Are there examples, best practices, or companion software for constructing an assignment from a question bank?

[moorepants]

After years of teaching a course you build a bank of questions, often these are variations or prior questions. You can cycle questions and their variations in exams and assignments over time if you track carefully the date used and differences in question variations.

I've used nbgrader for enough years now for the same courses that I'd really like to do something like make a single text file per question with metadata that includes date used and variation grouping and then run a command to assemble a new assignment notebook from new and old questions.

Has anyone already tackled and solved this problem? I haven't found anything yet with my internet searching, but maybe it exists as it seems like a common thing done in pen & paper assignments and exams.

If anything exists like this, I'd love some pointers to it. And if not, I'm interested in ideas on this. I think I will implement it if it doesn't exist.

[perllaghu]

I'm not aware of one.....

.... and can I clarify that you are looking to create a semi-random source notebook that would be generated & released in the normal way, rather than some system that would somehow distribute random different notebooks to different students?

The former seems achievable... the latter would, I think, break nbgrader in so many ways [or be a really really complicated setup involving different courses for different students, and all the complications that would entail for marking]

[moorepants]

I'm only seeking to generate a single assignment notebook (same one given to all students in that course offering) that is constructed from multiple files that each have a single question (and its grading tests, etc.).

Right now I could make a single question per notebook and then I could open a new notebook and drag the cells into the new notebook to assemble my new assignment. I could manually keep track of what year the question was last used and what all the variations of that question exist (a question might be associated with testing a single learning objective, for example). But...

I'm addicted to automating things (like many of us), so having a command assemble assignment p1_from_2024_v1,ipynb p2_from_1982_v1.ipynb -o assignment_2025.ipynb would be my desired workflow. And if p_from_2024_v1 was a markdown file instead of a notebook, that would allow me to edit plain text instead of having many notebooks open. So, plaintext would be icing on the cake.

[perllaghu]

I original thought was to havesomething like assemble --count 12 questions.ipynb - which would pull 12 random cells from the file questions.ipynb (which would have all the questions in it)

However I realised there are a whole pile of complications... like each "question" would need to be whole & complete in it's own right - which would exclude matched autograde answers from autograde tests; and like there'd be no way to include cells with instructions and/or guideance.

My next thought was to have assemble --count 12 questions_dir - where the directory questions_dir would have a large number of small ipynb files, each containing a set of cells that would count as a set [so instructions + answers + tests as needed], and the assemble would randomly concatenate 12 into a new file. I'm not sure how unique cell-ids are going to be, so there might be clashes there... and there's bound to be something else I've missed.

.... but I'm sure it would be doable --- I just don't know if anyone has [yet]

[moorepants]

I just want to do my research to see if there is prior art. But I think I will work on this regardless. If it would fit with nbgrader, I am open to contributing it.

[perllaghu]

If you can create something..... I think it could well be used beyond nbgrader.

.... however I can absolutely see a use for this sort of tool!

[moorepants]

Cool. I'll update here as we make progress. We plan to try to make something from now until July as I teach the course I need this for.