FIX-#4060: Fix _to_pandas() dtypes for empty frames. #4108
Closed
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
Signed-off-by: mvashishtha <mahesh@ponder.io>
mvashishtha
force-pushed
the
4060-fix-mixed-dtype
branch
from
c71b8f5
to
8312829
Compare
Signed-off-by: mvashishtha <mahesh@ponder.io>
Codecov Report
@@ Coverage Diff @@
## master #4108 +/- ##
==========================================
+ Coverage 85.61% 88.56% +2.94%
==========================================
Files 201 201
Lines 16852 16851 -1
==========================================
+ Hits 14428 14924 +496
+ Misses 2424 1927 -497
Continue to review full report at Codecov.
|
|
|
|
|
The unit tests revealed real bugs, so I'm trying to fix them. |
Signed-off-by: mvashishtha <mahesh@ponder.io>
dchigarev
reviewed
Jan 31, 2022
Comment on lines
+2425
to
+2428
| dtypes=op( | ||
| pandas.DataFrame(index=self.index, columns=self.columns), | ||
| pandas.DataFrame(index=right_frame.index, columns=right_frame.columns), | ||
| ).dtypes, |
There was a problem hiding this comment.
Besides the fact that we're executing UDF op twice (which could be really expensive), this is still a very unreliable construction. The output type isn't based on the source frames types at all, it always gives the same type in the result here.
Example
# Frames holding integer columns
>>> df1_i["a"].dtypes, df2_i["a"].dtypes
(dtype('int64'), dtype('int64'))
>>> (df1_i + df2_i).dtypes # int + int = int
a int64
dtype: object
# Frames holding strings:
>>> df1_s["a"].dtypes, df2_s["a"].dtypes
(dtype('O'), dtype('O'))
>>> (df1_s + df2_s).dtypes # string + string = string
a object
dtype: object
# trying to infer dtypes with the present logic:
# int + int = object -> wrong
>>> (pandas.DataFrame(index=df1_i.index, columns=df1_i.columns) + pandas.DataFrame(index=df2_i.index, columns=df2_i.columns)).dtypes
a object
dtype: object
>>> (pandas.DataFrame(index=df1_s.index, columns=df1_s.columns) + pandas.DataFrame(index=df2_s.index, columns=df2_s.columns)).dtypes
a object
dtype: objectAs far as I know, we don't have support for empty frames at all, that's why handling empty cases are very tricky. They should probably go as default to pandas or even be ignored if the mismatch isn't critical.
There was a problem hiding this comment.
@dchigarev you make a good point, however it is not so simple.
A common pattern we see is creating an empty dataframe and using append to add to it. There are lots of other usecases for empty dataframes even though I do not like them. Ultimately, we need a better way to handle them (maybe a new class?) but we can't just hand everything off to pandas because we don't get the dataframe back.
8 tasks
|
#4060 is closed, @mvashishtha feel free to reopen if needed |
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: mvashishtha mahesh@ponder.io
What do these changes do?
flake8 modin/ asv_bench/benchmarks scripts/doc_checker.pyblack --check modin/ asv_bench/benchmarks scripts/doc_checker.pygit commit -sdocs/development/architecture.rstis up-to-date