-
Notifications
You must be signed in to change notification settings - Fork 100
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
New question type request: numfunc n-tuple #271
Comments
Is there a reason that this approach would be significantly preferable over a multipart question where they input each component separately? |
By analogy, the current n-tuple and calculated n-tuple are all arguably preferable their multipart counterparts. From another perspective, I inspected the current atom question types and they fall into one of the following categories:
In this second category, numfunc is best suited to receive n-tuple support. I am not sure if complex and calccomplex should receive n-tuple support, but the rest in the second category probably do not benefit from n-tuple support at all and are best served by multipart. However, I fail to think of any pedagogical reasons why numfunc n-tuples would benefit students other than that a potential interface for question input looks nicer and more in-line with calculated n-tuples and numfunc. In the course that I am teaching now, the students seem to be equally fine with the multipart and the conditional versions. Edit: On the level of question design, official support for numfunc n-tuple would make the grading of parametric equations and vector fields more efficient and consistent across the question pool. Many options ($variable, $domain, $displayformat, $scoremethod) available to numfunc and calculated could also be made available to numfunc n-tuples automatically. But, of course, this question type has limited use outside of scope of vector-valued functions. |
Not sure what the proper procedure for feature request is, but I request a new type of question, the numfunc n-tuple. which is like an n-tuple or calculated n-tuple, but with numfunc entries.
Having coded questions whose answers are n-tuples of numfuncs, I see that this type of question could be very helpful in multivariable calculus. Example questions are gradients of functions, e.g., f(x,y,z)=xyz, where students have to respond with a vector of 3 numfuncs, each being yz, zx, and x*y. The current way in which I am coding this type of question is to use conditional questions together with some answer-parsing code can split student responses enclosed in single or double brackets (round or angle) into an array of strings.
If there is developer documentation, I would like to try to submit some commits. I am a bit short on time, so here is only my feature request.
The text was updated successfully, but these errors were encountered: