My Review of (ISYE 7406) Data Mining and Statistical Learning

           Grade: A
      Difficulty: 4/10
          Rating: 3/10
 Time commitment: 6 hours/week
-------------------------------

Overall

This is a sister course to ISYE 6740 CDA. While CDA is about theory, this course (ISYE 7406) focuses on application. The course covers KNN, linear regression, regularization, LDA, smoothing kernel techniques, tree based model, SVM, clustering and neural networks.

I took this course during summer semester. The pacing felt manageable. They reduced the number of homeworks from 6 in regular semester to 4 in summer.

Lecture Quality

• Lecture video is ok but the professor mostly regurgitates the slides. He tends to go over math formula and notations quickly without explaining individual terms so you may sometimes have to pause and explore on your own.

Assignments

• Homework (25%) :
  • There are 4 homework assignments, usually due every 2 weeks. Every homework is a mini data analysis project where you go through the whole process of EDA (exploratory data analysis), methodology discussion, modeling, analysis of results, so on. And you submit one big (~10 page) report, with the code attached in the appendix.
  • It felt a bit like more advanced version of 6501 weekly hw reports.
  • The focus is not code, not theory, but about analysis.
• Quiz (20%) :
  • There are 4 quizzes. Not proctored but timed (40 minutes). Each quiz is 5% of the course grade and consists of just several questions. Some are multiple choice questions while others are simple calculation questions that ask you to type in numbers into a blank textbox.
  • I found this stressful because I got a few questions wrong on technicalities under time pressure. For example, (I just make up an example here so I don't give away actual exact detail nor answers) I typed "1.234" when the question asked for the fourth decimal place, so I was expected to type "1.2340" -- it's unforgiving on these kinds of details. No partial credit. And one mistake like this costs ~1% of the course grade each time. It was stressful. No execuse. Everybody is playing the same game under the same rule. But it's just stressful :-)
• Project (25%) :
  • Just like CDA, the project is very much open ended. You can propose and work on any topic. But the focus is deep data analytics using ML models. I picked a famous census income dataset from UCI (University of California Irvine) ML repository, and compared a few ML algorithms for predicing income level based on individual attributes. Many people used kaggle datasets.
  • It's called group project but they allow a solo option. I did solo. Looking at the Canvas roster, it appeared about 80% of the class did solo projects.
• Peer Review (5%) :
  • There is mandatory peer review for every homework and project report. You have to anonymously peer review 3 class mates. And you also get their anonymous feedback in return. But the beauty of this is that the peer reviews are purely for feedback purpose and not part of the assignment grading. Only TA reviews and assigns score for your homework and project reports.
  • In my opinion, this course implemented peer review correctly. It's just for learning from each other and learning to give constructive feedback to each other. It's great. 5% of the course grade comes from giving thoughtful and detailed feedback.
• Final Exam (25%) :
  • The final exam is just like another homework. It's take-home format, and you have one week to analyze a given dataset and write a research report. So literally there is no difference from homework assignments. And yet it's worth 25% of the course grade.

Grading

• Homework: All manually graded by TA. The grading standard is somewhat random. My average HW score was 93%. The median class score was around 92%. The only tricky aspect is that the open-ended nature of homework report requirement which essentially just says "conduct a thorough end-to-end ML analysis". So the grading by TA can (unavoidably in a sense) feel random and subjective. Often TA took aways points for "you could've discussed more details on XYZ" (duh).
• Project: Manually graded by TA. The feedback from TA was extremely verbose and appeared to be AI-generated. Basically they gave me a massive laundry list of what's wrong and missing in my report. But I still got 100%.
• Quiz: Auto-graded. This is supposed to be easy but because of tiny human errors I mentioned previously, it got harder for me. I got 90%.
• Final Exam: Manually graded by TA. I got 100%. I guess they didn't have much time to grade it as it was the end of semester. The only feedback I got was basically "good job !".
• Peer Review: As long as you complete all your peer review tasks (there are 7 of them), you get 100%. I got 100%. I know many students forgot to do peer reviews for some of the homework & project reports and lost these easy points. You may think it's a rookie mistake but peer reviews are so forgettable. There is often a gap between the time you submit your report and the time you can do peer reviews. There is no separate item on Canvas module interface reminding students. It's kind of buried (hidden) in the Canvas interface. I almost forgot 2 out of 7 peer reviews. phew..
• Overall, no grading curve. They apply the usual 90/80/70% thresholds for A/B/C/D letter grade assignment. I earned 96% overall (a letter grade A).

Thoughts

• It's a decent course. The lecture quality is meh but the homework assignments are really hands-on and rewarding.
• I appreciate the course places an emphasis on learning to conduct rigorous data analysis and writing an academic report. It's an important practical skill set. But this super open-ended style of homework assignments does not vibe well for some students who may find this frustrating because there is no one correct answer. You may sink endless hours on your reports and can never feel sure whether you did well or not. They might prefer more specific and theoretically-focused homework (like the ones you see in ISYE 6740 CDA).
• It's not really a problem with this particular course but AI can do every homework instantly and perfectly. So the course staff really should re-think the homework format at this point.
• It's not a huge problem but often the instructions (for homework, project, exam) are somewhat poorly written with clear grammatical mistakes and occasional unusually casual tone (plus wording) indicating non-native English speaker wrote them. It's immediately noticeable once you read them. That's fine by me but the fact that nobody (between professor and an army of TAs including many native English speaker) bothered to fix them kind of shows the quality of operation of the instructor team. I say this because I'm a TA for other courses in the program and we carefully review and improve these things.

Course administration

• TAs and professor actively host office hours and very helpful on Piazza forum.
• I must give credits to the course instructor team for allowing solo option for group project and also making peer reviews not count for grade. Group project is the best part of the OMSA program for some students but also the worst nightmare experience for many students. By allowing a solo option, they accommodate students from both camps. Also peer review is often a disaster when it counts for grade while some argue it's a great opportunity to learn from each other and learn to give constructive critique to others work. This course implmeneted it perfectly. 5% of your grade comes from giving good detailed reviews to others, but others' peer reviews do not affect your grade. It's just perfectly done.
• Unsurprisingly, there were a few occasions in the semester TAs announced they detected AI usages in homework submissions. And TAs would additionally host oral exam to test some of the students suspected of using AI at the end of the semester.

FAQ

• [Prereq/preparation] How much math or coding do I need to review to prepare for this course ?
  • There is not much to it. If you survived ISYE 6501, that's enough prep for this course. They give starter template code. And math is never directly tested.
• Is the course assignment based on R programming language or Python ?
  • They give you sample code in R, but let you use any programming language. I used Python, but you can use literally anything (e.g. R, Python, Matlab, SAS, Java, etc)
• How hard is it to achieve letter grade A in this course ?
  • It's not impossible but there is an element of randomness in grading for the 4 homework assignments. But for the quiz, you can easily get 90% average. The project grading is also random. It's a bit like 6501 HW grading where you don't know whether you get 90% or 100%, it's random.
• Is it worth taking both DMSL and CDA ?
  • Yes, I think they complement each other well. While the individual ML topics may be overlapping, both courses focus on different aspects, so you will still learn a lot. CDA is like "let's focus on the convergence property of kmeans clustering algorithm" while DMSL is like "let's apply kmeans clustering to this dataset and analyze the results".
• If you have to choose one of DMSL vs CDA, which one do you recommend ?
  • CDA is objectively a better quality course, but it depends on your interest. If you want theoretical foundation, then CDA. If you want practice application (conducting data analysis project and writing reports), then DMSL. I personally found CDA a lot more fun and intellectually stimulating.

Resource

• Feel free to ping me for questions.

Reference

• Syllabus : You can read the syllabus at https://omsa.wiki