Preface

This book is a practical guide to data cleaning, broadly defined as all tasks necessary to prepare data for analysis. It is organized by the tasks usually completed during the data-cleaning process: importing data, viewing data diagnostically, identifying outliers and unexpected values, imputing values, tidying data, and so on. Each recipe walks the reader from raw data through the completion of a specific data-cleaning task.

There are already a number of very good pandas books. Unsurprisingly, there is some overlap between those texts and this one. However, the emphasis here is different. I focus as much on the why as on the how in this book.

Since pandas is still relatively new, the lessons I have learned about cleaning data have been shaped by my experiences with other tools. Before settling into my current work routine with Python and R about 10 years ago, I relied mostly on C# and T-SQL in the early 2000s, SAS and Stata in the 90s, and FORTRAN and Pascal in the 80s. Most readers of this text probably have experience with a variety of data-cleaning and analysis tools. In many ways the specific tool is less significant than the data preparation task and the attributes of the data. I would have covered pretty much the same topics if I had been asked to write The SAS Data Cleaning Cookbook or The R Data Cleaning Cookbook. I just take a Python/pandas-specific approach to the same data-cleaning challenges that analysts have faced for decades.

I start each chapter with how to think about the particular data-cleaning task at hand before discussing how to approach it with a tool from the Python ecosystem—pandas, NumPy, Matplotlib, and so on. This is reinforced in each recipe by a discussion of the implications of what we are uncovering in the data. I try to connect tool to purpose. For example, concepts like skewness and kurtosis matter as much for handling outliers as does knowing how to update pandas Series values.