What is Data Science?

Data, Databases, and the Extraction of Knowledge

Rene T., @becomingdatasci, November 2014

Lets start with: What is Data?

http://upload.wikimedia.org/wikipedia/commons/f/f0/DARPA
_Big_Data.jpg

http://fc01.deviantart.net/fs71/i/2012/326/3/4/cute_dog_by_tho
masmeadows345-d5lsah9.jpg

https://encryptedtbn2.gstatic.com/images?q=tbn:ANd9GcS9dKu3_Tzi-sWWyAqee5y0EhuvoIZNSya_rAKnuBBd0JYxPX7pw

http://www.freefoto.com/images/1351/06/1351_06_2---Books-Shakespeare-and-Company-Bookstore--The-Latin-Quarter-Paris_web.jpg

http://upload.wikimedia.org/wikipedia/commons/9/96/Bill_Nye
,_Barack_Obama_and_Neil_deGrasse_Tyson_selfie_2014.jpg

http://upload.wikimedia.org/wikipedia/commons/e/e4/Gr
een_Bank_100m_diameter_Radio_Telescope.jpg

https://c2.staticflickr.com/4/3273/3017878633_65beb1c7d6.jpg

https://c1.staticflickr.com/1/2/1349370_07
03fce74c.jpg

Around 100 hours of video are uploaded to YouTube every minute

it would take about 15 years to watch every video uploaded in one day

AT&T is thought to hold the worlds largest volume of data in one

unique database its phone records database is 312 terabytes in size,
and contains almost 2 trillion rows.

Every minute we send 204,000,000 emails, generate 1,800,000 Facebook

likes, send 278,000 Tweets, and up-load 200,000 photos to Facebook

570 new websites spring into existence every minute of every day.
http://smartdatacollective.com/bernardmarr/277731/big-data-25-facts-everyone-needs-know

http://pixabay.com/static/uploads/photo/2014/03/13/01/12/datacen
ter-286386_640.jpg

https://c2.staticflickr.com/2/1296/533233247_b6baa30fdb_z.jpg?zz=1

Video clip:
http://youtu.be/PBx7rgqeGG8?t=2m

https://c1.staticflickr.com/3/2300/2596366618_2d6cb01735.jpg

http://upload.wiki
media.org/wikipedi
a/commons/9/90/Ke
ncf0618FacebookNe
twork.jpg

http://upload.wikimedia.org/wikipedia/commons/b/bf/USDA_Hardine
ss_zone_map.jpg

http://upload.wikimedia.org/wikipedia/commons/1/1c/CMS_Higgs-event.jpg

What is a database?

Database
[dey-tuh-beys]
noun
A comprehensive collection of related data
organized for convenient access, generally in
a computer.

-dictionary.com

Types of Databases

http://www.oaddo.org

Databases You Use

Pretty much every website you interact with

Social Media
Banking
File Sharing
Search Engines

Online Shopping
Course Registration/Canvas
Travel
Etc. etc. etc..

You broadcast/generate data everywhere you go

Cell phones
Purchases
Driving (GPS)
Streaming music

Email
Posting status updates
Attending events
Etc. etc. etc..

https://www.google.com/maps/@38.8905569,-77.1721577,13z/data=!5m1!1e1

http://upload.wikimedia.org/wikipedia/commons/6/69/Netflix_logo.svg

https://c2.staticflickr.com/4/3324/3507973704_563846fe14_z.jpg?zz=1

How is data
collected about you
used to help you?

Who builds these systems?

Data Scientist
Computer Scientist
Data collection systems
Machine Learning
Algorithms
Interface Design
Design/Manage/Query
Databases
Data Aggregation
Data Mining

Mathematician
Statistical Models
Evaluation Metrics
Predictive Analytics
Data Visualizations

Business Person
Domain Expertise
Knowing what
questions to ask
Interpreting results for
business decisions
Presenting outcomes

Examples not a complete definition, and not all

simultaneously necessary skills

Data Science Venn Diagram by Drew Conway

http://static.squarespace.com/static/5150aec6e4b0e340ec52710a/t/51525c33e4b0b3e0d10
f77ab/1364352052403/Data_Science_VD.png?format=750w

From Doing Data Science by Cathy

ONeill & Rachel Schutt
http://www.becomingadatascientist.com/wpcontent/uploads/2014/06/DS_profile.png

http://semanticommunity.info/@api/deki/files/27057/Figure14.png?size=bestfit&width=484&height=541&revision=1

No need to be a unicorn, but do need to know something

about all of these areas, and become expert in some
(Sound familiar, ISAT students?)

Some other names for Data Scientist

Statistician

Pythonista

Data Mining Specialist

Financial Analyst

Biostatistician

Recommendation System

Social Science Researcher

Big Data Analyst

Information Architect

Spatial/GIS Analyst

Artificial Intelligence

Natural Language
Programmer

Computational Physicist

Engineer

Researcher

Neuroscientist

Data Visualization Designer

Data Science jobs pay an

average of $118,000 per year
It is estimated that by 2018, US could have a
shortage of 140,000+ people with advanced
analytical skills & need 1.5M managers/analysts
that can make decisions based on data analysis

Extraction of Knowledge

Also known as knowledge discovery

Goes beyond queries

Data Mining

Business Understanding
Data Understanding
Data Preparation
Modeling

Clustering
Classification
Regression

Evaluation

From Data Science for

Business by Provost & Fawcett

Images from ODU ECE 607 Lecture Slides by Prof. Jiang Li

Video clip: Interview with Neha Kothari, LinkedIN Data Scientist

http://youtu.be/8dxKe5cGHdA?t=17s

Data Science Example

Kaggle competition hosted by UPenn and Mayo Clinic to

detect seizures in intracranial EEG recordings

https://www.kaggle.com/c/seizure-detection

Current detection systems have high false positive rate, resulting

in unnecessary stimulation

Need to rapidly and automatically detect onset of seizure

Data provided

Matrix of EEG sample values

Time duration latency (time before seizure)

Sampling frequency

Channels (electrodes)

Human and Canine Data

Latency only provided in training data because when taking

real-life data, you wont know if or how long until seizure hits
thats what youre trying to predict

This is an important point in predictive analytics!

Competition winner Michael Hills published his method

FFT = Fast Fourier Transform

Correlation Coefficient r
Eigenvalues can think of this as a
scaling factor
Put all these values into a
Random Forest classifier

http://en.wikipedia.org/wiki/Fast_Fourier_transform

Ensemble learning method combines results of many weak decision

trees, turns out to be better classifier than one strong decision tree
Can now train a classifier for each patient

He wrote a computer program to help him experiment & quickly

validate result of each brute force approach, trying every
technique he could find

Determines primary frequencies in

EEG sample

Used the same evaluation technique kaggle competition would use

Line of scikit-learn Python code for training winning submission:

RandomForestClassifier(n_estimators=3000, min_samples_split=1,
bootstrap=False, random_state=0)

Kaggles evaluation method:

Judged on the mean area under the ROC curve (AUC) of two predictions.
Receiver Operating Characteristic = true positive vs false positive.
1)
2)

Predict the probability that a given clip is a seizure.

Predict the probability that the clip is within the first 15 seconds its respective
seizure (the technical term for time into the seizure is "latency").

The competition metric is the mean of these two AUCs:

Michael Hills winning submission scored 0.963

His model will label 963 of every 1000 true seizure clips as seizures
He won $5000 (much less than UPenn/Mayo would have had to pay a
Data Scientist to develop this as an employee or consultant!)
Currently another similar contest posted w/$25,000 prize

My Machine Learning project

Using JMU first-time donor (and non-donor) data from two previous years, could
I classify who was likely to become a donor for the first time during the next year?
Correctly classified 67% of first-time donors, got great feedback from professor,
plan to continue the study for my masters program final project.
You can read all about it on my blog! BecomingADataScientist.com
Code snippet using Random Forest Classifier

Other Examples

Galaxy Classification using Convolutional

Neural Networks
http://benanne.github.io/2014/04/05/galaxy-zoo.html

Choosing Facebook Audience for Content

Promotion using Random Forests
http://citizennet.com/blog/2012/11/10/random-forestsensembles-and-performance-metrics/

Predicting Wine Quality with Principal

Component Analysis
http://fastml.com/predicting-wine-quality/

Readmission Risk Score to decide which

patients to give additional follow-up help at
Mt. Sinai hospital
http://www.technologyreview.com/news/518916/ahospital-takes-its-own-big-data-medicine/

Data Visualization Example

http://labs.strava.com/heatmap/#12/-78.90549/38.44669/blue/bike

http://xkcd.com/1425/

How to get started

Recommended skills to pick up while at JMU

Programming

Any language is good to

start with. Gain core
understanding.

Python or R data analysis

experience a plus

Database design, SQL

Math

Calculus

Linear Algebra

Statistics (2 levels)

Advanced: Optimization /
Linear Programming

Research and Analysis

Science involving data

collection and interpretation

Working with messy real

life data

Business Analytics

Data Mining

Others

Business / Communication

Graphic Design

Take classes on campus or online!

Read, read, read

Doing Data Science by Cathy ONeil* & Rachel Schutt

Data Science for Business by Forster Provost & Tom Fawcett
Data Smart by John Foreman* (uses Excel)
Ill review other books as I read them:
http://www.becomingadatascientist.com/learning/
Blogs & News Feeds (FlowingData.com is a good one to start with)
Twitter look for curated lists of people to follow
https://twitter.com/BecomingDataSci/lists/women-in-datascience/members

*on Twitter and

willing to chat!

Free Online Courses

Python Fundamentals Codecademy http://www.codecademy.com/tracks/python

Machine Learning Coursera / Stanford https://www.coursera.org/course/ml

Data Analyst Nanodegree Udacity https://www.udacity.com/course/nd002

(includes Hadoop mini-course)

Applied Data Mining and Statistical Learning Penn State

https://onlinecourses.science.psu.edu/stat857/

Pretty comprehensive list here: http://www.kdnuggets.com/education/online.html

TED talks on Data

http://www.ted.com/search?q=data

Susan Etlinger* http://www.ted.com/talks/susan_etlinger_what_do_we_do_with_all_this_big_data

Need to spend more time on critical thinking skills[because we have

the] potential to make bad decisions far more quickly, efficiently, and with
far greater impact than we did in the past.

we need to be clear about ..the methodologies that we use, because if I

don't know what questions you asked, I don't know what questions you
didn't ask.

Explore

Volunteer to Analyze Data (DataKind)

Play with public data sets

http://101.datascience.community/2014/10/17/data-sources-for-cool-datascience-projects-part-1-guest-post/

https://www.opensciencedatacloud.org/publicdata/

http://catalog.data.gov/dataset

https://archive.ics.uci.edu/ml/datasets.html?format=&task=clu&att=&area=&nu

mAtt=&numIns=&type=&sort=nameUp&view=table

Data Science Competitions

(Kaggle also has knowledge competitions for learning)

What some of my followers on Twitter wish

they knew about data in college.

Questions?
Renee T.
[contact me via twitter or blog for email]
@becomingdatasci
http://www.becomingadatascientist.com