Data Science – CS109

Joe Blitzstein, Verena Kaynig-

Fittkau, Hanspeter Pfister
 Final Projects
http://cs109.github.io/2015/pages/projects.html
 Project Dates
• Week of Nov 23-27: Data collection and
cleaning finished
• Week of Nov 30- Dec 4: Exploratory analysis
finished, some modeling/visualizations, start
website
• Week of Dec 7-10: Modeling and/or prediction
finished, website/screencast/(maybe)
visualizations left
• Th, December 10: Final project due (11:59 pm)
• December 15: Best projects shown in class 

(location TBA)
 IPython Process Book
• important part of your project
• standalone document that fully describes your
project in detail
• overview and motivation
• related Work
• initial Questions
• data: source, scraping method, cleanup, etc.
• exploratory data analysis
• final analysis
• Use visualizations!
 Code – Zen (Picks)
• The Zen of Python, by Tim Peters
• >>> import this

• Explicit is better than implicit

• Simple is better than complex
• Flat is better than nested
• Readability counts
• Errors should never pass silently
• Now is better than never
• If the implementation is hard to explain, it's a
bad idea
 Website
• Use Github, Google Sites, Personal Sites
• Open to the public
• Links to Ipython notebook and data
• Visualizations
• Video
 Video
• 2 min limit
• Use a good microphone
• Do not just scroll through your notebook!
• Use what you learned about story telling
and presentations
• Do not put in too much
• Show demos! Show visualizations! Be
engaging
 Some Examples
• Predicting Hubway availability, L. Alexander, G. Goulet-
Langlois, J. Wolff
https://www.youtube.com/watch?v=2wK8jpNMjXI&feature=youtu.be

• AirBnB success, H. Husain, Y. Mao, L. Awad, X. Li

https://www.youtube.com/watch?v=raGjUj5qArc

• The green canvas, A. Hosny, J. Huang , Y. Wang 

https://vimeo.com/114379373
 Peer Assessment
• Preparation - were they prepared during team
meetings?
• Contribution - did they contribute productively
to the team discussion and work?
• Respect for others’ ideas - did they encourage
others to contribute their ideas?
• Flexibility - were they flexible when
disagreements occurred?

• You will also assess your own performance

What we learned
 A (cs109) Data Scientist Is...
• a programmer: Python
• a statistician: Statistical Learning Theory
• a computer scientist: Algorithms, Risk
Minimization
• an engineer: AWS, Spark, installs, debugging, etc
• a story teller: essays, presentations, projects
• an artist: visualizations, how to show information
 Some Key Principles
• use many data sources (the plural of anecdote is not data)

• understand how the data were collected (sampling is essential)

• weight the data thoughtfully (not all polls are equally good)

• use statistical models (not just hacking around in Excel)

• understand correlations (e.g., states that trend similarly)

• think like a Bayesian, check like a frequentist (reconciliation)

• have good communication skills (What does a 60% probability

even mean? How can we visualize, validate, and understand
the conclusions?)
 What we did do
• Learning a model: EDA, play-viz, simplicity, complexity, features, overfitting,
train-test, validation, cross-validation, regularization

• Statistics: Simple Distributions, LLN, CLT, Hypothesis tests, Multiple

Comparisons, Bootstrap, Bayesian Stats, Hierarchical Models, regression-to-
the-mean, regularization

• Machine Learning: (Empirical) Risk Minimization, Decision theory, Feature

selection, Pipelines, Dimensionality and reduction (PCA), Regression,
Classification, Clustering. Generative vs Discriminative vs Discriminant.

• Models: Perceptron, SVM, Logistic, LDA, NB, K-Means, LDA, Hierarchical

Clustering, Bagging, Boosting, Stacking, RF.

• Engineering: Data Model - SQL, Dataframes, Scraping, Cleaning, Imputing,

AWS, Containers, Struggling.

• Story Telling: Writing, Presenting,Visualization, Stories.

 What we didn’t do (but
could have)
• Optimization of a general loss: Gradient
Descent, Stochastic Gradient Descent
• Inference on a ROC curve (Inference in
General)
• Connect the Bayesian techniques to the
frequentist

• More Ops: Big Data, SQL, AWS, Docker…

 Machine Human
Data Management Human Cognition

Data Mining Perception

Machine Learning Visualization Story Telling

Business Intelligence Decision Making

Theory
Statistics
Data Science

Inspired by Daniel Keim, “Visual Analytics: Definition,

Process, and Challenges”
Netflix Prize
 Some Challenges
• massive data (500k users, 20k movies, 100m ratings)

• curse of dimensionality (very high-dimensional problem)

• missing data (99% of data missing; not missing at

random)

• extremely complicated set of factors that affect people’s

ratings of movies (actors, directors, genre, ...)

• need to avoid overfitting (test data vs. training data)

• what model? local? global? ensemble?

 Why do this?
“...10 years from now, each cancer patient
is going to want to get a genomic analysis
of their cancer and will expect
customized therapy based on that
information.”
Director, The Cancer Genome Atlas
(TCGA), Time Magazine, 6/13/11
“By 2018, the US could face a shortage of
up to 190,000 workers with analytical
skills”
McKinsey Global Institute

“The sexy job in the next 10 years will be

statisticians.” Data Scientists?
Hal Varian, Prof. Emeritus UC Berkeley
Chief Economist, Google
 External Courses
• https://www.coursera.org/learn/machine-learning/
(log in to access videos): Andrew Ng

• https://www.coursera.org/course/machlearning :
Pedro Domingos (also his book, The Master
Algorithm)

• Learning From Data (more mathy, with book): https://

work.caltech.edu/telecourse.html

• Harvard Computefest

• Stanford Statistical Learning MOOC: http://www.r-

bloggers.com/in-depth-introduction-to-machine-
learning-in-15-hours-of-expert-videos/
 Quick Reads
• A Few Useful Things to Know about Machine
Learning: https://homes.cs.washington.edu/~pedrod/
papers/cacm12.pdf

• Bayesian Reasoning and Machine Learning Chapter13

(http://web4.cs.ucl.ac.uk/staff/D.Barber/pmwiki/
pmwiki.php?n=Brml.Online)

• Murphy Chapter 1: https://www.cs.ubc.ca/~murphyk/

MLbook/
 Useful free books
• http://statweb.stanford.edu/~tibs/ElemStatLearn/

• http://www-bcf.usc.edu/~gareth/ISL/

• http://web4.cs.ucl.ac.uk/staff/D.Barber/pmwiki/pmwiki.php?
n=Brml.Online

• http://www.inference.phy.cam.ac.uk/itila/book.html

• https://github.com/CamDavidsonPilon/Probabilistic-
Programming-and-Bayesian-Methods-for-Hackers (also dead
tree now)

• http://www.stat.cmu.edu/~cshalizi/ADAfaEPoV/
 Some other resources
• Non free:Very recent but very relevant: https://
github.com/rasbt/python-machine-learning-book (also
read Sebastian’s Blog)

• simplystatistics.org

• The YHat Blog and the RODEO IDE.

• Chris Fonnesbeck’s BIOS 8366 at Vanderbilt: http://

stronginference.com/Bios8366/

• Follow Liz Cherney, Andrej Karpathy, Yhat, Chris

Albon, John Myles White, Cam Davidson Pilon and their
networks.
Next Step: Stats Courses
http://smbc-comics.com/index.php?id=3930
http://smbc-comics.com/index.php?id=3930
Enter here
(visitors Stat
welcome) Stat 110 Stat 111
100/101/102/104
Probability Inference
Statistics
Stat 135
Statistical
Computing Theoretical Foundation
Software
Stat 149
Stat 139
Generalized
Linear
Linear
Models
Models

Stat 121/
CS 109 Stat 186 Stat 107 Stat 131
Data Science Big Data Business/ Time
Financial Series
Applied Foundation
Stat 123
Applied Quantitative
Finance
Stat 140 Stat 186 Stat 160
Design of Causal Sample
Experiments Effects Surveys Stat 170
Quantitative
Design Foundation Analysis of Capital
Markets
Stat 171 + Econ
Stochastic
Stat 115 Processes
Computational
Biology
+ Life Sci
Next Step: CS / AM Courses
https://sites.google.com/a/g.harvard.edu/seasthreeyearcourseplan/
 Data Wrangling
• Python
• Pandas
• Spark
• Map Reduce
 CS124 -Data Structures and
Algorithms
Spring / Prof. Michael Mitzenmacher

• Fundamentals
• Graph Algorithms
• Greedy Algorithms
• Dynamic Programming
• Divide and Conquer
• Hashing
• Linear Programming
• Randomized Algorithms
• NP-completeness review
• Novel approaches to NP-complete problems
 CS 165 – Data Systems
Fall / Prof. Stratos Idreos

• Expected learning outcomes

• Become familiar with the history and evolution of
data systems design over the past 4-5 decades.
• Understanding the basic tradeoffs in designing
and implementing modern data systems.
• Being able to design a new data system given a
data-driven scenario and to built a prototype.
• Being able to understand which data system is a
good fit given the needs of an application. 
• Advanced C programming and debugging skills.
 CS207- Systems Development for
Computational Science
Spring / Dr. Rahul Dave

• Apply basic software development tool-chains, including

source-code control, testing frameworks, and documentation
tools, to the process of designing and implementing large
software systems;
• Apply design principles to the decomposition of software into
re-usable components, and to the production of those
components;
• Know how to approach an existing piece of software for
maintenance, extension, and modification;
• Design, develop, and deploy a set of software components to
produce a scalable, reliable, and reproducible experimental
system for scientific investigation;
• Use a variety of approaches to software development team
organization, and select techniques that are appropriate in
different circumstances.
 CS205 – Computing Foundations for
Computational Science
Fall / Dr. Ray Jones

• Apply basic computer science concepts such as

modularity, abstraction, and encapsulation to scientific
problems
• Recognize and recall computer architectures, algorithms,
and data structures that are relevant to computational
science
• Apply concepts of parallel programming and “parallel
thinking” to computational science
• Analyze and visualize large scientific data and implement
data-intensive computations on cluster and cloud
infrastructures
• Use open-source tools for large- and fine-grain parallel
computations, cloud computing, and visualization
 CS262 - Introduction to Distributed
Computing
Spring / Prof. Jim Waldo

• design and implementation of large

systems
• running on multiple computers connected
by a network.
• investigate the fundamental characteristics
of distributed systems
• investigate how to build systems that
exploit those fundamental characteristics.
 Visualization and Storytelling
• EDA
• Effective visualization
• Interactive visualization
• Story telling
 CS171-Visualizations
Spring / Prof. Hanspeter Pfister
www.cs171.org

• After successful completion of this course, you will be able to…

• Critically evaluate visualizations and suggest improvements and
refinements
• Apply a structured process to design interactive visualizations
• Iteratively generate visual encoding ideas using sketching and
prototyping
• Use principles of human perception and cognition in visualization
design
• Know various visualization methods to judge and create new designs
• Create web-based interactive visualizations using JavaScript and D3
• Work constructively as a member of a team to plan and carry out a
complex project
 Machine Learning
• Supervised learning:
– K-nearest neighbors
– SVM
– Decision Trees
– Random Forest
– Boosting
• Unsupervised learning:
– Dimensionality reduction
– Clustering
 CS181 – Machine Learning
Spring / Prof. Finale Doshi-Velez

•  Introduction and Course •  Neural Networks

Overview • Regression and Classification
• Clustering with K-Means and Trees
K-Medoids • Max-Margin Classification
•  Hierarchical Clustering • Support Vector Machines
• Principal Component Analysis •  Markov Decision Processes
•  Supervised Learning • Reinforcement Learning
•  Linear Regression • Partially-Observable Markov
•  Model Selection Decision Processes
•  Linear Classification • Expectation Maximization
• Classification and CV Review • Hidden Markov Models
• Probabilistic Classification
 CS182 – Artificial Intelligence
Fall / Prof. Sasha Rush

• After successful completion of this course, you will be able

to…
• choose the appropriate representation for an AI problem or
domain model, and construct domain models in that
representation
• choose the appropriate algorithm for reasoning within an AI
problem domain
• implement and debug core AI algorithms in a clean and
structured manner
• design and analyze the performance of an AI system or
component
• describe AI algorithms and representations and explain their
performance, in writing and orally
• critically read papers on AI systems
CS187 – Computational Linguistics
Fall / Prof. Stuart Shieber

Watson is the world Jeopardy champion. Siri responds

accurately to "Should I bring an umbrella tomorrow?". How do
they work? This course provides an introduction to the field of
computational linguistics, the study of human language using
the tools and techniques of computer science, with
applications to a variety of natural-language-processing
problems such as those deployed in Watson and Siri, and
covers pertinent ideas from linguistics, logic programming, and
statistical modeling. The course will include an experimental
practicum component covering skills in technical writing and
editing that should be of general use as well.
 CS283 – Computer Vision
Fall / Prof. Todd Zickler

The goal of computer vision is to create artificial vision systems

that can reliably extract information from images. The field is at
an exciting stage of development, spurred by the explosion of
online imagery, the emergence of new visual sensors, and the
growth of processing and storage capabilities. After decades of
hard work, we are finally starting see some stunning
applications. I argue that the time for vision is now.

Addressing the vision problem requires an understanding of

image formation (radiometry, optics, projective and differential
geometry) as well as a firm grasp of mathematical and
computational tools for image analysis. So this course, which
is designed to be a first course in computer vision, covers a bit
of both. It also blends theory and practice.
 AM207 - Monte Carlo Methods for
Inference and Data Analysis
Spring / Dr. Verena Kaynig

• Introduction to basic Monte Carlo methods

• Bayes formalism and sampling
• Stochastic Optimization
• Dynamic systems
• Advanced sampling methods
• Graphical Models
 CS105 – Privacy and Technology

(GOV1430)
Fall / Prof. Jim Waldo

• Privacy Concepts
• Surveillance
• Tapping and tracing
• Data aggregation, analytics, and privacy
• Data-intensive Science
• Biometrics and DNA
• Master of Science in CSE
– 1 year, 8 courses
• Master of Engineering in CSE
– 2 years, 8 courses, 1 thesis
• AB/SM in CSE option

• For PhD students:

– secondary field in CSE
– 4 courses
 IACS Compute Fest
• Skill building workshops
– Monday, January 11 - Friday, January 15
• Student computational challenge!
– Tuesday, January 19- Thursday, January 21
• IACS Symposium: Brain & Machines
– Friday, January 22

http://computefest.seas.harvard.edu