Open mdhaber opened 2 days ago
I'm basically happy with whatever the array API maintainers would find useful w.r.t. this.
I think anyone who is going to do that will be open source on GitHub anyways. We can use GitHub code search or maybe the API to do this. We've already done this sort of thing in the past to inform array API decisions https://github.com/data-apis/python-record-api
Right, and I've done that before, too. It might still be useful for users to report this sort of thing:
array-api-extra
functions instead of directly using array-api
functions. (However, we could get this by reaching out and asking.)So it's not information that we can't get otherwise, but it offloads a little work from us and gives users a mechanism for less formal feedback than an RFC at data-apis/array-api.
I don't see any issues with asking people to report issues or feedback in the documentation. That's something I usually do anyways. I'm -1 to any sort of actual "analytics".
Sounds good. Changed the title accordingly.
I'd suggest that we implement a way of collecting information about usage of
array-api-extra
to inform changes/additions to the array API standard. This doesn't have to be automatic; it could be as simple as documenting a procedure for users to comment on what they are using a given function for and why the existing standard was not enough for them.The simplest thing I can think of would be to include a standard note in function documentation requesting that users leave a note somewhere (e.g. an issue that corresponds with each function) or fill out a form. Perhaps this could get more sophisticated in time, but maybe start with something as simple as this?
My guess is that you'd get more comments like https://github.com/data-apis/array-api-extra/pull/7#issuecomment-2371828099 (which points out that having
expand_dims
accept more than oneaxis
is useful and not complicated for any backend to support). Comments in support of features and example use cases are helpful when it's time to file an issue atdata-apis/array-api
.