FEAT-#3043: allow pandas.DataFrame to be passed into merge and join functions
#4324
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
…ge function Signed-off-by: Anatoly Myachev <anatoly.myachev@intel.com>
|
Codecov Report
@@ Coverage Diff @@
## master #4324 +/- ##
==========================================
+ Coverage 86.91% 90.01% +3.09%
==========================================
Files 214 214
Lines 17367 17998 +631
==========================================
+ Hits 15095 16200 +1105
+ Misses 2272 1798 -474
📣 Codecov can now indicate which changes are the most critical in Pull Requests. Learn more |
|
|
Signed-off-by: Anatoly Myachev <anatoly.myachev@intel.com>
anmyachev
changed the title
pandas.DataFrame to be passed into merge and join functions
|
@devin-petersohn @YarShev in Can we add this feature only for these functions, leaving some inconsistency for the rest of the binary operations? What do you think? This is beneficial at least as long as this conversion exists, but I hope that in the future we will get rid of it. However, if we add this feature now, users will rely on it and we will have to support it further. FYI: this case is being considered because it speeds up one of our workloads. |
dchigarev
reviewed
Mar 18, 2022
| return self.join( | ||
| right, how=how, lsuffix=suffixes[0], rsuffix=suffixes[1], sort=sort | ||
| ) | ||
|
|
||
| return self.__constructor__( | ||
| query_compiler=self._query_compiler.merge( | ||
| right._query_compiler, | ||
| right._query_compiler | ||
| if not isinstance(right, pandas.DataFrame) |
There was a problem hiding this comment.
Suggested change
| if not isinstance(right, pandas.DataFrame) | |
| if isinstance(right, (pd.DataFrame, pd.Series)) |
| @@ -1258,14 +1258,18 @@ def join( | |||
| if on is not None: | |||
| return self.__constructor__( | |||
| query_compiler=self._query_compiler.join( | |||
| other._query_compiler, | |||
| other._query_compiler | |||
| if not isinstance(other, pandas.DataFrame) | |||
There was a problem hiding this comment.
Suggested change
| if not isinstance(other, pandas.DataFrame) | |
| if isinstance(other, (pd.DataFrame, pd.Series)) |
| @@ -448,7 +448,8 @@ def merge(self, right, **kwargs): | |||
| sort = kwargs.get("sort", False) | |||
|
|
|||
| if how in ["left", "inner"] and left_index is False and right_index is False: | |||
| right = right.to_pandas() | |||
| if not isinstance(right, pandas.DataFrame): | |||
There was a problem hiding this comment.
- A non-negative condition is easier to read
- Checking that
rightis not a pandas.DataFrame does not ensure that it has a.to_pandasmethod
Suggested change
| if not isinstance(right, pandas.DataFrame): | |
| if isinstance(right, BaseQueryCompiler): |
| @@ -495,7 +496,8 @@ def join(self, right, **kwargs): | |||
| sort = kwargs.get("sort", False) | |||
|
|
|||
| if how in ["left", "inner"]: | |||
| right = right.to_pandas() | |||
| if not isinstance(right, pandas.DataFrame): | |||
There was a problem hiding this comment.
Suggested change
| if not isinstance(right, pandas.DataFrame): | |
| if isinstance(right, BaseQueryCompiler): |
| @@ -1387,19 +1391,25 @@ def merge( | |||
| raise ValueError("Cannot merge a Series without a name") | |||
| else: | |||
| right = right.to_frame() | |||
| if not isinstance(right, DataFrame): | |||
| if not isinstance(right, (DataFrame, pandas.DataFrame)): | |||
There was a problem hiding this comment.
we unconditionally converted right to pd.DataFrame above, why do we want to check for pandas.DataFrame here?
Comment on lines
+1400
to
+1403
| if not isinstance(right, DataFrame): | ||
| raise TypeError( | ||
| f"Can only merge Series or DataFrame objects, a {type(right)} was passed" | ||
| ) |
There was a problem hiding this comment.
why the above check is not enough?
| @@ -179,6 +180,21 @@ def test_merge(): | |||
| pd.merge("Non-valid type", modin_df2) | |||
|
|
|||
|
|
|||
| @pytest.mark.parametrize("data", test_data_values, ids=test_data_keys) | |||
| def test_merge_pandas(data): | |||
| modin_df, pandas_df = pd.DataFrame(data), pandas.DataFrame(data) | |||
There was a problem hiding this comment.
this has to be imported from modin.pandas.test.utils
Suggested change
| modin_df, pandas_df = pd.DataFrame(data), pandas.DataFrame(data) | |
| modin_df, pandas_df = create_test_dfs(data) |
|
|
||
| def test_join_pandas(): | ||
| data = test_data["int_data"] | ||
| modin_df, pandas_df = pd.DataFrame(data), pandas.DataFrame(data) |
There was a problem hiding this comment.
Suggested change
| modin_df, pandas_df = pd.DataFrame(data), pandas.DataFrame(data) | |
| modin_df, pandas_df = create_test_dfs(data) |
|
@anmyachev what's the status of this PR? |
|
I think @devin-petersohn was specifically against mixing Modin and pandas in that way, even though it's a "low hanging fruit" optimization in this case. I feel that this should be better addressed by implementing #4605 and allowing the end user to easily "force" Modin to wrap certain pandas dataframe in this specific "small df" QueryCompiler. |
|
@pyrito for now, this most likely will not be merged, @vnlitvinov is right. By the way, I have an interesting case, which may not be very relevant to this task, but still. This call works for pandas, but fails for modin because modin tries to initialize ray on the worker (if I remember correctly). In this case, the following change helps: import modin.pandas as pd
def agg_func(x):
return int(pd.Series.count(x) > 0)
res = some_df.pivot_table(..., agg_func=agg_func) |
@anmyachev this fails since the function is running on a ray worker - maybe we can wrap deployed functions with a function that just imports pandas as pd, so if the user tries to use pd inside an apply function (for example) it's scope will have pandas instead of Modin, and it'll work? |
The following example doesn't work. def agg_func(x):
return int(pandas.Series.count(x) > 0)
def wrapper(x):
import pandas as pd
return agg_func(x)
res = some_df.pivot_table(..., agg_func=wrapper)if I understand correctly, |
|
@anmyachev We could modify the scope and replace modin.pandas with pandas by modifying the globals dict (although I'm not sure if we want to get into this level of meta-programming?) |
|
We need to first decide on an approach. |
|
@RehanSD this approach is not completely right, you should create new dict for I would guess that:
|
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: Anatoly Myachev anatoly.myachev@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 -sdocs/development/architecture.rstis up-to-date