Update copy analysis result and experiment data behavior #1432
Conversation
coruscating
added
the
backport stable potential
coruscating
force-pushed
the
fix-copy-analysis-results
branch
from
cb2091b
to
3c26036
Compare
| @@ -204,16 +204,26 @@ def clear(self): | |||
| with self._lock: | |||
| self._data = pd.DataFrame(columns=self.DEFAULT_COLUMNS) | |||
|
|
|||
| def copy(self): | |||
| def copy(self, new_ids: bool = True): | |||
There was a problem hiding this comment.
Nothing uses new_ids=False, so should we just not include that option for now? It's a question of why a user is calling copy. If making a copy, it seems likely the user would want to change something and then in that case likely that new IDs should be used?
| new_ids: Whether to generate new IDs for copied entries. Defaults to True. | ||
|
|
||
| Returns: | ||
| A new shallow copied DataFrame object. |
There was a problem hiding this comment.
I don't like the use of DataFrame here because the DataFrame is an internal container of the data not directly exposed to the user. Also, maybe the point about being shallow is better in a .. note:: block like it was and here it could just say A copy of the analysis result table? I think the original note was not clear enough. The default columns of the data frame are all strings and numbers which are all immutable so the only difference between being shallow and deep is saving some memory. However, I think it's possible that the user could add a column with mutable objects? The concern there is that then mutating the objects in the copied table would also mutate the ones in the original. Maybe this is a pretty minor concern -- would a user add mutable objects to an analysis result table?
| exp_data.add_data(self._get_job_result(1)) | ||
| copied = exp_data.copy(copy_results=False) | ||
| self.assertEqual(exp_data.data(), copied.data()) | ||
| self.assertFalse(copied.analysis_results()) | ||
| self.assertEqual(exp_data.provider, copied.provider) | ||
|
|
||
| def test_copy_analysis_results(self): |
There was a problem hiding this comment.
This is good. Should there be a test of copy in test_analysis_results_table.py as well?
Summary
Fixes #1396. When an
ExperimentDataobject was copied, it didn't update the analysis result IDs and their associated experiment IDs, resulting in the copied experiment entry not having analysis results in the cloud service.Details and comments
AnalysisResultTableisn't very efficient because it has to loop through the rows and call_create_unique_hash()for each ID. Maybe this logic can be refactored.