Hacker News Reader

Comment by acbart | Hacker News Reader

This will never work because it absolutely fails to scale, at a time when we have a complete lack of computing educators. Customized projects for each student? I have 159 students in my CS1 this semester, and I can barely keep track of them with the massive amounts of data I collect. The idea of making individualized assignments is beautiful but unrealistic.

View on HN

Most code reviews I've done at work take between 5 minutes to 10 minutes. To assume the worst possible case of 200 students, 20 deliverable per student, and a 10 minutes meeting per MR you'd be at ~28 days (~84 8hr-days). Get 1 TA and you now have 42 days of the semester dedicated to 1:1 time (35%) and the remainder of the semester dedicated to producing learning content. Learning content (lecture videos, etc) can be produced ahead of time and in this case you can replace this with more 1:1 tutoring time effectively allowing you to give each student ~10hr (10 min * 20 sessions * 1/.35) direct 1:1 time over the course.

From the students perspective all they need to do is:

1. Write a "project proposal" where they define a project, define goals, schedule a checkin with you to see if that project is large enough (1/20th the grade of the class)

2. Watch lecture videos and go through exercises on their own time.

3. Write code and bring it in to teacher. (18/20th of their grade)

4. Demo their finished project to professor (1/20th their grade)

This does work for later classes. I never had the ability to test this workflow out for earlier (100-level) classes as all of those classes I TA'd for followed the same model you're talking about.

The 200+ level classes that followed the model I'm talking about had ~10% to 30% pass rate which was in line (~2x) with the pass rate of similarly leveled courses from our college's Math and Physics department which was used as a sanity check.

edit: that's not to say you're incorrect that it would be difficult to do at scale, this is just the way I've seen it done at moderate scale (20 to 40 students).

Code reviews I have done for unrelated bits of code often take me 90 minutes. And that's for code that fits into the idea of my product space, but not super familiar to my everyday life. I don't think 5-10 minutes is anywhere close to reasonable.

The hard part is giving the student meaningful feedback. Especially to those who don't understand something important. It takes time to understand what they do and don't understand, and explain it at their level.

It's easy (even a bit fun) to grade and give feedback for good students. Meanwhile it's a huge pain to do for the weak students. You need to first figure out what they heck they're trying to do before you can tell them how to fix it, which may take a lot. I suspect code reviews for professionals is much more like the good student category.

If you're asking the students to define unique projects themselves, you run into new problems as students dig up the most obscure blogs they can find on the internet to download and give you. It's a never-ending adversarial struggle.

You'll be surprised at what "weak" code you'll see in the real world and have to explain to someone it's not maintainable or clean or whatever.

This is especially true of junior data scientists (in my anecdotal experience so far)

I teach classes of ~500 students and classes of ~100 students.

While having individual projects is very nice and it works for smaller classes, it does not scale to average students.

I have once used individual projects to grade the ~100-student class and the amount of effort it takes is unbelievable.

For example, just your number "1." point, write a project proposal, is not so easy. It takes hours to write a good proposal and several back-and-forths between instructor and student to agree on a reasonable (not too advanced/not too basic) project.

10 to 30% pass rate? And that was very high compared to other departments?

Do you mind sharing roughly where this was (country and quality of university)? I thought we were in an era of grade inflation where everyone gets an A.

This was course a 300-level test course compared to a 300-level test course from these other departments. The hope is that when they become "real" courses the pass rate is higher. I think this counts people who are re-taking the course or who dropped it as failing. We had ~50% of our class drop within the first month.

Part of it was related to their expectations of the course: "you mean I have to write code and the class is based off of that and not memorizing test answers? I'm out"

I think it can be automated in the more basic courses at least. But wouldn't solve the problem of students getting someone else to do the project for them anyway.

So much this: cs61a at Berkeley for example has almost 2000 students in a single class, the idea of personalized anything with that many students is total unfeasible.

I thought in the US people paid money for going to university, so that classes are well-funded enough... /sarcasm

Berkeley is a public university, so many students qualify for federal aid and wind up paying very little or nothing at all,the full cost of in state tuition to those who can afford it is about $6000 per semester.

This is ridiculous. Why do Americans pay thousands of dollars for a bit fancier version of MOOC? When I studied we had 200 people on a course sitting on lectures, 30 on practical sessions and everyone could have a personal session during the professor’s office hours.

There is some kind of a massive waste of money going on.

Reply on news.ycombinator.com