1 of 61

1

Applied Data Analysis (CS401)

Maria Brbić / Robert West

Lecture 1

Intro to ADA

11 Sep 2024

2 of 61

Important websites

2

http://ada.epfl.ch

Your main entry point. All materials linked from there.

https://github.com/epfl-ada/2024

Used for exercises, homework, project, and final exam.

https://go.epfl.ch/ada2024-ed

Main communication channel. Sign in with your EPFL email address (or simply access via Moodle).

3 of 61

Previous instructor

3

​

  • Robert West
  • On sabbatical in Microsoft Research, Seattle

​

​

​

4 of 61

About your instructor this year

4

​

  • Born in Tučepi, Croatia

​

​

​

  • Education:

University of Zagreb, Croatia

University of Tokyo

Stanford University, USA

​

  • Assistant Professor at EPFL since Sep ‘22

Machine Learning for Biomedicine (MLBio) lab

​

~ 3.850.000 people

~1500 people

5 of 61

Our research @ MLBio

5

My Research

Biomedicine

MLBio

Lab

Challenges

Methods

Machine

Learning

6 of 61

Our research @ MLBio

6

  • Collaborate with biologists and medical researchers and have new “untouched datasets” or collect datasets
  • Gain new insights from these datasets 🡪 what are interesting questions that need new AI algorithms to be answered?
  • Apply AI algorithms we develop to advance biomedical research and drive new discoveries in biology and medicine
  • Develop new AI methods
  • Unsupervised learning, generative models, open-world learning etc

7 of 61

7

Our research @ MLBio

8 of 61

Data analysis

“... the process of inspecting, cleaning, transforming, and modeling data with the goal of discovering useful information, suggesting conclusions, and supporting decision-making.”

“Data analysis has multiple facets and approaches, encompassing diverse techniques under a variety of names, in different business, science, and social science domains.”

8

9 of 61

Applied data analysis

  • This course is about breadth, not depth
  • “What methods, principles, and tools are out there?”, rather than “How can I become an expert in deep learning for computer vision applied to images of cats?”
  • Data science is a fast-paced, shifting field
  • Obsessing on one tool or technique won’t pay off in a few years
  • Be ready to explore and keep learning on your own

9

Complementary courses:

Machine learning

NLP

DIS

Data viz

Goal of this class: Enable you to conduct a

full-fledged data science project from start to finish

That being said, depth matters too…

10 of 61

Let’s abbreviate this course as Ada, not A-D-A, in honor of Ada Lovelace, “the world’s first computer programmer.”

https://en.wikipedia.org/wiki/Ada_Lovelace

10

11 of 61

11

12 of 61

Syllabus

  • Handling data
    • “Slicing and dicing”: obtaining, preparing, juggling data
  • Visualizing data
    • Exploration of data, communication of results
  • Describing data
    • How to support (and be suspicious of) claims about data
  • Regression analysis for disentangling data
    • How to disentangle datasets with correlated variables
  • Causal analysis of observational data
    • How to deal with “found data”
    • Correlation != causation

12

13 of 61

Syllabus (cont’d)

  • Learning from data
    • Supervised learning
    • Unsupervised learning
    • Applied aspects of machine learning
  • Handling specific types of data
    • Handling text data
    • Handling network data
  • Scaling to massive data

13

14 of 61

Grading

  • 35% Homework assignments (2)
    • Involving skills required from data scientists
    • Groups of 5 students
    • Homework of 2017, 2018, 2019, 2020, 2021, 2022, 2023
  • 35% Final exam
    • Mini data analysis project
    • Done on laptop, individually, on campus
    • Final exams of 2017, 2018, 2019, 2020, 2021, 2022, 2023
  • 30% Project (more details soon)
    • Your own freestyle data analysis
    • Done in groups of 5 students (same as for homework)
    • Milestones spread throughout the semester
    • Projects of 2017, 2018, 2019, 2020, 2021, 2022, 2023

14

15 of 61

Grading

  • 35% Homework assignments (2)
    • Involving skills required from data scientists
    • Groups of 5 students
    • Homework of 2017, 2018, 2019, 2020, 2021, 2022, 2023
  • 35% Final exam (date TBD)
    • Mini data analysis project
    • Done on laptop, individually, on campus
    • Final exams of 2017, 2018, 2019, 2020, 2021, 2022, 2023
  • 30% Project (more details soon)
    • Your own freestyle data analysis
    • Done in groups of 5 students (same as for homework)
    • Milestones spread throughout the semester
    • Projects of 2017, 2018, 2019, 2020, 2021, 2022, 2023

15

This class will be hard work,

but it will get you a job.

16 of 61

Grading (cont’d)

  • To obtain a meaningful grade distribution, scaling/shifting will be applied to each of {homework, project, exam, quizzes} before taking weighted average (standard practice at EPFL)
  • While intermediate grades are a good indication of where you stand, remember there might be some wiggle

→ Don’t rely on intermediate grades to decide whether you can afford to skip the exam etc.

16

17 of 61

Deadlines

  • Homeworks
    • Homework 1
      • Release Oct 4th 2024
      • Due Oct 18th 2024
    • Homework 2
      • Release Nov 15th 2024
      • Due Nov 29th 2024
  • Final exam
    • Date TBD
  • Project deliverables
    • Project milestone P1
      • Due Oct 4th 2024
    • Project milestone P2
      • Due Nov 15th 2024
    • Project milestone P3
      • Due Dec 20th 2024

17

All deadlines are 23:59 CET

18 of 61

Meeting logistics: Lectures

  • Wednesdays 8:15–10:00
  • If you want to see it live, come to class! (No live streaming)
  • Lectures are also recorded and made available after class

18

19 of 61

Meeting logistics: Lab sessions

  • Fridays 13:15–14:45
  • In person only:
    • CO 1
    • CM 1 120
    • CM 1 121
  • Labs are complementary to lectures, not simply more detail on same
  • You solve exercises that we make available the day before, can ask questions and get help from assistants
  • In certain weeks: homework/project office hours (probably on Zoom, in parallel to exercises)

19

20 of 61

Weekly quizzes

  • Available online on Moodle after every lecture
  • 5 questions, to be answered within 10 minutes of starting
  • Quiz 1: no real questions, just to let you get familiar with the setup
  • Quiz 2: the first quiz with lecture material questions
  • Quiz i is about lecture material of week i
  • Goal:
    • Engage continuously with course material
    • Think (not just find right slide)
  • Not graded, for you to recap lecture materials

20

21 of 61

Project

  • We’ll provide a number of datasets
  • You need to form and pitch a crisp project idea
  • Free to combine with other datasets (at your own risk)
  • Goal: not a loose collection of results – tell a story with the data!
    • Data stories of 2017, 2018, 2019, 2020, 2021, 2022, 2023
    • Nice example data story

21

22 of 61

Homework and projects: GitHub

  • De-facto standard for managing and sharing code
  • All students in this class need a GitHub account
  • Homework and project submissions done via GitHub
  • ADA Github repository:

https://github.com/epfl-ada/2024

​

22

23 of 61

Main communication channel:

  • Class forum, available via Moodle
  • Also accessible directly, outside of Moodle:�https://go.epfl.ch/ada2024-ed �(sign in using the same email address as for Moodle)
  • Central place to ask all class-related questions
  • Don’t send us email
  • Mandatory! We’ll send important announcements on Ed only
  • Help each other (without cheating, of course)

23

24 of 61

Watch-at-home videos

  • Throughout the semester, we’ll release videos with supplemental information; e.g.,
    • Intro to lab sessions (already available!)
    • Project instructions
    • Homework 1 postmortem
    • Homework 2 postmortem

24

25 of 61

General note on communication

  • Multiple platforms used in ADA for various tasks (as in real life): Ed, GitHub, Google docs, ADA website
  • To avoid confusion,
    • familiarize yourself with communication guidelines
    • all materials will be linked from the website as a central point of entry: https://ada.epfl.ch
    • all discussions will take place on Ed

25

26 of 61

Commercial break

26

ADA students: sharp like teeth!

27 of 61

Group registration

  • Must form teams within 2 weeks, starting now (in time for release of Homework 1)!
  • Get started immediately to find 4 other teammates
  • By Fri 27 Sep 23:59, complete the registration form (to be done by each team member individually):�https://go.epfl.ch/ada2024-team-registration
  • Can change team after Homework 1 (but try to avoid it)

27

28 of 61

Prerequisites

Basics of

  • probabilities and stats
  • databases
  • programming
    • You won’t survive if you can’t program
    • Homework, exam: Python required
    • Project: up to you, but we support only Python
    • Brush up your Python skills (many great online courses out there)

28

29 of 61

Python environments

  • Homeworks and exams to be done as Jupyter Notebooks
  • You will submit a pre-executed .ipynb file
    • We don’t care how you produce it
    • Option 1: local Python installation (e.g., Anaconda + JupyterLab)
    • Option 2: Google Colab = notebook hosted by Google
    • Option 3: noto = notebook hosted by EPFL
  • To get started: come to Friday’s lab session (“Exercise 0”)
  • “Homework 0”: do it yourself at home after lab session (optional, not graded)
  • Doing Homework 0 is the best way of making sure you’re set up correctly for later homework, project, exam

29

30 of 61

Python++

30

31 of 61

31

POLLING TIME

​

  • “What is your prior experience with Python?”
  • Scan QR code or go to�https://app.sli.do/event/iPMVwwrojFn9EaNp6iiDDx

32 of 61

Maria Brbić

Instructor

Head TAs

Marija Šakota

Saibo Geng

Aoxiang�Fan

Tim�Davidson

Sepideh

Mamooler

Artyom�Gadetsky

Yist

Yu

Shawn Fan

Yulun

Jiang

TAs: Teaching assistants (PhD students)

Mohammad

Masani

Silin�Gao

Shuo

Wen

Bettina

Messmer

Ivan

Zakazov

Shuqi

Wang

SAs: Student assistants (Master students)

Stefan

Krsteski

Xi

Lei

Matea

Tashkovska

Elisa

Billard

Pablo

Menéndez

Jiaming Jiang

Yannis Laaroussi

Allocio

Jeanne

Jakhongir Saydaliev

Sebastien

Chahoud

Yagiz

Gençer

Tianhao

Dai

Tugba

Tümer

Shuangqi

Li

​

Ziyi

Zhang

​

Yunzhen

Yao

33 of 61

  • Help each other on Ed
  • Participate actively in classes and labs
  • Give us feedback

33

34 of 61

Feedback

34

Give us feedback on this lecture here: https://go.epfl.ch/ada2024-lec1-feedback

​

Feedback form available for each lecture and lab session

​

  • What did you (not) like about this lecture?
  • What was (not) well explained?
  • On what would you like more (fewer) details?
  • …

​

35 of 61

Questions?

35

36 of 61

What is data science?

36

37 of 61

Why now?

37

I WISH I HAD BECOME A DATA SCIENTIST BACK IN 2023…

38 of 61

“Data science”

  • Most science is (or should be) based on data, per definitionem
  • So how is “data science” different from plain old “science”?

38

39 of 61

Data volume explodes

Eric Schmidt, Google (2010)

39

“Between the dawn of civilization and 2003, we only created five exabytes of information; now [in 2010] we’re creating that amount every two days.”

40 of 61

Data variety explodes

Text (indexed Web pages, email), networks (Web graph, knowledge graph), images, maps, logs (search logs, server logs, GPS logs), speech, …

40

41 of 61

Needed: A method to the madness

  • Scientific method 1.0:�- Focused on “Model the data”�- Scientist has hypothesis prior to analyzing the data
  • Scientific method 2.0:

- Data-driven science � - Systematic cycle (see diagram)� - “Explore the data” becomes increasingly important�

41

� Data as a first-class citizen

42 of 61

Scientist 2.0

42

43 of 61

Hilary Mason, chief scientist at bit.ly

43

“A data scientist is someone who can obtain, scrub, explore, model, and interpret data, blending hacking, statistics, and machine learning. Data scientists not only are adept at working with data, but appreciate data itself as a first-class product.”

​

44 of 61

Josh Wills, Data Scientist at Slack

44

45 of 61

45

46 of 61

46

47 of 61

More data often beats better algorithms

47

http://www.incompleteideas.net/IncIdeas/BitterLesson.html

48 of 61

21st-century politics

49 of 61

https://www.pewresearch.org/politics/wp-content/uploads/sites/4/2019/06/PP_2019.06.19_Political-Discourse_FINAL.pdf

We ask: Do these subjective impressions reflect the true state of US political discourse?

ADA will teach you the tools to answer such questions using data (see next slides)

50 of 61

Syllabus, revisited

  • Handling data
  • Visualizing data
  • Describing data
  • Regression analysis for disentangling data
  • Causal analysis of observational data
  • Learning from data
  • Handling text data
  • Handling network data
  • Scaling to massive data

50

Data: https://github.com/epfl-dlab/Quotebank

Web interface: https://quotebank.dlab.tools/

51 of 61

Syllabus, revisited

  • Handling data
  • Visualizing data
  • Describing data
  • Regression analysis for disentangling data
  • Causal analysis of observational data
  • Learning from data
  • Handling text data
  • Handling network data
  • Scaling to massive data

51

52 of 61

Syllabus, revisited

  • Handling data
  • Visualizing data
  • Describing data
  • Regression analysis for disentangling data
  • Causal analysis of observational data
  • Learning from data
  • Handling text data
  • Handling network data
  • Scaling to massive data

52

“Is the effect real, or could it have been produced by chance?”

53 of 61

Syllabus, revisited

  • Handling data
  • Visualizing data
  • Describing data
  • Regression analysis for disentangling data
  • Causal analysis of observational data
  • Learning from data
  • Handling text data
  • Handling network data
  • Scaling to massive data

53

54 of 61

Syllabus, revisited

  • Handling data
  • Visualizing data
  • Describing data
  • Regression analysis for disentangling data
  • Causal analysis of observational data
  • Learning from data
  • Handling text data
  • Handling network data
  • Scaling to massive data

54

“What caused the observed increase in negativity?”

55 of 61

Syllabus, revisited

  • Handling data
  • Visualizing data
  • Describing data
  • Regression analysis for disentangling data
  • Causal analysis of observational data
  • Learning from data
  • Handling text data
  • Handling network data
  • Scaling to massive data

55

Quotes were attributed to speakers by a machine learning algorithm

56 of 61

Syllabus, revisited

  • Handling data
  • Visualizing data
  • Describing data
  • Regression analysis for disentangling data
  • Causal analysis of observational data
  • Learning from data
  • Handling text data
  • Handling network data
  • Scaling to massive data

56

Research question (“Did political discourse become more negative?”) is a question about language == text

57 of 61

Syllabus, revisited

  • Handling data
  • Visualizing data
  • Describing data
  • Regression analysis for disentangling data
  • Causal analysis of observational data
  • Learning from data
  • Handling text data
  • Handling network data
  • Scaling to massive data

57

“Who speaks about whom in what way?” → Construct “who-mentions-whom” network

58 of 61

Syllabus, revisited

  • Handling data
  • Visualizing data
  • Describing data
  • Regression analysis for disentangling data
  • Causal analysis of observational data
  • Learning from data
  • Handling text data
  • Handling network data
  • Scaling to massive data

58

Extracting 235 million quotes from 127 million news articles (about 1 terabyte of raw text) spanning 12 years requires big-data tools (e.g., Spark)

59 of 61

Curious to learn more?

59

Full paper available at https://www.nature.com/articles/s41598-023-36839-1

60 of 61

TODO before Friday’s lab session

  • Sign up for Ed here and familiarize yourself with it
  • If you’re not on GitHub yet, sign up for GitHub
  • Start looking for 4 teammates
    • You may use “Group formation” category on Ed
  • Check out Google Colab and noto (to see if you want to use either of them)
  • Check out Exercise 0 here (in prep for Fri lab session)

​

60

61 of 61

Any feedback? -- Let us know!

61

Give us feedback on this lecture here: https://go.epfl.ch/ada2024-lec1-feedback

​

Feedback form available for each lecture and lab session

​

  • What did you (not) like about this lecture?
  • What was (not) well explained?
  • On what would you like more details?
  • What would you like the instructor to wear next time?
  • …

​