Proposal: Outsource installation of packages (models, data plugins) - WARNING - breaking API change#346
Proposal: Outsource installation of packages (models, data plugins) - WARNING - breaking API change#346edwardchalstrey1 wants to merge 6 commits intomainfrom
Conversation
|
I think I'd be sad if this feature went away, esp. if we're targetting less experienced users. I'd rather have the info message (i.e. this PR) as a fallback if the package can't be installed automatically. |
@quantumjot that's understandable and it's the reason you added this feature to begin with - I think this somewhat comes at the expense of things being more likely to go wrong with installation where we try to automate it (due to dependencies etc) - my alternative idea for less experienced users which makes use of @ots22 UI feature: #325 |
|
If we did merge this, we could also acknowledge the breaking change via switching to version 1.0.0. https://semver.org/ |
Closes #253 (see discussion on this issue)
Idea here is to circumvent a lot of the problems we have around package installation for models in the catalog (and more recently data plugins) by completely removing this as a feature of scivision. This branch does that, and I've created a branch on the
scivision_examplesrepo with the same name that updates the intro (Koala) notebook (see here) - take a look under the "Loading the model" header.Essentially we raise a warning if the package is not installed that explains how to install it (we already had this warning in place in fact!).
I want to spark some discussion here, so feel free to voice your disagreements in the comments below!
TODO:
Further idea
Perhaps there should be a function like:
get_install_code('scivision_classifier')which simply prints out:
... or alternatively links to the package repo README for packages that are more complex to install