FIX-#3571: fallback to 'sort=True' on categorical keys in groupby #3715
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
…n groupby Signed-off-by: Dmitry Chigarev <dmitry.chigarev@intel.com>
dchigarev
commented
Nov 19, 2021
| @@ -2894,7 +2894,7 @@ def value_counts( | |||
| ): | |||
| if subset is None: | |||
| subset = self._query_compiler.columns | |||
| counted_values = self.groupby(by=subset, sort=False, dropna=dropna).size() | |||
| counted_values = self.groupby(by=subset, dropna=dropna, observed=True).size() | |||
There was a problem hiding this comment.
sort=False -> sort=Trueso the users won't see the warning thrown by groupbyobserved=False -> observed=Truebecause newly added tests forDataFrame.value_countsshowed that it's have to be set (see the example):
Example
>>> df = pandas.DataFrame({"a": [1, 1, 2, 2], "b": [2, 2, 3, 3]}, dtype="category")
>>> df
a b
0 1 2
1 1 2
2 2 3
3 2 3
>>> df.value_counts()
a b
1 2 2
2 3 2
dtype: int64
>>> df.groupby(["a", "b"]).size()
a b
1 2 2
3 0
2 2 0
3 2
dtype: int64
>>> df.groupby(["a", "b"], observed=True).size()
a b
1 2 2
2 3 2
dtype: int64
dchigarev
commented
Nov 19, 2021
Comment on lines
+1726
to
+1728
| by = np.array(["a"] * 50000 + ["b"] * 10000 + ["c"] * 1000) | ||
| random_state = np.random.RandomState(seed=42) | ||
| random_state.shuffle(by) |
There was a problem hiding this comment.
the bug is partitioning dependent, that's why I'm carrying the same data all around
There was a problem hiding this comment.
To clarify - you mean the shuffle is in place so that the bug gets tripped (bc if we didn't shuffle and the values weren't replicated across the partitions, we wouldn't hit the bug?)
There was a problem hiding this comment.
Right. For this data shuffled with this specific seed we know for sure that there is a problem caused by the incorrect order of categories in the reduce phase of groupby (see the issue explanation).
This problem might not be visible when the values are in a different order, so I'm using exact data from the #3571 report as a solid reproducer of the issue.
dchigarev
commented
Nov 19, 2021
Comment on lines
+294
to
+296
| # The bug only occurs in the case of Categorical 'by', so we might want to check whether any of | ||
| # the 'by' dtypes is Categorical before going into this branch, however triggering 'dtypes' | ||
| # computation if they're not computed may take time, so we don't do it |
There was a problem hiding this comment.
modin_frame._dtypes attribute is None most of the time, its computation requires calling MapReduce and actual data materialization. I don't think we want to trigger this on every groupby(sort=False)
Codecov Report
@@ Coverage Diff @@
## master #3715 +/- ##
==========================================
+ Coverage 85.47% 88.67% +3.19%
==========================================
Files 197 197
Lines 16341 16345 +4
==========================================
+ Hits 13968 14494 +526
+ Misses 2373 1851 -522
Continue to review full report at Codecov.
|
|
|
|
RehanSD
reviewed
Nov 19, 2021
vnlitvinov
previously approved these changes
Nov 25, 2021
Signed-off-by: Dmitry Chigarev <dmitry.chigarev@intel.com>
vnlitvinov
approved these changes
Nov 25, 2021
|
@RehanSD do you have any more comments? |
|
Looks good to me! |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
Signed-off-by: Dmitry Chigarev dmitry.chigarev@intel.com
What do these changes do?
flake8 modin/ asv_bench/benchmarks scripts/doc_checker.pyblack --check modin/ asv_bench/benchmarks scripts/doc_checker.pygit commit -ssort=Falseparameter and categoricalbyproduces incorrect aggregation results #3571docs/developer/architecture.rstis up-to-dateSince this PR doesn't fix the actual problem but just hiding it, we should keep the #3571 open after the PR is merged.