MATH2931: Higher Linear Models

2 x assignments (15%, 20%), 1 final exam (60%), 5% class participation.

Even though the pre-requisite is MATH2801 (DN) / MATH2901, I

I think the structure of this course really says a lot about the course. Even with previous lecturers, the course has not fared well so it says more about the content taught in the course than the lecturer. The course ties statistical modeling with the theory of linear models so a lot of what you need to know (particularly with the higher course) really comes down to coding it up on a language such as MATLAB, Python, and R. I, on the other hand, enjoy the theory so it didn't really sit well with me.

The course, as a whole, was unstructured and I didn't feel like I gained too much from doing this course. It was confusing to follow the lecture content because course content was organised by timestamps more than specific topics so it was really in your best interest to continuously keep up with the course itself. Zdravko typesetted his lecture material on Overleaf live so hearing the the keyboard clacks while listening to him speak was a bit distracting at times.

One of the biggest downfalls with this course is that Zdravko never really emphasised the coding in the course. It's a core component of the course with one question specifically dedicated to coding in the final exam, so him not really emphasising the coding component in lectures disheartened me from wanting to even attempt the question (thankfully, it's optional).
Preparing for the final exam was a bit of a headache, purely because we aren't given too many resources to work off of besides going over the tutorial problems repeatedly. It's not a terrible course but it's not a memorable one either.

2 x 2 hour lectures, 2 x 1 hour tutorials.

4/5.

Yes

Dr. Zdravko Botev.

Live scribbles are available after each lecture.

2.5/5.

None prescribed, but recommended:
- (Less advanced) Dirk P. Kroese and Joshua CC Chan. Statistical modeling and computation.
- (More advanced) Dirk P. Kroese et al. Data Science and Machine Learning: Mathematical and Statistical Methods.

20T3.

72 (CR).

3 x Assignments (10% each), finals weighted 55%.

MATH2901 is a prerequisite. For this course, some elements from MATH2501/MATH2601 are implicitly assumed (although not explicitly examined).

This course is basically the continuation of MATH2901 and essential to any statistics major student. It takes the concepts of statistical inference introduced in its predecessor and essentially seeks to introduce basic model fitting and analysis. Much of the content in this course revolves around R; you are not required to write R code but you will need to interpret given code in assignments and in the exam.

For me, this course felt significantly more dry and bland than its precursor. The first half of the course introduces all the essentials to model fitting and the concepts behind it, but it gradually turns into just grind and rote. It becomes more memorisation in the later half, and whereas the proofs are decent they start becoming very convoluted. It's more or less about how to fit a model that does whatever it does, and just what deductions you can make out of it. You also need to know the uses of various forms of measure (e.g. Mallow's C_p and the PRESS statistic for goodness of fit).

This course would've been rated a 1/5, but every course is made better by the presence of Libo and that can't be denied.

I don't regard this as a difficult mark despite getting a considerably lower mark in it than MATH2901. I just find it a lot less interesting.

It should be remarked again that linear algebra (MATH2501 OR MATH2601) is not a prerequisite for this course. Linear algebra is just an aid used for the proofs in this course. Remember that MATH2931 assumes MATH2901, WHICH assumes MATH1231/41/51, so elementary linear algebra concepts should not be foreign. Stuff like spectral decomposition, may, however, be a bit unfamiliar.

Note: The lectures for this course are combined with its ordinary counterpart MATH2831. This is due to the cohorts being appreciably smaller than that of MATH2801/MATH2901. MATH2831 students aren't expected to deal with much of the linear algebra components and have a few less things to memorise.

3 x 1 hour lectures, 1 hour tutorial

2/5

Yes, but you miss quite a fair bit of what's done on the blackboard.

Dr Libo Li

As with MATH2901, Libo releases his lecture notes.

2.5/5

N/A

2017/2

80 DN