Data Science with Python

Python Integration with Hadoop MapReduce and Spark

 Learning Objectives

By the end of this lesson, you will be able to:

Explain why Python should be integrated with Hadoop

Outline the ecosystem and architecture of Hadoop

Explain the functioning of MapReduce

Discuss Apache Spark functions and their benefits

Write Python programs for Hadoop operations

 Quick Recap: Need for Real-Time Analytics

We have seen how big data is generated and understood that to extract insights, proper
analysis of data is more important than its size.
 Quick Recap: Need for Real-Time Analytics

Real-time analytics is the rage right now because it helps extract information
from different data sources almost instantly.
 Quick Recap: Need for Real-Time Analytics

Demographic Orders Products Visits/Channels Social Media Customer Support

s Sentiments and Surveys

Real-Time
Analytics

Higher Conversion Targeted Marketing Better Customer Increased

Satisfaction Sales
 Quick Recap: Why Python
Data Scientists all over the world prefer Python because it is an easy-to-use language that has diverse
libraries required for analytics.

Acquire

Wrangle

Explore

Model

Data Science
Visualize
Bokeh
 Disparity in Programming Languages
However, Big Data can only be accessed through Hadoop which is completely developed and implemented in
Java. Also, analytics platforms are coded in different programming languages.

Hadoop
Infrastructure
(HDFS)

Python

Spark Big Data

Analytics
Platform
Data Science

Big Data Platform

 Integrating Python with Hadoop
As Python is a Data Scientist’s first language of choice, both Hadoop and Spark provide Python APIs that
allow easy access to the Big Data platform.

Hadoop
Infrastructure
(HDFS)

Python Python
APIs
Spark Big Data
Analytics
Platform
Data Science

Big Data Platform

 Need for Big Data Solutions in Python
There are several reasons for creating Big Data solutions in Python.

Multiple Data Processing

Frameworks

Multiple
Programming
Languages

Data Scientist Python APIs

Multiple Big Data
Vendors
 Hadoop: Core Components

Hadoop

HDFS MapReduce
(Hadoop Distributed File System)

• MapReduce is a data processing framework to

• It is responsible for storing data on a cluster
process data on the cluster
• Data is split into blocks and distributed across
• Two consecutive phases: Map and Reduce
multiple nodes in a cluster
• Each map task operates on discrete portions of
• Each block is replicated multiple times
data
o Default is 3 times
• After map, reduce works on the intermediate
o Replicas are stored on different nodes
data distributed on nodes
 Hadoop: System Architecture

This example illustrates the Hadoop system architecture and the ways to store data in a cluster.

Data
Sources Name node

Large file

Data nodes
(Hadoop cluster)

File blocks
(64MB or 128MB)
Secondary name
node
 MapReduce
The second core component of Hadoop is MapReduce, the primary framework of the HDFS architecture.

Input
HDFS

sort Output
HDFS
Split 0 map copy
merge
HDFS
reduc part 0
replicatio
e
n

Split 1 map

HDFS
reduce part 1
replicatio
n
Split 2 map
 MapReduce: Mapper and Reducer
Let us discuss the MapReduce functions, mapper and reducer, in detail.

Mapper Reducer
• Mappers run locally on the data • All intermediate values for a given
nodes to avoid the network traffic. intermediate key are combined together
• Multiple mappers run in parallel into a list and given to a reducer.
processing a portion of the input • This step is known as shuffle and sort.
data. • The reducer outputs either zero or
• The mapper reads data in the form of more final key-value pairs. These
key-value pairs. are written to HDFS.
• If the mapper write generates an output, it is written
in the form of key-value pairs.
 Hadoop Streaming: Python API for Hadoop

Hadoop Streaming acts like a bridge between your Python code and the Java-based HDFS, and lets
you seamlessly access Hadoop clusters and execute MapReduce tasks.

Hadoop
Streaming

Data Scientist Python API

 Mapper in Python

Python supports map and reduce operations:

Suppose you have list of numbers you want

to square = [1, 2, 3, 4, 5, 6 ]

Square function is written as follows:

def square(num):
return num * num

You can square this list using the following code:

squared_nums = map(square, numbers)

Output would be:

[1, 4, 9, 16, 25, 36]
 Reducer in Python

Suppose you want to sum the squared numbers:

[1, 4, 9, 16, 25, 36]

Use the sum function to add two numbers

def sum(a, b ):
return a + b

You can now sum the numbers using the reduce function
Import functools as f
sum_squared = f. reduce(sum, a)

Output would be:

[91]
 Setting Up Cloudera QuickStart Virtual Machine

Cloudera provides enterprise-ready Hadoop Big Data platform which supports Python as well.
To set up the Cloudera Hadoop environment, visit the Cloudera link:

http://www.cloudera.com/downloads/quickstart_vms/5-7.html

Cloudera recommends that you use 7-Zip to extract these files. To download and install it, visit the link:
http://www.7-zip.org/
 Cloudera QuickStart VM: Prerequisites

• These 64-bit VMs require a 64-bit host OS and a virtualization product that can support a 64-bit guest OS.
• To use a VMware VM, you must use a player compatible with WorkStation 8.x or higher:
• Player 4.x or higher
• Fusion 4.x or higher
• Older versions of WorkStation can be used to create a new VM using the same virtual disk (VMDK file), but
some features in VMware tools are not available.
• The amount of RAM required varies by the run-time option you choose
 Launching VMware Image

To launch the VMware, visit the VMware link:

https://www.vmware.com/products/player/pl https://www.vmware.com/products/fusion/fu
ayerpro-evaluation.html sion-evaluation.html
QuickStart VMware Image
Launch VMware player with Cloudera VM

Launch Terminal

Account:
username: cloudera
password: cloudera
 QuickStart VM Terminal

Step 1 Step 2

Unix command :
• pwd to verify present working directory
• ls -lrt to list files and directories
Using Hadoop Streaming for Calculating Word Count

Demonstrate how to create a MapReduce program and use Hadoop Streaming to

determine the word count of a document
 MapReduce Uses Disk I/O Operations

HDFS HDFS HDFS HDFS

read write read write
Iteration 1 Iteration 2

Input

Query 1 Result 1

Query 2 Result 2

Input Query 3
Result 3
 Apache Spark Uses In-Memory Instead of Disk I/O

HDFS
read
Iteration 1 Iteration 2
Memory Memory (RAM)
(RAM)
Input

Query 1 Result 1

Query 2
Result 2

Input Query 3
Result 3
Distributed memory

10-100 X faster than network and disk

 Hardware Requirements for MapReduce and Spark

Hard Drives Hard Drives

CPUs
CPUs Memory
MapReduce Spark
 Apache Spark Resilient Distributed Systems (RDD)

Some basic concepts about Resilient Distributed Datasets (RDD) are:

• The main programming approach of Spark is RDD.

• They are fault-tolerant collections of objects spread across a cluster that you can
operate on in parallel. They can automatically recover from machine failure.
• You can create an RDD either by copying the elements from an existing collection or by
referencing a dataset stored externally.
• RDDs support two types of operations: transformations and actions.
• Transformations use an existing dataset to create a new one.
• Example: Map, filter, join
• Actions compute on the dataset and return the value to the driver program.
• Example: Reduce, count, collect, save

If the available memory is insufficient, then the data is written to disk.

 Advantages of Spark

Faster: 10 to 100 times faster than Hadoop MapReduce

• Simple data processing framework

Simplified: • Interactive APIs for Python for faster application
development

Efficient: Has multiple tools for complex analytics operations

Integrated: Can be easily integrated with existing Hadoop

infrastructure
 PySpark: Python API for Spark

PySpark is the Spark Python API which enables data scientists to access Spark programming model.

PySpark

Data Python API

Scientist
 PySpark: RDD Transformations and Actions

Transformation Action

Transformation Description Action Description

Returns RDD, formed by

map() passing data element of the Returns all elements of the
collect()
source dataset as an array

Returns RDD based on Returns the number of

filter() count()
selection elements present in the dataset
Maps items present in the Returns the first element in the
flatMap() first()
dataset and returns sequence dataset
Returns key value pairs where Returns number of elements (n)
reduceByKey() values for which each key is take(n) as specified by the number in
aggregated by value the parenthesis

SparkContext or SC is the entry point to spark for the spark application

 Spark Tools

Spark MLlib
Spark GraphX
SQL (machine
Streaming (graph)
learning)

Spark

Interactive Python APIs

 Setting Up Apache Spark

To set up the Apache Spark environment, access the link:

http://spark.apache.org/downloads.html

Please use 7-Zip to extract these files.

Setting Up Environmental Variable of Apache Spark

[installed directory]\spark-1.6.1-bin-
hadoop2.4\spark-1.6.1-bin-hadoop2.4

[installed directory] \spark-1.6.1-bin-

hadoop2.4\spark-1.6.1-bin-
hadoop2.4\bin
Integrating Jupyter Notebook with Apache Spark

Setup the
pyspark
notebook
specific
variables

Run the pyspark

command

Check SparkContext
Using PySpark to Determine Word Count

Demonstrate how to use the Jupyter integrated PySpark API to determine the word count
of a given dataset
Word Count

Determine the word count of the given Amazon dataset:

• Create a MapReduce program to determine the word count of the Amazon
dataset
• Submit the MapReduce task to HDFS and run it
• Verify the output
Word Count

Use the given dataset to count and display all the airports based in New York using
PySpark. Perform the following steps:
• View all the airports listed in the dataset
• View only the first 10 records
• Filter the data for all airports located in New York
• Clean up the dataset if required
Knowledge Check
Knowledge
Check
What are the core components of Hadoop? Select all that apply.
1

a. MapReduce

b. HDFS

c. Spark

d. RDD
 Knowledge
Check
What are the core components of Hadoop? Select all that apply.
1

a. MapReduce

b. HDFS

c. Spark

d. RDD

The correct answer a and b

is
MapReduce and HDFS are the core components of Hadoop.
Knowledge
Check
MapReduce is a data processing framework which gets executed _____.
2

a. at DataNode

b. at NameNode

c. on client side

d. in memory
 Knowledge
Check
MapReduce is a data processing framework which gets executed _____.
2

a. at DataNode

b. at NameNode

c. on client side

d. in memory

The correct answer a

is
The MapReduce program is executed at the data node and the output is written to the disk.
Knowledge
Check Which of the following functions is responsible for consolidating the results produced by each
of the Map() functions/tasks?
3

a. Reducer

b. Mapper

c. Partitioner

d. All of the above

 Knowledge
Check Which of the following functions is responsible for consolidating the results produced by each
of the Map() functions/tasks?
3

a. Reducer

b. Mapper

c. Partitioner

d. All of the above

The correct answer a

is
Reducer combines or aggregates results produced by mappers.
Knowledge
Check
What transforms input key-value pairs to a set of intermediate key-value pairs?
4

a. Mapper

b. Reducer

c. Combiner

d. Partitioner
 Knowledge
Check
What transforms input key-value pairs to a set of intermediate key-value pairs?
4

a. Mapper

b. Reducer

c. Combiner

d. Partitioner

The correct answer a

is
Mapper processes input data to intermediate key-value pairs which are in turn processed by reducers.
 Key Takeaways

You are now able to:

Explain why Python should be integrated with Hadoop

Outline the ecosystem and architecture of Hadoop

Explain the functioning of MapReduce

Discuss Apache Spark functions and their benefits

Write Python programs for Hadoop operations

 Stock Market Data Analysis

Import the financial data using Yahoo data reader for the following
companies:
• Yahoo
• Apple
• Amazon
• Microsoft
• Google

Perform fundamental data analysis

• Fetch the previous year’s data
• View the values of Apple’s stock
• Display the plot of closing price
• Display the stock trade by volume
• Plot all companies’ data together for closing prices

© Copyright 2015, Simplilearn. All rights reserved.

 Stock Market Data Analysis

Perform Daily Return Analysis and show the relationship between

different stocks
• Plot the percentage change plot for Apple’s stock
• Show a joint plot for Apple and Google
• Use PairPlot to show the correlation between all the stocks

Perform risk analysis

© Copyright 2015, Simplilearn. All rights reserved.

 Titanic Data Set Analysis

On April 15, 1912, the Titanic sank after colliding with an iceberg, killing
1502 out of 2224 passengers and crew. This tragedy shocked the world
and led to better safety regulations for ships. Here, we ask you to
perform the analysis through the exploratory data analysis technique.
In particular, we want you to apply the tools of machine learning to
predict the survived passengers.

© Copyright 2015, Simplilearn. All rights reserved.

 Titanic Data Set Analysis

The details of these projects and their scope are below:

Data acquisition of the Titanic dataset

• Train dataset
• Test dataset

Perform the Exploratory Data Analysis (EDA) for train dataset

• Passengers age distribution
• Passengers survival by age
• Passengers survival breakdown
• Passengers class distribution
• Passengers embarkation by locations

© Copyright 2015, Simplilearn. All rights reserved.

 Titanic Data Set Analysis

Perform machine learning to train the model and

• Create user defined function to load train data set
• Create user defined function to load test data set
• Create machine model
• Train the machine
• Predict whether a passenger survived the tragedy or not
• Persist the mode for future re-use

© Copyright 2015, Simplilearn. All rights reserved.

 Thank You

© Copyright 2015, Simplilearn. All rights reserved.