Data Science in the Real World: Making a Difference

Понравилась презентация – покажи это...

Слайд 0

Data Science in the Real World: Making a Difference Srinath Perera Director Research WSO2, Apache Member (@srinath_perera) srinath@wso2.com StatDay 2015 @ University of Colombo

Слайд 1

Outline Making sense of World’s Data Building Data Systems Changing Dynamics of Data Analysis with Big Data ( Sensor Data) Challenges and Open Problems

Слайд 2

Michael Stonebraker “But then, out of nowhere, some marketing guys started talking about ‘big data, That’s when I realized that I’d been studying this thing for the better part of my academic life.”

Слайд 3

Michael Stonebraker “But then, out of nowhere, some marketing guys started talking about ‘big data, That’s when I realized that I’d been studying this thing for the better part of my academic life.” ACM Turing Award, 2015

Слайд 4

A Day in Your Life Think about a day in your life? What is the best road to take? Would there be any bad weather? How to invest my money? How is my health? There are many decisions that you can do better if only you can access the data and process them. http://www.flickr.com/photos/kcolwell/5512461652/ CC licence

Слайд 5

Слайд 6

What can We do with Data? Optimize (World is inefficient) 30% food wasted farm to plate GE Save 1% initiative (http://goo.gl/eYC0QE ) Trains => 2B/ year US healthcare => 20B/ year Save lives Weather, Disease identification, Personalized treatment Technology advancement Most high tech research are done via simulations

Слайд 7

Building Data Processing Systems

Слайд 8

Data Science Architecture

Слайд 9

Data Processing Technologies Landscape

Слайд 10

Batch Processing Store and process Slow (> 5 minutes for results for a reasonable usecase) Programming model is MapReduce Apache Hadoop Spark Lot of tools built on top Hive Shark for (SQL style queries), Mahout (ML), Giraph (Graph Processing)

Слайд 11

Usecase: Big Data for development Done using CDR data People density noon vs. midnight (red => increased, blue => decreased) Urban Planning People distribution Mobility Waste Management E.g. see http://goo.gl/jPujmM From: http://lirneasia.net/2014/08/what-does-big-data-say-about-sri-lanka/

Слайд 12

Value of some Insights degrade Fast! For some usecases ( e.g. stock markets, traffic, surveillance, patient monitoring) the value of insights degrades very quickly with time. E.g. stock markets and speed of light We need technology that can produce outputs fast Static Queries, but need very fast output (Alerts, Realtime control) Dynamic and Interactive Queries ( Data exploration)

Слайд 13

Complex Event Processing

Слайд 14

Predictive Analytics If we know how to solve a problem, that is if we know a finite set of rules, then we can programs it. For some problems (e.g. Drive a car, character recognition), we do not know a finite fix rule set. Instead of programming, we give lot of examples and ask the computer to learn (often called Machine Learning) Lot of tools R ( Statistical language) Sci-kit learn (Phython) Apache Spark’s MLBase and Apache Mahout (Java)

Слайд 15

Usecase: Predictive Maintenance Idea is to fix the problem before it broke, avoiding expensive downtimes Airplanes, turbines, windmills Construction Equipment Car, Golf carts How Build a model for normal operation and compare deviation Match against known error patterns

Слайд 16

Communicate: Dashboards Idea is to given the “Overall idea” in a glance (e.g. car dashboard) Support for personalization, you can build your own dashboard. Also the entry point for Drill down How to build? Expose data via JSON Build Dashboard via Google Gadget and content via HTML5 + java scripts (Use charting libraries like Vega or D3)

Слайд 17

Communicate: Alerts and Triggers Detecting conditions can be done via Event Processing system ( e.g. CEP) Key is the “Last Mile” Email SMS Push notifications to a UI Pager Trigger physical Alarm

Слайд 18

Case Study: Realtime Soccer Analysis Watch at: https://www.youtube.com/watch?v=nRI6buQ0NOM

Слайд 19

Changing Dynamics

Слайд 20

Large Observational Datasets Stats are easy with designed experiments You got to select a representative set You have a control group You have lot and lot of data and lot and lot of computing power ( compared to what you had) Two reactions!!

Слайд 21

“It is better to be roughly right than precisely wrong.” ? John Keynes In the long run, we are all Dead!!

Слайд 22

Challenges: Causality Correlation does not imply Causality!! ( send a book home example [1]) Causality do repeat experiment with identical test If CAN’T do a randomized test (A/B test) With Big data we cannot do either Option 1: We can act on correlation if we can verify the guess or if correctness is not critical (Start Investigation, Check for a disease, Marketing ) Option 2: We verify correlations using A/B testing or propensity analysis [1] http://www.freakonomics.com/2008/12/10/the-blagojevich-upside/ [2] https://hbr.org/2014/03/when-to-act-on-a-correlation-and-when-not-to/

Слайд 23

Curious Case of Missing Data http://www.fastcodesign.com/1671172/how-a-story-from-world-war-ii-shapes-facebook-today, Pic from http://www.phibetaiota.net/2011/09/defdog-the-importance-of-selection-bias-in-statistics/ WW II, Returned Aircrafts and data on where they were hit? How would you add Armour?

Слайд 24

More Data Beat a Clever Algorithm Observed by large internet companies Also seen over keggle Competitions E.g. SVM vs. Logistic regression Read “A Few Useful Things to Know about Machine Learning” (Pedro Domingos)

Слайд 25

Challenges: Feature Engineering In ML feature engineering is the key [1]. You need features to form a kernel. Then you can solve with less data. Deep learning can learn best feature (combination) via semi or unsupervised learning [2] Bekkerman’s talk https://www.youtube.com/watch?v=wjTJVhmu1JM Deep Learning, http://cl.naist.jp/~kevinduh/a/deep2014/

Слайд 26

Challenges: Taking Decisions (Context)

Слайд 27

Challenges: Updating Models Incorporate more data We get more data over time We get feed back about effectiveness of decisions (e.g. Accuracy of Fraud) Trends change Track and update model Generate models in batch mode and update Streaming (Online) ML, which is an active research topic

Слайд 28

Challenges: Lack of Labeled Data Most data is not labeled Idea of Semi Supervised learning Provide Data + Examples + Ontology, and algorithm find new patterns Lot of Data Few example sentences Often uses Expectations Maximization (EM) Algorithm Watch Tom Mitchell’s Lecture https://www.youtube.com/watch?v=psFnHkIjHA0 Ontology: People, Cities Relationships: like, dislike, live in Examples: Bob (People) lives in Colombo (City)

Слайд 29

Two Takeaways Do your data Processing as part of a Bigger system Think Systems, automate, make a difference Realtime vs Batch Use tools ( Do not reinvent the wheel) Think how dynamics are changing (Uncontrolled experiments, lot of Data) Do not be a data Pessimist However, do not do stupid things either

Слайд 30