YEAR 3 SEM 2 (AY17/18) MODULE REVIEW

This semester, I decided to take a plunge and took on the challenge of taking 6 modules.
Well, that's the worst mistake ever.

I cannot emphasise this more but PLEASE DON'T DO THIS!
It's really f*&^%ing crazy. Initially I thought that it wouldn't be that difficult since statistics modules should probably relate to each other. NAHHHHHHHHH!!!! My cap dropped 0.52 this semester but oh well , I'm still eligible for FYP.

So how does it feel like to take 6 stats mods?
Every week on average, I have 4 submissions.
Once I'm done with one project, I have another to do. Most of the time I'm just rushing countless of submissions/assignments... It feels like having shit falling on my face everyday, everywhere I go.
Plus I had cca commitments as well as 3 tuition students.. So you can probably imagine how dark my eye rings were after this extremely torturous semester. Do not even attempt.

Modules taken this semester:

1. ST4233 Linear Models

Lecturer: Zhang Jin Ting

Grade expected: C
Grade Obtained: D+  (first time getting a grade below B- , hmm..)
Module difficulty: 9/10
Webcast: Yes

About the lecturer: 
The prof's chinese accent is really one of a kind, took me months to get used to it.
So yes good luck to y'all because this module is compulsory.
I wouldn't take another module under him although he can be quite funny sometimes.

What you learn in this module: 
I don't know man.. Just irrelevant proofs without any numbers, there's just so many X and X transpose and X inverse...
Actually I got A- ST3131 Linear Regression so I thought that ST4233 wouldn't be a problem. Well, it turns out that these modules are really not that similar.
4233 is all about proving. MA1101 Linear Algebra is especially important in this module, which is really bad news for me because I can't remember anything from that module (think about linear independence, eigenvalues eigenvectors, decomposition idk what). Will be good if you can revise your LA in advance, or take this module together with friends so that y'all can help each other out for the assignments.

This module was really painful for me because I took it alone and there's actually weekly assignments which are the weekly tutorials. I only submitted them till week 6 and didn't bother anymore for week 7 onwards because I simply couldn't do them/ was too busy with my other 5 modules

2 x 2h lectures (inclusive of a 1h tutorial for the last hour of the week)

Edit: Great, 6 months later then I realised that the tutorial answers are actually re-used from past year tutorials. Can't believe that I really attempted (and failed) to do all the tutorials from scratch (sighs)

Exams difficulty:
His midterms are doable, his finals is probably doable..? Got a shock when he asked us to calculate all the SSR SST SSE, MSR MSE stuff when 100% of his tutorials were just cheem proving questions. Obviously I died there because I forgot the formulas hahahaha

2. DSA4212 Optimisation for large-scale data-driven inference

Lecturer: Prof Alex

Grade expected: B-
Grade Obtained: F (wow)
Module difficulty: 10/10
Webcast: Yes
Hidden/untold pre-requisite (this is just my personal opinion): ST4240

Another super super painful module this semester, the F here caused my cap score to drop so much, so I actually had to go down to school to pay $10 for results appeal/review.

Paying $10 to review your script 
I'll talk about this since I realised that other new module reviewers out there are all too smart/lucky to not kena this shit. Basically, after the release of results, you have to chiong/rush down to school to fill up a form and pay $10/module you wish to appeal or review. The funny thing is that usually it's the smart people (who score A- B+) who are doing the appeals HAHAHA

After losing your $10, you need to wait for an email from the faculty to tell you how's the review going etc. For my case, they just told me that my marks are all correctly counted.. Just wanna let you guys know that actually the $10 that you pay for review is almost as good as a donation to the faculty, because you will never get to see your paper.

"Walao I paid $10 leh, can they at least see my shitty script or not!?"
- No they won't. Told you it's a donation already......

Then you must be wondering "huh, then I pay the $10 for what!?!!?" In return for your $10 kind donation, DSAP/Stats department will send you an email to notify you that your script has been checked and counted correctly, that's all. So far I've never heard of any successful appeal cases.


Alright back to the review:
Initially I took this elective module as a form of enrichment, to find out more about the D word that everyone's going crazy over now - DATA SCIENCE. I'm really interested in pursuing a career in the data science/analytics field, plus I thought that by scoring A- for my CS1010S, this module might not kill me that badly..? So I took it with a few friends. Worst decision ever.

It is the first time that this module is being offered and I've read soooo many good reviews on how Prof Alex is an extremely good lecturer, so I was really pretty excited for this mod. Well he is surely a good lecturer, explains concepts and theorems very well. However, he doesn't return your mid-term scripts, neither does he go through the paper with you. The rationale behind this is probably because the learning process is more important than the results, which I definitely agree. But from a dumb/average student point of view, I actually never got to know what went wrong and what should have been the correct answers instead, so that's quite sad for me.

This module is also very heavy on linear algebra (eigenvalues, eigenvectors, diagonalisation, spectral decomposition, singular value decomposition) and also Hessian Matrix... I need to warn you that the learning curve for this module can be quite steep at the start, so all the best & work very hard.

Mod format:
15% Project
25% - Mid terms
60% Finals
Both exams are closed book no cheatsheet.

I died so bad for this module because of the no cheatsheet regulation. I understand that it's to ensure that we memorise everything but sometimes there's seriously too many formulas to memorise and considering the fact that I have 5 other modules worth of formulas to memorise, man.... I really wanted to die.

Oh and the project/assignments... There's 2 of them in total, but we have no idea what is the weightage for each. They are on really interesting topics such as recommendation model, image recognition (? sorry I forgot already, bad memories).
Basically my group members and I spent a lot, A LOT of time on these 2 assignments. These are the most time consuming stats projects I've ever done in NUS. They're difficult because the prof wouldn't really spoon-feed you/take much about them during lectures, therefore you really need to google your way out. The easiest way to score well for your projects would be to just group with the smartypants/tankers. The tankers will tank the project for you, so you can still score well without knowing any shit about the project (this is very true :))

All the coding is done in Python but we do it on Jupyter notebook/Anaconda instead of IDLE.
The painful part of the project is that one epoch/ one run of the entire assignment code takes 1h+ on average, and if you're like me (who has a very laggy computer), it'll take a few hours... My group mates and I spent days running the code (initially our code took 6 hours to run) I definitely recommend you to run the code on a PC because all our MacBooks took at least 4hrs for one epoch (oh gosh).

Did I regret taking this module?  Actually if I ignore the 'F', I definitely wouldn't regret because Prof Alex is a good lecturer + the theorems and concepts which are being taught in the class are all super useful in real life, if you wish to do data science analytics in the future (logistic regression, matrix/recommendation system, backtracking, singular value decomposition, image blurring etc). But I don't really understand how the F happened considering that I spent the most time on this module but scored the worst. FYI, my other group mates scored C, D+, D+ respectively. I feel that all of us got really bad grades as there probably isn't any bell curve to bolster the 'losers' in the lecture group of 50+ people.

Some other essential things that are taught in this module (not an exhaustive list):
Naive/Stochastic/Proximal Gradient Descent, Backtracking approach, Logistic Regression, Ridge Regression, Penalisation etc

Gradient descent is really important and in fact, it's being taught in Data Mining ST4240.
I personally feel that it really might be better to take 4240 first before taking DSA4212...

Edit: Now that I'm almost done with ST4240, i really advise you guys to take ST4240 first. The speed that Alex goes for ST4240 is so much more manageable than DSA4212. Plus, I feel that DSA4212 is really a much more higher level module of ST4240. It'll be good for you to build your foundation with ST4240 first and not just suicide by diving into DSA4212 straight.
I need to warn you all that the people who are reading DSA are really smart people (class size 50+) but the people who are reading 4240 are more of your 'normal' people...

How to do better for this module: 
1. Revise your linear algebra
2. Go and read up on common optimisation techniques/ideas/examples (logistic regression, gradient descent, backtracking, image recognition, recommendations) I'll recommend you to check out Andrew Ng's Coursera course on 'Machine Learning'. He explains really well and gives you a sneak peek as to what is to come if you ever take DSA4212.
3. If you don't understand what was being taught for that lecture, faster get it settled. Don't let it roll because it's not funny to let your questions accumulate for this module
4. If Prof Alex asks you "have you learnt this before??" during the lecture, and if it's some chim thing that you've really never heard of, please just raise your hand for your own sake. Because if you don't, he will skip through that and go through the rest of the content at lightning speed, the next moment you'll realise that you're screwed for the exams.
5. Go and learn about MLE, he will test it, and he will make sure that he tells you that he will test it (hahahaha but I still couldn't solve it anyway)
6. Pair up with smartypants. There are many ppl who group with them and get really good grades for their assignments. My group mates and I, we are just a bunch of average stats major who struggled really bad with his assignments.
7. Will recommend you to take this module only after taking ST4240 Data Mining (but you can just take directly if you're already well-versed in Machine learning/big data stuff). Prof Alex also teaches the fundamental Machine learning concepts in 4240, but he goes through much slower in 4240 since it's a bigger cohort, so yeah.

Overall: 
I'm extremely traumatised by this module and I wouldn't recommend unless you're a smartypants/ you believe that the learning process is more important than the results. However i believe that I did really badly because of bad timetable planning too. I know there are a few people who have gotten A. You've got this :)! Friendly suggestion: Take ST4240 first.

3. ST3240 Multivariate Statistical Analysis

Lecturer: Prof Loh Wei Liem

Grade expected: B+
Grade Obtained: B (I'm happy enough!!!)
Module difficulty: 3/10
Webcast: Yes

Topics that are taught in this module: Cluster Analysis, Linear Discriminant Analysis, Categorising/Classification

Prof Loh WL is mah man.

He explains everything so well and I'm just super thankful to have him for the sem/mod lah really. He's super chill and lax and he makes me laugh because he's too chill. It's good, we all need a chill slack mod like this. His voice is very soothing toooooooo

Studied for the module in 24 hours time and I got a B. WHATTTTT!?
24 hours because I was really tortured by the other modules. My exam schedule for this sem was thurs, friday, saturday, monday, tuesday, thursday (non-stop)

Tutorial attendance was counted, you just need to write your name.
Group Project was really easy because it's always on the fisher's iris data set (he's so nice!!!)
Finals was really quite easy/manageable. The best part? It's open book and you can bring in as many tutorials and notes as you want (HE'S THE BEST, HE'S MAH MAN!!!)

Frankly speaking, I barely spent much time on this module (energy channelled to the wrong modules - look above). I only did Tutorial 1... (sorry prof ily) but I did manage to rush through all his webcasts /lecture notes within 1 day - damn godlike)

Awesome mod to take, but an even more awesome prof.
If it's being taught by some other profs, I wouldn't advise you to take this module, because the mod can be quite difficult if the lecturer is not mah man.

I'm pretty sure that I would have gotten B+ at least if given more time for this module.

4. ST3233 Applied Time Series Analysis

Lecturer: Vik Gopal

Grade expected: B
Grade Obtained: B-
Module difficulty: 7/10
Webcast: Yes

Super. Freaking. Lucky. That. Vik. Was. Teaching. This. Mod.

I mean, Vik is really probably the best lecturer in DSAP (imo) - so caring, knowledgeable, patient, and speak such good english.....!!?!?!!?!

Anyway, I didn't do that well for this module because I scored below average for midterms.
His marking is really quite strict + the whole idea of doing it on the computer was just so weird so I didn't check the time and ended up not finishing the paper. Finals was manageable I guess.
Mid terms was more on coding/application but finals was purely theoretical so do expect more theoretical stuff, e.g. proving of covariance/independence

Assignments were easy - 2 data camp courses + a 8 pages max report on time series of your interest.
I didn't do well for the report because I only had 4 hours to rush it out (ha ha ha was doing DSA project because it  was due that day too. Oh, did I mention that I actually had 4 assignment deadlines on that day? Don't take 6 cores please)
If you have more time/planned your timetable well, it'll be easy to get B for this module. :-)

It's a useful mod to take because it teaches you random walk (with drift), AR models etc.
But if it's under vik, just. take. it.  (even though my experience with vik is usually a 'B' grade)

5. ST3246 Statistical Models for Actuarial Science

Ah-ha, that rare thing that is only offered every 2 years --> CHIONG AH JUST TAKE

Lecturer: Prof Lim Tiong Wee

Grade expected: D
Grade Obtained: C (holyguacamole)
Module difficulty: 8.5/10
Webcast: No

Prof lim is a good lecturer, he explains things well just that this module is problematic on its own.
Just a word of caution, there is NO webcast for this module.

And frankly speaking, I initially thought that this module would be easy.
Damn, turns out to be really difficult. I couldn't keep up with lectures as there wasn't any webcast. The theorems are really quite difficult to understand within a single sitting. And the best thing is that the lecture content is just constantly jumping around- maybe the first lecture of the week he might be doing pg 20-30 then the second lecture of the week he'll just focus on page 15-20? I don't know man, it was really really hard to keep up with the lectures.

Probably wouldn't score this bad if there's webcasts :-(

C was a huge surprise for me because I thought that I would be failing this module... I couldn't do anything for the finals. Yes I know it's crazy but I really didn't write anything for the first 15mins and felt like leaving after 30mins. Never submitted such an empty script in NUS before so it's really crazy how it's a C and how I scored a F for DSA (LOL).

I couldn't really put in much time for this module as I was too busy with ST4233 and DSA4212. :(

The assignments really saved my life/grade.

Theorems and calculations are too complicated.
Wished that I had time to do the tutorials because frankly speaking, I think I only did one or two tutorials at most.

His finals wasn't very doable, but I think that the person who sat beside me during finals just kept writing and writing... But I know that in general, most people couldn't do it. I have a friend who didn't finish the paper but scored 'A' for this module. HAHA

Will I recommend this module?
Actually, NO. Retreat if possible. Unless you tell me you're really super into actuarial/actuarial is your cup of tea, then I guess it'll be a good module for you to thrash the people who can't be bothered with actuarial (like me). Basically in this module you really just learn about how to do calculations such as expectation cost of an insured holder, how much you should price the premium... Personally, I really just found it boring and unnecessarily hard hahaha

6. ST3236 Stochastic Analysis I

Heard so much bad stuffs about this module that's why I cleared it so late.

Lecturer: Prof Subhro Ghosh (Math department)

Grade expected: C
Grade Obtained: B- (o m g)
Module difficulty: 8.5/10
Webcast: Yes

I'm super lucky to take this module under Subhro, seriously.
It was really bad to take under Prof Ajay (stats department) so I took it under math department - best decision of the sem!!

He really explains everything so well and I can't believe that I can actually understand all the markov chain shit.  He only has a mid term, a finals paper and 4 assignments.

I didn't do well for his midterm because once again I had 3 other midterms to settle zz
But his MT paper is really easy and they're exactly the same as the tutorials/assignments wtfff
His finals paper was a joke because it was almost 95% similar to the mock finals paper that he gave us + went through with us during lecture. I think 80% of the people left the exam venue by the 1.5h mark hahahahah!! Someone even left after 47minutes.

I think it's really great that he emphasises so much on the understanding of pivotal theorems and concepts instead of just blasting us questions on all the super abstract theorems etc. He actually didn't test all the super abstract theorems and I think he only focused on 3-4 theorems.

Definitely highly recommend you to take this mod under him :-)
Do all his tutorials, assignments and mock papers and you'll pass with flying colours.