors could have easily split the material into two books, one on time series with Python, and a second on deep learning models. Granted, having all the tools in one place is excellent for a reference, but can take quite some time during a casual read. This is a second edition, so the added material likely expanded the initial volume.Overall, “Modern Time Series Forecasting with Python, 2nd Ed” was a truly comprehensive look at time series solutions, initial ML builds, incorporating deep learning solutions, and measuring success. The book covered the initial math at every step, suggested use cases, provided coding samples, and linked to a GitHub for further learning. If you are working with any time-series solutions, I’d purchase this book and keep it on your desk.

Unlike the title there is nothing modern about this book. It starts with creating synthetic data sets using ..gulp... timesynth - that was written in 2017 and can't be installed with pip - enough said. The examples are elementary. 500+ pages!! But lots of figures and most of the pages have quarter of space blank. Probably can condense the material in 300 pages max. The notebooks show lots of warnings. It is important to provide datasets for the example and make sure that the examples in each chapter are self-contained - (a very good example is Raschka's book on Machine Learning) - and that they work. The example depend on some pre-processed data set from chapter 2. If somehow at any point something is messed up then you have to go through the whole step of "pre-processing". I purchased this book after reading reviews and going through table of contents. I realized that this is a very bad idea. To make matter worse I did not get a chance to go through it after it was delivered it and now it is too late to return it. Finally, not a single topic is covered at any reasonable level of depth - just scratches surface and moves on. Isn't forecasting an important area related to time series? Instead of going through the book I can google and find better examples and discussion on various topics.Never buy a book on programming if you do not have time to go through the material and checking out the code.

As someone working hands-on with forecasting problems in real-world data pipelines, Modern Time Series Forecasting with Python (2nd Edition) hits the perfect balance between theory and practical application. Manu Joseph and Jeffrey Tackes have done an outstanding job explaining time series modelling with clear examples, intuitive explanations, and production-ready code snippets.What sets this book apart is its deep dive into both traditional statistical approaches and modern machine/deep learning techniques — from ARIMA and Prophet to PyTorch-based deep learning models. The chapters on feature engineering, backtesting strategies, and handling seasonality are particularly well-written and applicable in industry scenarios.The use of pandas and PyTorch throughout the book makes it super relevant for practitioners today. It’s not just about getting good results on a dataset — it’s about building scalable, explainable, and robust forecasting pipelines.Highly recommend this for data scientists, ML engineers, or anyone looking to level up their time series forecasting skills with a Python-first approach.