skip to primary navigationskip to content
Home

Department of Computer Science and Technology

Data Science

Course pages 2021–22

Data Science

Course arrangements

  • Lecture notes These printed notes have some sections marked *, which are non-examinable. If you spot a mistake in the printed notes, please first check if it's corrected in the extended online notes (which I will update continually) and if not then email me.
  • Supervisions Your college may have have arranged either three or four supervisions for this course. If three, the suggested breakdown is:
    – Not for supervision: example sheet 0
    – Supervision 1: example sheet 1
    – Supervision 2: example sheet 2, and half of 3
    – Supervision 3: half of example sheet 3, and example sheet 4

Videos and example sheets

The videos have exactly the same material as a normal lecture, but they are short because there's no need for the lecturer to pause. You have a pause button — use it! You'll need to watch these videos several times, with the printed notes beside you. When there are handwritten equations in the videos, you should copy out the equations yourself to make certain you understand the method. When the video presents code, you should download the code and try it yourself. This will make the example sheets much easier. — The optional in-person sessions are digressions about off-syllabus topics and why it's all worth studying. THE IN-PERSON SESSIONS ARE OPTIONAL. THEY ARE NOT EXAMINABLE MATERIAL.
Example sheet 0 and solutions (prerequisites)
Lecture 1
1. Learning with probability models (4:08) fitting, climate
1.1 Specifying probability models (15:20) fitting
1.2 Standard random variables (3:21)
8 Oct, 11am Optional in-person: introduction [slides]
Lecture 2
1.3 Maximum likelihood estimation (17:35) fitting
1.4 Numerical optimization (8:01)
Lecture 3
1.5 Likelihood notation (10:00)
1.6 Generative models (8:14) fitting
1.7 Supervised learning (14:18) fitting
Lecture 4 Mock exam question 1 and walkthrough (23:36)
15 Oct, 11am Optional in-person: prediction and parameters [slides]
Lecture 5
2.1 Linear modelling (13:27) lm
(Videos for section 2 are unchanged from last year)
2.2 Feature design (19:39)
2.3 Diagnosing a linear model (5:29)
Lecture 6
2.4 Probabilistic linear modelling (9:45)
2.5 Linear algebra and least squares (12:07)
Lecture 7
2.6 Interpreting parameters (20:03)
Example sheet 1
22 Oct, 11am Optional in-person: climate dataset challenge [slides]
Lecture 8
Part II, 4.1 Bayes's rule for random variables (17:27)
4.2 Bayes's rule calculations (15:37)
The printed notes have an error in Example 4.2.1. See extended notes for correction.
4.3 Deriving the likelihood (9:17)
Lecture 9
5.1 Monte Carlo integration (13:21)
5.2 Bayes's rule via computation (19:25) bayes.ipynb
7, 7.1 Bayesianism (16:54)
Lecture 10
7.2–3 Finding the posterior (24:45)
7.4 Bayesian readouts (6:28)
29 Oct, 11am Optional in-person: R2 and model choice [slides]
Climate confidence challenge: find a 95% Bayesian confidence interval for the rate of temperature increase in the UK. Submit your answer on Moodle by 9am on Friday 5 November.
Lecture 11
6.1–6.2 Empirical cdf (15:10)
6.3 The empirical distribution (6:06)
8. Frequentism (3:58)
Lecture 12
8.1 Resampling / confidence intervals (13:03) freq.ipynb
8.2 Hypothesis testing (23:37)
8.4 Non-parametric resampling (7:00)
Lecture 13
5 Nov Optional in-person: the problem of induction [slides]
Climate confidence challenge: find a 95% frequentist confidence interval for the rate of temperature increase in the UK. Submit your answer on Moodle by 9am on Friday 12 November.
Lecture 14
10, 10.1 Markov chain models (12:09) markov.ipynb
10.2 Likelihood and memorylessness (4:56)
10.3–4 Sequence models with history (9:57)
Lecture 15
11 Markov chain behaviour (3:22) markov.ipynb
11.1–2 Calculations with Markov chains (11:37)
11.4–6 Stationarity and limiting behaviour (12:38)
Lecture 16
11.7 Drift analysis (11:23)
Example sheet 4
  • Stoat-finding challenge: submit your answer on Moodle by midnight on 1 December. Top three submissions win stylish and elegant Data Stoat t-shirts.
12 Nov Optional in-person: Bayesian versus frequentist? [slides]

Study at Cambridge

About the University

Research at Cambridge