[SPARK-32098][PYTHON] Use iloc for positional slicing instead of direct slicing in createDataFrame with Arrow #28928
Conversation
This file contains hidden or 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
…taFrame with Arrow
Member
Author
|
I think this should be ported back through branch-2.4 ... |
|
Test build #124513 has finished for PR 28928 at commit
|
BryanCutler
approved these changes
Jun 25, 2020
Member
BryanCutler
left a comment
BryanCutler
left a comment
There was a problem hiding this comment.
LGTM. I don't know all the ins and outs of using different kinds of index types, but iloc is the preferred way to select by position now.
BryanCutler
pushed a commit
that referenced
this pull request
Jun 25, 2020
…ct slicing in createDataFrame with Arrow
When you use floats are index of pandas, it creates a Spark DataFrame with a wrong results as below when Arrow is enabled:
```bash
./bin/pyspark --conf spark.sql.execution.arrow.pyspark.enabled=true
```
```python
>>> import pandas as pd
>>> spark.createDataFrame(pd.DataFrame({'a': [1,2,3]}, index=[2., 3., 4.])).show()
+---+
| a|
+---+
| 1|
| 1|
| 2|
+---+
```
This is because direct slicing uses the value as index when the index contains floats:
```python
>>> pd.DataFrame({'a': [1,2,3]}, index=[2., 3., 4.])[2:]
a
2.0 1
3.0 2
4.0 3
>>> pd.DataFrame({'a': [1,2,3]}, index=[2., 3., 4.]).iloc[2:]
a
4.0 3
>>> pd.DataFrame({'a': [1,2,3]}, index=[2, 3, 4])[2:]
a
4 3
```
This PR proposes to explicitly use `iloc` to positionally slide when we create a DataFrame from a pandas DataFrame with Arrow enabled.
FWIW, I was trying to investigate why direct slicing refers the index value or the positional index sometimes but I stopped investigating further after reading this https://pandas.pydata.org/pandas-docs/stable/getting_started/10min.html#selection
> While standard Python / Numpy expressions for selecting and setting are intuitive and come in handy for interactive work, for production code, we recommend the optimized pandas data access methods, `.at`, `.iat`, `.loc` and `.iloc`.
To create the correct Spark DataFrame from a pandas DataFrame without a data loss.
Yes, it is a bug fix.
```bash
./bin/pyspark --conf spark.sql.execution.arrow.pyspark.enabled=true
```
```python
import pandas as pd
spark.createDataFrame(pd.DataFrame({'a': [1,2,3]}, index=[2., 3., 4.])).show()
```
Before:
```
+---+
| a|
+---+
| 1|
| 1|
| 2|
+---+
```
After:
```
+---+
| a|
+---+
| 1|
| 2|
| 3|
+---+
```
Manually tested and unittest were added.
Closes #28928 from HyukjinKwon/SPARK-32098.
Authored-by: HyukjinKwon <[email protected]>
Signed-off-by: Bryan Cutler <[email protected]>
(cherry picked from commit 1af19a7)
Signed-off-by: Bryan Cutler <[email protected]>
BryanCutler
pushed a commit
that referenced
this pull request
Jun 25, 2020
…ct slicing in createDataFrame with Arrow
When you use floats are index of pandas, it creates a Spark DataFrame with a wrong results as below when Arrow is enabled:
```bash
./bin/pyspark --conf spark.sql.execution.arrow.pyspark.enabled=true
```
```python
>>> import pandas as pd
>>> spark.createDataFrame(pd.DataFrame({'a': [1,2,3]}, index=[2., 3., 4.])).show()
+---+
| a|
+---+
| 1|
| 1|
| 2|
+---+
```
This is because direct slicing uses the value as index when the index contains floats:
```python
>>> pd.DataFrame({'a': [1,2,3]}, index=[2., 3., 4.])[2:]
a
2.0 1
3.0 2
4.0 3
>>> pd.DataFrame({'a': [1,2,3]}, index=[2., 3., 4.]).iloc[2:]
a
4.0 3
>>> pd.DataFrame({'a': [1,2,3]}, index=[2, 3, 4])[2:]
a
4 3
```
This PR proposes to explicitly use `iloc` to positionally slide when we create a DataFrame from a pandas DataFrame with Arrow enabled.
FWIW, I was trying to investigate why direct slicing refers the index value or the positional index sometimes but I stopped investigating further after reading this https://pandas.pydata.org/pandas-docs/stable/getting_started/10min.html#selection
> While standard Python / Numpy expressions for selecting and setting are intuitive and come in handy for interactive work, for production code, we recommend the optimized pandas data access methods, `.at`, `.iat`, `.loc` and `.iloc`.
To create the correct Spark DataFrame from a pandas DataFrame without a data loss.
Yes, it is a bug fix.
```bash
./bin/pyspark --conf spark.sql.execution.arrow.pyspark.enabled=true
```
```python
import pandas as pd
spark.createDataFrame(pd.DataFrame({'a': [1,2,3]}, index=[2., 3., 4.])).show()
```
Before:
```
+---+
| a|
+---+
| 1|
| 1|
| 2|
+---+
```
After:
```
+---+
| a|
+---+
| 1|
| 2|
| 3|
+---+
```
Manually tested and unittest were added.
Closes #28928 from HyukjinKwon/SPARK-32098.
Authored-by: HyukjinKwon <[email protected]>
Signed-off-by: Bryan Cutler <[email protected]>
(cherry picked from commit 1af19a7)
Signed-off-by: Bryan Cutler <[email protected]>
Member
|
merged to master, branch-3.0 and branch-2.4 |
Member
Author
|
Thank you @BryanCutler and @ueshin! |
gatorsmile
reviewed
Jun 26, 2020
| # Slice the DataFrame to be batched | ||
| step = -(-len(pdf) // self.sparkContext.defaultParallelism) # round int up | ||
| pdf_slices = (pdf[start:start + step] for start in xrange(0, len(pdf), step)) | ||
| pdf_slices = (pdf.iloc[start:start + step] for start in xrange(0, len(pdf), step)) |
Member
There was a problem hiding this comment.
Thank you for fixing this!
While standard Python / Numpy expressions for selecting and setting are intuitive and come in handy for interactive work, for production code, we recommend the optimized pandas data access methods, .at, .iat, .loc and .iloc.
Is it the only place?
Member
Author
There was a problem hiding this comment.
As far as I can tell, yes.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
5 participants
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.
What changes were proposed in this pull request?
When you use floats in the index of pandas, it creates a Spark DataFrame with a wrong result as below when Arrow is enabled:
This is because direct slicing uses the value as index when the index contains floats:
This PR proposes to explicitly use
ilocto positionally slide when we create a DataFrame from a pandas DataFrame with Arrow enabled.FWIW, I was trying to investigate why direct slicing refers the index value or the positional index sometimes but I stopped investigating further after reading this https://pandas.pydata.org/pandas-docs/stable/getting_started/10min.html#selection
Why are the changes needed?
To create the correct Spark DataFrame from a pandas DataFrame without a data loss.
Does this PR introduce any user-facing change?
Yes, it is a bug fix.
Before:
After:
How was this patch tested?
Manually tested and unittest were added.