Computational Data Science

CMPT 353, Summer 2021

Greg Baker

https://coursys.sfu.ca/2021su-cmpt-353-d1/pages/

This Course

It's Computational Data Science. We'll come back to what that is.

Course web site: in CourSys, https://coursys.sfu.ca/2021su-cmpt-353-d1/pages/ .

Online Offering Strategy

My online instruction plan: change as little as possible from in-person offerings.

  • Don't remove material; don't add material.
  • Keep what can be done online.
  • Adapt the rest as best possible.

Online Offering Strategy

Lectures will be pre-recorded.

In the lecture time, they will be available as a YouTube Premier (≈ watch party) in ≈50 minute chunks. Greg will be available in YouTube chat to answer questions during that time.

They will be available as regular YouTube videos for viewing later.

Online Offering Strategy

Technology choices:

  • Lectures: YouTube.
  • Discussion forum: Piazza.
  • Office hours by video chat: Zoom (or ask in discussion).

Online Offering Strategy

Requirements for you:

  • A PC with a webcam.
  • … that is powerful enough to run a VM: at least 8 GB memory, 20 GB disk, reasonably decent processor (not too old, not a Celeron or other low-spec).
  • Windows, Mac, Linux all good.
  • A stable Internet connection.
  • Participation during lecture time for quizzes.

Grades

  • Weekly exercises: 12 × 3.5% = 42%
  • Project: 32%
  • Quizzes: 4 × 5% = 20%
  • Final quiz: 6%

Exercises

Due Fridays. My goal: make sure you actually try out the things we have talked about and see the reality of applying them.

Will contain some short problems to get you used to the tools, expanding to something a more interesting real problem.

Project

In the lectures/​exercises, I intend to explore what I consider the core of data science.

The project will let you integrate those techniques, and explore ideas on the edges of that, depending what interests you.

Project

I will (hope to?) post project topic options related to topics like…

  • natural language processing
  • image recognition
  • signal processing

Or if you have something else in mind, we can discuss it.

Project

A few details:

  • Groups of 2–3 (or something else?).
  • Take the given problem. Use the techniques from the course, and explore others to sensibly attack the problem.
  • In a report, summarize your methods, findings, and what worked/​didn't.

Quizzes/Exam

Quizzes: 5% each. Dates may change if necessary, but planned during lecture times:

  • June 11 (Friday of week 5)
  • July 2 (Friday of week 8)
  • July 23 (Friday of week 11)
  • August 6 (Friday of week 13)

Final Quiz: whenever our final exam is scheduled.

Us

Instructor: Greg Baker <ggbaker@sfu.ca>.

Office hour: Thursdays 13:00–14:00 by video conference and discussion forum.

Us

TAs:

  • Ali Arab
  • Ghazal Saheb Jam

Office hours: details later.

Lectures and Labs

Tuesdays: lectures as expected.

Fridays: usually no lectures. The TAs and I will all be available for consultation during the lecture time in discussion forum and video chat, mostly for the weekly exercises.

References

Textbooks: none.

Possible reference material:

References

Possible reference material (continued):

Programming

Python 3 will be the primary programming language language used in the course. If you aren't comfortable with it, you need to be (very) soon.

StackExchange Data Science tags (as of April 2021):

LanguageTagged Qs
Python5217
R1287
Matlab148
Java51
Scala43

Programming

This will be a programming-heavy course. If you don't really like programming, this might not be the course for you.

The programming style will be very library-heavy, which is realistic in the modern world. We will use many libraries: NumPy, Pandas, matplotlib, scikit-learn, statsmodels, ….

Programming

That means you'll spend a lot of time reading the docs and fighting to make the tools do what you want them to, and less implementing the logic yourself. That's also realistic.

The code you would have written would almost certainly have been slower and worse.

Programming

I will feel free to increase the amount of assignment work a little from my usual level because of the missing hour of lecture.

Expectations

To get credit for this course, I expect you to demonstrate that you know how to use programming techniques to manipulate and analyse data. That means:

  • A pass on the weighted average of the stuff where you demonstrate programming ability: exercises + project.
  • A pass on the weighted average of the quizzes.

Failure to do these may result in failing the course.

Expectations

Academic Honesty: it's important, as always.

If you're using an online source, leave a comment.

def this_function(p1, p2):
    # adapted from http://stackoverflow.com/a/21623206/1236542
    ...

That's all I ask, but remember to do it.

Expectations

You are expected to do the work in this course yourself (or as a group for the project).

Independent work is not copying a solution from somewhere but understanding it. If you work with another student, we shouldn't be able to tell from the results.

More details on course web site.

Expectations

The quizzes are structured as regular tests: individual work, but open book.

It will be hard to monitor that, but cheating on our quizzes will be treated as equivalent to cheating on an in-person exam, with corresponding penalties: I will be asking for a grade of FD in the course for any academic dishonesty on quizzes.

Computational Data Science?

Computational Data Science: data science, but with computation as the focus.

But what is data science?

Data Science?

According to Wikipedia: an interdisciplinary field about processes and systems to extract knowledge or insights from data in various forms…

According to Pat Hanrahan, Tableau Software: [The combination of] business knowledge, analytical skills, and computer science.

According to Daniel Tunkelang, LinkedIn: [The ability to] obtain, scrub, explore, model and interpret data, blending hacking, statistics and machine learning.

Data Science?

According to Joel Grus: There's a joke that says a data scientist is someone who knows more statistics than a computer scientist and more computer science than a statistician.… We'll says that a data scientist is someone who extracts insights from messy data.

Data Science?

According to Drew Conway, Alluvium:

Data Science?

My definitions:

Data Science
You get some data. Then what do you do to get answers from it? Whatever that is, that's data science.
Computational Data Science
You get some data. You know how to program. Then what do you do?

Why Data Science?

Why is data science suddenly so popular?

There's more data being collected: web access logs, purchase history, click-through rates, location history, sensor data, ….

Sometimes the volume of data is big: too big to manage easily. That's where big data starts.

Why Data Science?

People want answers/​insights from that data: Is the marketing campaign working? Is the UI actually usable? What if we did X instead of Y?

New techniques: Machine learning lets us attack questions that were previously unanswerable. Computer scientists are realizing that statistics is important; statisticians are realizing that computer science is important.

Topics (1)

  • Data science: what is it? How does data become useful?
  • Data processing tools: Python + NumPy + Pandas; analysis tools in Python.
  • Data aquisition. Or where do we find data?
  • Getting data into shape: cleaning; extract/​transform/​load.

Topics (2)

  • Making sense of data: statistics. Or it turns out that stats course was useful.
  • Making sense of data: machine learning. Or it's like AI, except it works.
  • Data analysis strategies.

Topics (3)

  • Big data tools: Apache Spark and a compute cluster.
  • Data visualization and communicating results.