rongpenl / data-scientists-guide-apache-spark

Best practices of using Spark for practicing data scientists in the context of a data scientist’s standard workflow.

The Data Scientist's Guide to Apache Spark

This repo contains notebook exercises for a workshop teaching the best practices of using Spark for practicing data scientists in the context of a data scientist’s standard workflow. By leveraging Spark’s APIs for Python and R to present practical applications, the technology will be much more accessible by decreasing the barrier to entry.

Materials

For the workshop (and after) we will use a Discord chatroom to keep the conversation going: https://discord.gg/avj79xZ

And/or please do not hesitate to reach out to me directly via email at [email protected] or over twitter @memoryphoneme

The presentation can be found on Slideshare here.

Prerequisites

Prior experience with Python and the scientific Python stack is beneficial. Also knowledge of data science models and applications is preferred. This will not be an introduction to Machine Learning or Data Science, but rather a course for people proficient in these methods on a small scale to understand how to apply that knowledge in a distributed setting with Spark.

Setup

• Local Installation
• Cluster Deployment

SparkR with a Notebook

1. Install IRKernel

install.packages(c('rzmq','repr','IRkernel','IRdisplay'), repos = c('http://irkernel.github.io/', getOption('repos')))

IRkernel::installspec()

2. Set environment variables:

# Example: Set this to where Spark is installed
Sys.setenv(SPARK_HOME="/Users/[username]/spark")

# This line loads SparkR from the installed directory
.libPaths(c(file.path(Sys.getenv("SPARK_HOME"), "R", "lib"), .libPaths()))

# if these two lines work, you are all set
library(SparkR)
sc <- sparkR.init(master="local")

Data

link = 'http://hopelessoptimism.com/static/data/airline-data'

The notebooks use a few datasets. For the DonorsChoose data, you can read the documentation here and download a zip (~0.5 gb) from: http://hopelessoptimism.com/static/data/donors_choose.zip

IPython Console Help

Q: How can I find out all the methods that are available on DataFrame?

• In the IPython console type sales.[TAB]

• Autocomplete will show you all the methods that are available.

• To find more information about a specific method, say .cov type help(sales.cov)

• This will display the API documentation for that method.

Spark Documentation

Q: How can I find out more about Spark's Python API, MLlib, GraphX, Spark Streaming, deploying Spark to EC2?

• Go to https://spark.apache.org/docs/latest

• Navigate using tabs to the following areas in particular.

• Programming Guide > Quick Start, Spark Programming Guide, Spark Streaming, DataFrames and SQL, MLlib, GraphX, SparkR.

• Deploying > Overview, Submitting Applications, Spark Standalone, YARN, Amazon EC2.

• More > Configuration, Monitoring, Tuning Guide.

References

Setup

• Spark on Windows 7

History of Computing

• Why CPUs aren't getting any faster
• Hadoop: A brief History
• The State of Spark: And where we are going next
• https://blogs.apache.org/foundation/entry/the_apache_software_foundation_announces50

Original Papers

• Matei's Thesis
• Original (RDD) Paper

Data Science with Spark

• http://blog.cloudera.com/blog/2015/07/how-to-do-data-quality-checks-using-apache-spark-dataframes/

Distributed Computing

• Distributed Systems for Fun and Profit
• Resilience Engineering: Learning to Embrace Failure
• Chaos Monkey

Spark Internals

• PySpark Internals
• A deeper understanding of Spark's internals

Spark Performance

• Tuning and Debugging in Apache Spark
• reduceByKey vs groupByKey
• Advanced Spark
• What's the difference between cache() and persist()
• Monitoring and Instrumentation

Spark Deployment

• A Tale of two clusters: Mesos vs. YARN

Plotly + Spark

• https://plot.ly/ipython-notebooks/apache-spark/
• https://plot.ly/python/ipython-notebooks/
• https://plot.ly/python/matplotlib-to-plotly-tutorial/#6.1-Matplotlib-to-Plotly-conversion-basics

word2Vec

The word2vec tool takes a text corpus as input and produces the word vectors as output. It first constructs a vocabulary from the training text data and then learns vector representation of words. The resulting word vector file can be used as features in many natural language processing and machine learning applications.

Theory/Application

• Efficient Estimation of Word Representations in Vector Space
• Distributed Representations of Words and Phrases and their Compositionality
• Distributed Representations of Sentences and Documents
• deeplearning4j tutorial (with applications)
• Modern Methods for Sentiment Analysis
• word2vec: an introduction

Tools

• Spark MLlib (Python, Scala)
• deeplearning4j (Java)
• gensim: word2vec and doc2vec (Python)

Books on Spark

• Learning Spark: Lightning-Fast Big Data Analytics
  By Holden Karau, Andy Konwinski, Patrick Wendell, Matei Zaharia
  Publisher: O'Reilly Media, June 2014
  http://shop.oreilly.com/product/0636920028512.do
  Introduction to Spark APIs and underlying concepts.

• Spark Knowledge Base
  By Databricks, Vida Ha, Pat McDonough
  Publisher: Databricks
  http://databricks.gitbooks.io/databricks-spark-knowledge-base
  Spark tips, tricks, and recipes.

• Spark Reference Applications
  By Databricks, Vida Ha, Pat McDonough
  Publisher: Databricks
  http://databricks.gitbooks.io/databricks-spark-reference-applications
  Best practices for large-scale Spark application architecture. Topics include import, export, machine learning, streaming.

Learning Scala

• Scala for the Impatient
  by Cay S. Horstmann
  Publisher: Addison-Wesley Professional, March 2012
  http://www.amazon.com/Scala-Impatient-Cay-S-Horstmann/dp/0321774094
  Concise, to the point, and contains good practical tips on using Scala.

Video Tutorials

• Spark Internals
  By Matei Zaharia (Databricks)
  https://www.youtube.com/watch?v=49Hr5xZyTEA

• Spark on YARN
  By Sandy Ryza (Cloudera)
  https://www.youtube.com/watch?v=N6pJhxCPe-Y

• Spark Programming
  By Pat McDonough (Databricks)
  https://www.youtube.com/watch?v=mHF3UPqLOL8

Community

• Community
  https://spark.apache.org/community.html
  Spark's community page lists meetups, mailing-lists, and upcoming Spark conferences.

• Meetups
  http://spark.meetup.com/
  Spark has meetups in the Bay Area, NYC, Seattle, and most major cities around the world.

• Mailing Lists
  https://spark.apache.org/community.html
  The user mailing list covers issues and best practices around using Spark. The dev mailing list is for people who want to contribute to Spark.