Time series analysis on AWS book presentation
Want to learn how to build forecasting models and detect anomalies in your time series data while using managed services available from the AWS cloud platform? There’s a book for this!
During the pandemic period, I had the wonderful opportunity to write a book about time series analysis with Packt Publishing. This was my first book and I grew a whole lot during this journey, but more on this in a future post I guess! This book is a deep dive into three AI/ML managed service provided on the AWS cloud, that can deal with time series data. On AWS, managed services are services where the end users only bring their data and configure some parameters. All the other tasks, considered as undifferentiated heavy lifting, are performed on the users’ behalf by the service. This includes the automation of all the infrastructure management, security, planning for scalability, decommissioning unused resources, prepare the data to feed them to various machine learning algorithms, etc.
The three services you will discover in this book are Amazon Forecast (for time series forecasting), Amazon Lookout for Equipment (for multivariate time series anomaly detection particularly suited to predictive maintenance use cases) and Amazon Lookout for Metrics (provided anomaly detection and root cause analysis dedicated to business metrics).
As soon as the book was published I shared the news with my network and I was overwhelmed by the warm welcome it received!
High level overview of the book
Given the interest shown by the community, I decided to share a bit more detail about what you can expect in each part of this book:
- Chapter 1 could have been a part on itself as it gives a general introduction about time series data and what makes this type of dataset stands out from the crowd!
- Chapters 2 to 7 are dedicated to Amazon Forecast: by the end of this first part you will know how to prepare your datasets, train a model and request new forecasts to predict future values for your time series data. For more details about this part, cruise over the following link:
- Chapters 8 to 12 are dedicated to Amazon Lookout for Equipment. If you have sensor data collected from your machines, pieces of industrial equipment or manufacturing processes, you will learn how to train new anomaly detection models, schedule regular executions to detect anomalies and build dashboards to provide your end users with valuable insights. For more details about this section, you can read the following article:
- The last chapters (chapters 13 to 15) are dedicated to Amazon Lookout for Metrics: if you have univariate or multiple time series data, the third part of this book will help you train models that can automatically detect when something goes wrong and even diagnose why it went wrong to support your root cause analysis. This last part is described in more details in this article:
Let’s now have a look at the first introductory chapter…
Outline and summary of chapter 1
In this first chapter, you will discover the different families of time series data and you will have a detailed overview of the different approaches you can use to perform your time series analysis depending on the insights you can to extract from your datasets. Here is a complete outline of this first chapter:
- What is a time series dataset?
- Recognizing the different families of time series: building upon this article I wrote last year, you will learn how to distinguish between univariate time series data, continuous multivariate data, event-based multivariate data and multiple time series data.
- Adding context to time series data: how do you build and/or leverage labels for time series dataset, how can multiple related time series be used to support your analysis, what type of metadata can be useful to reinforce your predictions?
- Learning about common time series challenges: this section will give you an overview about the challenges encountered when dealing with time series datasets. In particular, you will learn about technical challenges (file structure, storage considerations or data quality), behavioral challenges (stationarity or level shifts), missing insights and context and visualization challenges (building and expanding upon this article, this one or this one):
- Selecting an analysis approach: this section will list many approaches you can apply on raw time series data or time series summarized as tabular datasets (extracting features with libraries such as tsfresh, hctsa or featuretools and applying some well-known analysis such as PCA or k-NN). You will also have an overview of how you can leverage symbolic transformations and imaging techniques, transforming your time series into bags-of-words or images before processing them as pictures, something somewhat similar to what I wrote about here:
- Solving for your use case: the last part of this introductory chapter will focus of key use cases you can tackle with time series data (from forecasting to anomaly detection through virtual sensors or activity detection) and the kind of approaches that I’ve seen successful to address them.
Conclusion
Upon completion of this first chapter you will be equipped with a better understanding of the tools and techniques you can leverage to successfully leverage machine learning to uncover business insights from your time series data. You will understand how time series data can vastly differ from one another and you will have a good command of the families of preprocessing, transformation and analysis techniques.
In the next post, I will present the content of the first part of the book, chapters 2 to 7, dedicated to time series forecasting with Amazon Forecast.
The book is now available worldwide on Amazon. Here are a few links:
I hope you’ll find it interesting, feel free to leave me a comment here and don’t hesitate to subscribe to my Medium email feed if you don’t want to miss my upcoming posts! Subscription link is here:
Want to support me and future work? Join Medium with my referral link!
Enjoy!
