Closed charlesbluca closed 7 months ago
Update: This is no longer an issue in the latest branch-24.06
checkout and the dtypes
are unchanged after a parquet round trip. Using the same python script by @charlesbluca and adding extra prints we get:
Input: datetime64[us] datetime64[us]
Output: datetime64[us] datetime64[us]
@charlesbluca could you also please verify at your end that this is fixed now. Thank you!
Describe the bug When reading in a parquet table containing datetime columns using
dask_cudf.read_parquet
, the datetime columns have mismatched precisions between the Dask object's meta and the computed frame.Steps/Code to reproduce bug
Expected behavior I would expect the precision of the datetimes to match between the Dask meta and computed object.
Environment overview (please complete the following information)
Environment details
Click here to see environment details