The Example-Rich, Hands-On Guide to Data Munging with Apache Hadoop^TM

Data scientists spend much of their time "munging" data: handling day-to-day tasks such as data cleansing, normalization, aggregation, sampling, and transformation. These tasks are both critical and surprisingly interesting. Most important, they deepen your understanding of your data's structure and limitations: crucial insight for improving accuracy and mitigating risk in any analytical project.

Now, two leading Hortonworks data scientists, Ofer Mendelevitch and Casey Stella, bring together powerful, practical insights for effective Hadoop-based data munging of large datasets. Drawing on extensive experience with advanced analytics, the authors offer realistic examples that address the common issues you're most likely to face. They describe each task in detail, presenting example code based on widely used tools such as Pig, Hive, and Spark.

This concise, hands-on eBook is valuable for every data scientist, data engineer, and architect who wants to master data munging: not just in theory, but in practice with the field's #1 platform-Hadoop.

Coverage includes

• A framework for understanding the various types of data quality checks, including cell-based rules, distribution validation, and outlier analysis
• Assessing tradeoffs in common approaches to imputing missing values
• Implementing quality checks with Pig or Hive UDFs
• Transforming raw data into "feature matrix" format for machine learning algorithms
• Choosing features and instances
• Implementing text features via "bag-of-words" and NLP techniques
• Handling time-series data via frequency- or time-domain methods
• Manipulating feature values to prepare for modeling

Data Munging with Hadoop is part of a larger, forthcoming work entitled Data Science Using Hadoop. To be notified when the larger work is available, register your purchase of Data Munging with Hadoop at informit.com/register and check the box "I would like to hear from InformIT and its family of brands about products and special offers."

Ofer Mendelevitch, director of data science at Hortonworks, is responsible for helping customers realize Big Data benefits with Hadoop. He previously served as entrepreneur in residence at XSeed Capital and was vice president of engineering at Nor1. As director of engineering at Yahoo!, he led engineering and data science R&D teams responsible for several large-scale, computational advertising projects.

Casey Stella, Hortonworks principal architect, helps clients of all sizes solve data science problems with Hadoop. He was previously architect and software engineer at Explorys, a start-up spun out of the Cleveland Clinic, focusing on data mining and medical informatics using Hadoop and HBase. He has worked on several ventures dealing with massive amounts of data, including scientific programming in the oil industry, VoIP scalable server infrastructure, and metadata repositories at Oracle.

Preface vi

About the Authors viii

Data Munging with Hadoop 1

Why Hadoop for Data Munging? 2

Data Quality 2

What Is Data Quality? 2

Dealing with Data Quality Issues 3

Using Hadoop for Data Quality 8

The Feature Matrix 9

Choosing the "Right" Features 10

Sampling: Choosing Instances 10

Generating Features 12

Text Features 13

Time-Series Features 16

Features from Complex Data Types 17

Feature Manipulation 18

Dimensionality Reduction 19

Summary 22