CSCI E-84

A Practical Approach to Data

Science
Ramon A. Mata-Toledo, Ph.D.
Professor of Computer Science
Harvard University
Unit 1 - Lecture 1
January, Wednesday 27, 2016

Required Textbooks

Main Objectives of the Course

This course is an introduction to the field of data science
and its applicability to the business world.
At the end of the semester you should be able to:
Understand how data science fits in an organization and
gather, select, and model large amounts of data.
Identify the appropriate data and the methods that will
allow you to extract knowledge from this data using the
programming language R.

What This Course is NOT is an in-depth

calculus based statistic research
environment .

What is this course all about? It IS is a practical hands-on

approach to understanding the methodology and tools of
Data Science from a business prospective. Some of the
technical we will briefly consider are:

Statistics
Database Querying
SQL
Data Warehousing
Regression Analysis
Explanatory versus Predictive Modeling

Brief History of Data Science/Big Data

The term data science can be considered the result of three
main factors:
Evolution of data processing methods
Internet particularly the Web 2.0
Technological Advancements in computer processing
speed/storage and development of algorithms for
extracting useful information and knowledge from big
data

Brief History of Data Science/Big Data

(continuation)
However, Already seventy years ago we encounter the
first attempts to quantify the growth rate in the volume of
data or what has popularly been known as the information
explosion (a term first used in 1941, according to
the Oxford English Dictionary). "

Selected Milestones in the History of

Data Science/Big Data
Theoretical/Academic:
(Source: http://www.forbes.com/sites/gilpress/2013/05/09/a-very-short-history-of-big-data/)

November 1967 - B. A. Marron and P. A. D. de Maine publish Automatic

data compression in the Communications of the ACM, stating that The
information explosion noted in recent years makes it essential that storage
requirements for all information be kept to a minimum.

April 1980 - I.A. Tjomsland gives a talk titled Where Do We Go From Here? at
the Fourth IEEE Symposium on Mass Storage Systems, in which he says Those
associated with storage devices long ago realized that Parkinsons First Law may
be paraphrased to describe our industry Data expands to fill the space
available. I believe that large amounts of data are being retained because users
have no way of identifying obsolete data; the penalties for storing obsolete data
are less apparent than are the penalties for discarding potentially useful data.

July 1986 - Hal B. Becker publishes Can users really absorb data at todays
rates? Tomorrows? in Data Communications. Becker estimates that the
recoding density achieved by Gutenberg was approximately 500 symbols
(characters) per cubic inch500 times the density of [4,000 B.C. Sumerian] clay
tablets. By the year 2000, semiconductor random access memory should be storing
1.25X10^11 bytes per cubic inch.

1996 - Digital storage becomes more cost-effective for storing data than paper
according to R.J.T. Morris and B.J. Truskowski, in The Evolution of Storage
Systems IBM Systems Journal, July 1, 2003.

1997 - Michael Lesk publishes How much information is there in the world?
Lesk concludes that There may be a few thousand petabytes of information all
told; and the production of tape and disk will reach that level by the year 2000. So
in only a few years, (a) we will be able [to] save everythingno information will
have to be thrown out, and (b) the typical piece of information will never be
looked at by a human being.

August 1999 - Steve Bryson, David Kenwright, Michael Cox, David Ellsworth, and
Robert Haimes publish Visually exploring gigabyte data sets in real time in
the Communications of the ACM. It is the first CACM article to use the term Big Data
(the title of one of the articles sections is Big Data for Scientific Visualization). The
article opens with the following statement: Very powerful computers are a blessing to
many fields of inquiry. They are also a curse; fast computations spew out massive
amounts of data. Where megabyte data sets were once considered large, we now find
data sets from individual simulations in the 300GB range. But understanding the data
resulting from high-end computations is a significant endeavor. As more than one
scientist has put it, it is just plain difficult to look at all the numbers. And as Richard W.
Hamming, mathematician and pioneer computer scientist, pointed out, the purpose of
computing is insight, not numbers.
October 2000 - Peter Lyman and Hal R. Varian at UC Berkeley publish How Much
Information? It is the first comprehensive study to quantify, in computer storage terms,
the total amount of new and original information (not counting copies) created in the
world annually and stored in four physical media: paper, film, optical (CDs and DVDs),
and magnetic.

The Google White Papers

Foreshadowing of the data explosion

Developed 2 white papers

Map Reduce
Google File System (GFS)

Basic precept was based on a distributed file system

Clusters of smaller, cheaper computers

Selected Software for Handling Big Data

SPSS is a widely used program for statistical analysis in social science. It is also used
by industry and academic researchers. The original SPSS manual (Nie, Bent & Hull,
1970) has been described as one of "sociology's most influential books" for allowing
ordinary researchers to do their own statistical analysis. In addition to statistical
analysis, data management (case selection, file reshaping, creating derived data)
and data documentation (a metadata dictionary was stored in the datafile) are
features of the base software. SPSS was acquired by IBM in 2009.
SAS was developed at North Carolina State University from 1966 until 1976, when
SAS Institute was incorporated. SAS was further developed in the 1980s and 1990s
with the addition of new statistical procedures, additional components and the
introduction of JMP. A point-and-click interface was added in version 9 in 2004. A
social media analytics product was added in 2010.

R is a programming language and software environment for statistical

computing and graphics supported by the R Foundation for Statistical Computing.
R was created, in 1997, by Ross Ihaka and Robert Gentleman at the University of
Auckland, New Zealand, and is currently developed by the R Development Core
Team, of which Chambers is a member.
R is named partly after the first names of the first two R authors and partly as a
play on the name of S programming language.
Source:
(https://www.stat.auckland.ac.nz/~ihaka/downloads/Interface98.pdf).

Software Tools
MongoDB is a cross-platform document-oriented database developed by
MongoDB Inc, in 2007. Classified as a NoSQL database, MongoDB eschews the
traditional table-based relational database structure in favor of JSON-like
documents with dynamic schemas (MongoDB calls the format BSON), making
the integration of data in certain types of applications easier and faster.
MongoDB is free and open-source software.

Software Tools - Continuation

Hadoop
Open source from Apache [written in Java]
Implementation of the Google White Papers
Now associated with a collection of technologies

Design Choice
Hadoop Example
Very good at storing files
Optimizes use of cheap resources - no RAID
needed here
Provides data redundancy
Good at sequential reads
Not so good at high speed random reads
Cannot update a file must replace

Hadoop Ecosystem other Technologies

Yet Another Resource Negotiator (YARN)

Scoop imports tables from RDBMS

Flume deals with event data like web logs
Hive imposes metadata over flat data so SQL code can be used

Impala high speed analytic using distributed queries

Hbase NOSQL db for Hadoop stored data
Mahout machine learning algorithms

Oozie workflow manager

Zookeeper configuration management

Big Data
The processing of massive
data sets that facilitate
real-time data driven decision-making
Digital data grows by
2.5 quintillion (1018)
bytes every day

Units of Information storage

The smallest addressable unit in a computers memory is the
byte. A byte is equal to 8 consecutive binary digits or bits.

In the physical sciences a Kilo (K) stand for 1000 units. For
example: 1 Km = 1000 meters (m)
1 Kg = 1000 grams (g)
However, in the computer field when we refer to Kilobyte or
Kb the K = 1024 bytes.

Units of Information storage

(continuation)

Larger information units and their names are shown in the figure below.
Source (https://en.wikipedia.org/wiki/Units_of_information)

1 ZB is equivalent to 1 followed by 21 zeroes.

1 YB is equivalent to 1 followed by 24 zeroes.

The problem with data..

over abundance, fragmentation,
erroneous, heterogeneity, duplication,
untrustworthy, unstructured.

What makes it Big Data?

Volume
Velocity
Variety
Veracity
Validity
Value
Volatility

Diya Soubra (2012), The 3Vs that Define Big Data

Volume > 2.8

21)
(10 ZB

Digital universe will reach 40 zettabytes (ZB) by 2020

57 times all grains of sand on the beaches

= weight of 424 Nimitz aircraft carriers
Facebook (2012) processed 2.5 million pieces of content
each day (= 500+ terabytes of data daily)
Amazon sold 958,333 items every hour (on Cyber Monday)
in 2016. (If you do the math, thats how many in a day???)

Velocity Data in Motion

Data Delivery Streaming Data

Increasing rate of data flow into the

organization. It considers incoming
and outgoing flow of data. The
faster we can capture, analyze, and
transform data into useful
information the better.

Results delivery Real time, Near-real time, Periodic

Disease Outbreak
Facial Recognition
Traffic

Variety ALL types of data

Big data comes from EVERYWHERE
It is not a relational database with
billions of structured rows its a
mix of structured and multistructured data.

While we are talking about Data

Veracity the 4th V of Big Data
Dirty data is a tough issue
Uncertain data cant be cleaned
Errors can snowball fast with big data
Examples:
GPS signals bouncing around buildings in NYC
Weather conditions sun,clouds, rain icon

Validity
Validity refers to the issue that the data being stored and
mined is clean and meaningful to the problems or
decisions that need to be made.

Value and Volatility

Value refers usefulness and relevancy of the information
extracted from the data as opposed to the practice of
collecting data for archival or regulatory purposes.
Volatility addresses the issue of how long is the data
valid and for how long should it be stored. Some analysis
is required to determine when, in the future, the data is
no longer relevant.

In Summary

Uses of Big Data - Current

Customer-Centric
Personalize the experience

Customer Recommendations
Streaming Routing

Online ad targeting

Predictive Analytics
Manage the risk

Credit Card Fraud

Crop Forecasts

Uses of Big Data - Future

Self Quantification

New material development

(Materials Genome Project)
Pay as You Go

What is Data Science?

What does it take?

Is there anything else?

Tools in
Data
Science

Some Order in the Buzz

I/O The Real Issue

Processors arent the
problem, its getting
data off the hard disk
and for those hopeful that Solid State drives
would be a solution they wont be replacing hard
drives anytime soon

Algorithms
What it really is all about
In Big Data, common association is with
Predictive Algorithms

Room for new ideas but some believe

we have enough of these for now
Others say what you need is more data,
simpler algorithms

What types of algorithms?

Sorting

Classifying

Searching

Predictive
Learning via Rule
Fit Ensembles

Streaming
Filtering

Deterministic
behavior
algorithms

Summary
From scientific discovery to business intelligence,
"Big Data" is changing our world
Big Data permeates most (all?) areas of computer
science
Opening the doors to lots of opportunities in the
computing sciences
Its not just for Data Scientists.

Links to more info about material presented in this lecture:

IDC (2012) Digital Universe in 2020 - http://www.emc.com/collateral/analyst-reports/idc-the-digital-universe-in-2020.pdf
Diya Soubra (2012) The 3Vs that define Big Data - http://www.datasciencecentral.com/forum/topics/the-3vs-that-define-big-data
Self Quantification - http://www.topcoder.com/blog/big-data-mobile-sensors-visualization-gamification-quantified-self/
Materials Genome Project - http://www.theverge.com/2013/9/26/4766486/materials-genome-initiative-mit-and-harvard
Pay as You Go - http://www.businessweek.com/articles/2012-10-15/pay-as-you-drive-insurance-big-brother-needs-a-makeover
Montgomery County Traffic Cameras - http://www6.montgomerycountymd.gov/tmctmpl.asp
Trending Now: Using Social Media to Predict and Track Disease Outbreaks http://www.ncbi.nlm.nih.gov/pmc/articles/PMC3261963/
Health Map - http://healthmap.org/en/
Manhunt Boston Bombers. Aired May 29, 2013 on PBS - Nova Program on Facial Recognition http://www.pbs.org/wgbh/nova/tech/manhunt-boston-bombers.html
Netflix and Big Data - http://technologyadvice.com/wp-content/themes/newta/how-netflix-is-using-big-data-to-get-people-hooked-on-itsoriginal-programming.html#.Ul1WIBAlgnZ
Data Never Sleeps - http://www.domo.com/learn/infographic-data-never-sleeps
Google Research Publication: The Google File System - http://research.google.com/archive/gfs.html
Google Research Publication: MapReduce - http://research.google.com/archive/mapreduce.html

Links to more info about material presented in this talk:

Sorting the World: Google Invents New Way to Manage Data - http://www.wired.com/science/discoveries/magazine/1607/pb_sorting
Hadoop - http://hadoop.apache.org/
What lies at the core of Hadoop? http://blog.enablecloud.com/2012/06/what-lies-at-core-of-hadoop.html
Distributed Average Consensus with Least-Mean-Square Deviation http://www.stanford.edu/~boyd/papers/pdf/lmsc_mtns06.pdf
HDFS in Cartoon https://docs.google.com/file/d/0Bzw6KHOtbT4MmRkZWJjYzEtYjI3Ni00NTFjLWE0OGItYTU5OGMxYjc0N2M1/edit?usp=drive_web&pli=1
Big-Data Computing: Creating revolutionary breakthroughs in commerce, science, and society - Computer Research
Association CCC led paper - http://www.cra.org/ccc/resources/ccc-led-white-papers
How Quantum Computers and Machine Learning Will Revolutionize Big Data Quanta Magazine http://www.wired.com/wiredscience/2013/10/computers-big-data/
Predictive Apps http://www.information-management.com/blogs/application-developers-ignore-big-data-at-your-own-peril10024904-1.html
Real World Use of Big Data Ford Focus - http://www.stthomas.edu/gradsoftware/files/BigData_RealWorldUse.pdf
Eatery Massive Health Experiment - https://eatery.massivehealth.com/
Waze www.waze.com
CS2013 Ironman v1.0 draft - http://ai.stanford.edu/users/sahami/CS2013/