Book Image

Python Data Cleaning Cookbook

By : Michael Walker
Book Image

Python Data Cleaning Cookbook

By: Michael Walker

Overview of this book

Getting clean data to reveal insights is essential, as directly jumping into data analysis without proper data cleaning may lead to incorrect results. This book shows you tools and techniques that you can apply to clean and handle data with Python. You'll begin by getting familiar with the shape of data by using practices that can be deployed routinely with most data sources. Then, the book teaches you how to manipulate data to get it into a useful form. You'll also learn how to filter and summarize data to gain insights and better understand what makes sense and what does not, along with discovering how to operate on data to address the issues you've identified. Moving on, you'll perform key tasks, such as handling missing values, validating errors, removing duplicate data, monitoring high volumes of data, and handling outliers and invalid dates. Next, you'll cover recipes on using supervised learning and Naive Bayes analysis to identify unexpected values and classification errors, and generate visualizations for exploratory data analysis (EDA) to visualize unexpected values. Finally, you'll build functions and classes that you can reuse without modification when you have new data. By the end of this Python book, you'll be equipped with all the key skills that you need to clean data and diagnose problems within it.
Table of Contents (12 chapters)

Removing duplicated rows

There are several reasons why we might have data duplicated at the unit of analysis:

  • The existing DataFrame may be the result of a one-to-many merge, and the one side is the unit of analysis.
  • The DataFrame is repeated measures or panel data collapsed into a flat file, which is just a special case of the first situation.
  • We may be working with an analysis file where multiple one-to-many relationships have been flattened, creating many-to-many relationships.

When the one side is the unit of analysis, data on the many side may need to be collapsed in some way. For example, if we are analyzing outcomes for a cohort of students at a college, students are the unit of analysis; but we may also have course enrollment data for each student. To prepare the data for analysis, we might need to first count the number of courses, sum the total credits, or calculate the GPA for each student, before ending up with one row per student. To generalize from...