Adapter from multiindex to ListDataset #2976
Conversation
|
@fkiraly, I added the unit test to the end of sktime/datatypes/tests/test_panel_converters.py I'm having some |
|
Hi Tan!
__all__ = ["convert_from_multiindex_to_listdataset"]
from sktime.datatypes._adapter.pd_multiindex_to_list_dataset import convert_from_multiindex_to_listdataset |
I see. But can you explain what is the reason for having |
|
@fkiraly In my local machine, I have a failure at |
@fkiraly For some reasons, the tests are now finished. Both the adapter and unit tests are ready. |
|
excellent! Will look & probably merge after the 0.13.0 release. |
fkiraly
left a comment
There was a problem hiding this comment.
Looks great! I would merge if not for what looks like an accidental mistake.
I think there is a commit/merge accident, there are two files pd_multiindex_to_listdataset, one without py extension and one with. I suspect you want only the one with.
I am not going to delete it though, since the content seems different - could you kindly have a look?
My bad. The one without |
|
Added back conversion (unit test in progress) |
There was a problem hiding this comment.
Nice! wasn't expecting the back-converesion, just wanted to merge.
I suppose the back-conversion is native to nested_univ, so a clean version has that as a return type and does further conversions depending on the desired result type.
Either way, I think this is good to go in in order to unblock @AurumnPegasus, we can refactor and add more unit tests later.
|
FYI @AurumnPegasus |
Reference Issues/PRs
Partial solution to Issue #2860
Continuation from PR# 2893
What does this implement/fix? Explain your changes.
Implement an adapter to convert multiindex format to ListDataset in gluon-ts
The method takes as input a pd-multiindex DataFrame, (optional) a list of categorical feature, (optional) the startdate and (optional) frequency of the dataset.
Does your contribution introduce a new dependency? If yes, which one?
Dependency: gluon-ts
What should a reviewer concentrate their feedback on?
Testing with different dataset in sktime to weed out the bugs. So far, Arrowhead, ItalyPoowerDemand and StandWalkJump have been tested.
Any other comments?
PR checklist
For all contributions
For new estimators