Book Image

Hands-On Data Analysis with Scala

By : Rajesh Gupta
Book Image

Hands-On Data Analysis with Scala

By: Rajesh Gupta

Overview of this book

Efficient business decisions with an accurate sense of business data helps in delivering better performance across products and services. This book helps you to leverage the popular Scala libraries and tools for performing core data analysis tasks with ease. The book begins with a quick overview of the building blocks of a standard data analysis process. You will learn to perform basic tasks like Extraction, Staging, Validation, Cleaning, and Shaping of datasets. You will later deep dive into the data exploration and visualization areas of the data analysis life cycle. You will make use of popular Scala libraries like Saddle, Breeze, Vegas, and PredictionIO for processing your datasets. You will learn statistical methods for deriving meaningful insights from data. You will also learn to create applications for Apache Spark 2.x on complex data analysis, in real-time. You will discover traditional machine learning techniques for doing data analysis. Furthermore, you will also be introduced to neural networks and deep learning from a data analysis standpoint. By the end of this book, you will be capable of handling large sets of structured and unstructured data, perform exploratory analysis, and building efficient Scala applications for discovering and delivering insights
Table of Contents (14 chapters)
Free Chapter
Section 1: Scala and Data Analysis Life Cycle
Section 2: Advanced Data Analysis and Machine Learning
Section 3: Real-Time Data Analysis and Scalability

Streaming a k-means clustering algorithm using Spark

The k-means algorithm is an unsupervised machine learning (ML) clustering algorithm. The objective of this algorithm is to build k centers around which data points are centered, thereby forming k clusters. The most common implementation of this algorithm is generally done using batch-oriented processing. Streaming-based clustering algorithms are also available for this, with the following properties:

  • The k clusters are built using initial data
  • As new data arrives in minibatches, existing k clusters are updated to compute new k clusters
  • It also possible to control the decay or decrease in the significance of older data

At a high level, the preceding steps are quite similar to the word count problem that we solved using the streaming solution. The goal of the k-means algorithm is to partition the data into k clusters. If the...