Deep Learning with Structured Data

Tabular data is everywhere, but turning it into a useful deep-learning model takes more than choosing an architecture. In Deep Learning with Structured Data, Mark Ryan works through the practical decisions behind that process, using a Toronto transit dataset and a streetcar-delay prediction project as a guide.

The focus is the complete applied workflow: prepare the data, build and train a Keras model, evaluate experiments, and take the result toward deployment. For readers with intermediate Python and machine-learning experience, this is a grounded look at where deep learning can fit into real-world structured-data work.

Follow a project from raw data to deployment

The book moves through the stages of an applied data-science project rather than treating modeling as an isolated task. It begins with the problem and the data, then addresses exploration, cleansing, transformation, model construction, training, and experimentation. The later chapters turn to deployment and possible next steps.

Give structured data the preparation it needs

Data preparation receives substantial attention. Ryan covers exploring and cleaning the dataset, then transforming it for model use. That progression makes the connection between the source data and the eventual model easier to follow—and keeps the practical work of preparing tabular information in view alongside the deep-learning techniques.

Build, test, and refine a Keras model

With the data prepared, the book develops a Keras deep-learning model and explores its training and performance. The experiments help frame model building as an iterative process: train, measure, and examine what the results suggest before moving forward.

Think beyond the training run

Deployment is part of the project, not an afterthought. The book also discusses monitoring and ways to extend the streetcar-delay example, including the potential value of additional data. Its closing guidance considers how to judge whether another structured-data problem might be a suitable candidate for deep learning.

Who will find this useful?

This book is aimed at readers with intermediate Python and machine-learning skills who want to explore deep learning for tabular data and relational-database use cases. Data-science practitioners will find an end-to-end project perspective, from working with imperfect data to considering a model’s place in a deployed system.

For a practical treatment of structured-data deep learning, Ryan keeps the emphasis on the work around the model as well as the model itself.