Book Image

Essential PySpark for Scalable Data Analytics

By : Sreeram Nudurupati
Book Image

Essential PySpark for Scalable Data Analytics

By: Sreeram Nudurupati

Overview of this book

Apache Spark is a unified data analytics engine designed to process huge volumes of data quickly and efficiently. PySpark is Apache Spark's Python language API, which offers Python developers an easy-to-use scalable data analytics framework. Essential PySpark for Scalable Data Analytics starts by exploring the distributed computing paradigm and provides a high-level overview of Apache Spark. You'll begin your analytics journey with the data engineering process, learning how to perform data ingestion, cleansing, and integration at scale. This book helps you build real-time analytics pipelines that help you gain insights faster. You'll then discover methods for building cloud-based data lakes, and explore Delta Lake, which brings reliability to data lakes. The book also covers Data Lakehouse, an emerging paradigm, which combines the structure and performance of a data warehouse with the scalability of cloud-based data lakes. Later, you'll perform scalable data science and machine learning tasks using PySpark, such as data preparation, feature engineering, and model training and productionization. Finally, you'll learn ways to scale out standard Python ML libraries along with a new pandas API on top of PySpark called Koalas. By the end of this PySpark book, you'll be able to harness the power of PySpark to solve business problems.
Table of Contents (19 chapters)
1
Section 1: Data Engineering
6
Section 2: Data Science
13
Section 3: Data Analysis

Upgrading pandas to PySpark using Koalas

pandas is the defacto standard for data processing in standard Python, the same as Spark has become the defacto standard for distributed data processing. The pandas API is Python-related and leverages a coding style that makes use of Python's unique features to write code that is readable and beautiful. However, Spark is based on the JVM, and even the PySpark draws heavily on the Java language, including in naming conventions and function names. Thus, it is not very easy or intuitive for a pandas user to switch to PySpark, and a considerable learning curve is involved. Moreover, PySpark executes code in a distributed manner and the user needs to understand the nuances of how distributed code works when intermixing PySpark code with standard single-node Python code. This is a deterrent to an average pandas user to pick up and use PySpark. To overcome this issue, the Apache Spark developer community came up with another open source library...