Repository navigation
Extract parameter and model description from scikit-learn docstrings #175
Description
and store in the description of the flow or the description of the parameters.
{{ message }}
and store in the description of the flow or the description of the parameters.
Activity
PGijsbers commentedon Apr 4, 2018
Comment for reference of whoever is interested in picking up this issue.
As far as I can tell this is not resolved yet, but most work has been done (?).
Parameter meta information can be passed to OpenMLFlow, so I think the goal is to pass this information correctly when creating a flow from a sklearn model. To me it looks like the right place to start this is here.
Neeratyoy commentedon Jul 25, 2019
@mfeurer it seems that if the str length of a flow description is > 1024, we get an
OpenMLServerException: Problem validating uploaded description fileIs that expected?
Steps to recreate:
task = openml.tasks.get_task(403)clf = sklearn.linear_model.SGDClassifier()r, f = openml.runs.run_model_on_task(clf, task, return_flow=True)dummy_str = '-' * 1024f.description = dummy_strf = f.publish() # this worksopenml.utils._delete_entity('flow', f.flow_id)f.flow_id = Nonedummy_str = '-' * 1025f.description = dummy_strf = f.publish() # this does not workNow w.r.t. this feature request, will not the flow description become excessively large given that sklearn has a good 2-3 line description for each parameter (roughly on an average). This would also entail changing the
__repr__for OpenMLFlow.What do you suggest?
mfeurer commentedon Jul 26, 2019
I'd suggest to store the first 1024 characters of the docstring on OpenML (but always stopping before the "Read more here" or "Parameters"). I don't see a real issue with repr for now. We can change that if this becomes a real issue for users?
mfeurer commentedon Sep 13, 2019
Added via #756