ML (CS7641) - An OMSCS Review
Like I mentioned here about how I went wild with two courses this semester, I’ll talk about the heavier one in this article. Oh boy, ML CS7641 is indeed very heavy! All the fear that you hear people talk about in OMSCentral is indeed valid. It is definitely not an easy course.
The answer to the usual “why I took this course?” is obvious at-least for people who know me - I dabble in machine learning and I want to do the ML specialization. Before I go into the details, let me start with something positive by saying - this is a very doable course. If you have some ML background, then this course should be an easy A if you put the hard work in. Yes, it is a time consuming and hard-work demanding course, even if you are in this field, to get an A. I had fun this semester with multiple trips with friends and family, went to watch the FIFA World Cup, had good social weekends and still landed with an A - hope this motivates you before you read the rest of the blog.
Firstly, the lectures - decent material, but with a lot of humor (good humor or bad, I’ll let you guys decide after you take the course). The lectures do go into some depth and breadth. Some topics are pretty detailed with mathematical proofs. The lectures are useful when it comes to the exams. They also subtly set you up for the projects. I’d recommend going through the lectures at-least once with good concentration.
I think this is the first time I might have actually gone into the readings suggested by the course. Very much required. I found myself going through the famous textbook by Tom Mitchell. The other readings that you can find in the files section is pure gold. It’s a lot of literature and the syllabus clearly tells you what to read and when. They were solid reads that prepped me up significantly for the projects and exams.
The projects hold the maximum weightage, and there are 4 of them, each covering one significant portion of the syllabus. Think of each as a mini research track in itself - you research some datasets, you run experiments on them, you compare and contrast each and everything you can find, you analyze them and write a 10 pager. Do this 4 times and you are pretty much done with the semester. You start off by picking 2 datasets. Spend time on this, you will be stuck with that dataset for the remainder of the course and changing it will be a mess. The experiments you do later in the course depend on how well-picked the datasets are. So do that well. You get approximately 1 month for each project, and you need one month. You CANNOT miss a single week, well maybe you can (I did miss a week for two projects and still did well). But don’t live dangerously like me, I’m sure I could have done better if I had that extra week. The experiments that you run take time, a lot of time, especially in project 2, and you won’t even know what experiments to run and what to analyze unless you start doing it. Hence the emphasis on the “start early”.
Office-hours is A MUST if you want to do good in the projects. The project page does not give you a lot of details - very open ended problem statements and you won’t know what the TAs grade you for. There’s no official rubric. The only place where you’ll get hints and understand what the TAs will look for are the office hours - watch the recordings!
Exams aren’t easy. Midterm is long and short. Long on questions, short on time. They say it is intentional. No MCQs. All are long form answer questions where you have reason, justify or explain something. The answers are strictly graded. The final exam is short on questions and long on time so usually students do better here.
The course is curved, so expect grading to reflect that. There’s a good chance you simply have a 60% aggregate and still land up with an A (focus on the word chance). Don’t fret if you don’t see marks in the 80s or 90s - high chance that you may not (I’ll be happier if I’m wrong) just do your best to stay above the average and you should do good.