`parquet_url` incorrectly provided for non-arff formats
Opened this issue · 0 comments
PGijsbers commented
As you can see here, anything that is not 'Sparse_ARFF'
gets a parquet_url
. This is incorrect, see e.g., www.openml.org/d/1243 which has Rimage
data format (not supported).
The check should probably be changed to == "ARFF"
instead based on the list of stored formats (the test server also sports additional formats, including lower-case 'sparse_arff'
):
Went ahead and labeled this wontfix
since API V2 is on the way and it doesn't really impact people (I imagine all datasets that are not ARFF never got out of processing anyway). It's a minor fix though so be my guest.
We might also consider deactivating any of the non-arff datasets (as part of the dataset clean up).