You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
…orkflow (#1429)
closes#1426
It appears there was a change in 24.08 that broke a notebook demonstrating a `merge` on two geometry columns. It seems like the merge result tries to reconstruct a `GeoDataFrame` from a `dict[Any, GeoSeries | Series]` but the `Series.index` alignment requires the types to be recognized cudf types (not `"geometry"`)
I don't think this alignment is entirely necessary though since it goes through the `_split_out_geometry_columns`/`_recombine_columns` methods which appears to be used on operations that maintain row ordering so index alignment isn't required.
This PR instead passes a `dict[Any, GeoColumn | Column]` to `cudf.DataFrame._from_data` given that this row ordering is preserved.
(This PR also includes the fix for #1427)
Authors:
- Matthew Roeschke (https://github.com/mroeschke)
Approvers:
- Mark Harris (https://github.com/harrism)
URL: #1429
Version
24.10
On which installation method(s) does this occur?
No response
Describe the issue
On 24.10, several unit tests are failing:
(build link)
Minimum reproducible example
No response
Relevant log output
No response
Environment details
No response
Other/Misc.
Not observing this on 24.08, so I believe it's 24.10-specific.
This is happening for both wheels and conda packages, across all combinations of CPU architecture, Python version, and CUDA version tested.
The text was updated successfully, but these errors were encountered: