[SPARK-29143][PYTHON][ML] Pyspark feature models support column setters/getters #25908

huaxingao · 2019-09-24T00:24:49Z

What changes were proposed in this pull request?

add column setters/getters support in Pyspark feature models

Why are the changes needed?

keep parity between Pyspark and Scala

Does this PR introduce any user-facing change?

Yes.
After the change, Pyspark feature models have column setters/getters support.

How was this patch tested?

Add some doctests

…rs/getters

SparkQA · 2019-09-24T00:50:21Z

Test build #111246 has finished for PR 25908 at commit 2ced6ea.

This patch passes all tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
class LSHParams(JavaParams, HasInputCol, HasOutputCol):
class LSH(JavaEstimator, LSHParams, JavaMLReadable, JavaMLWritable):
class LSHModel(JavaModel, LSHParams):
class BucketedRandomProjectionLSHParams(JavaParams):
class BucketedRandomProjectionLSH(LSH, BucketedRandomProjectionLSHParams,
class BucketedRandomProjectionLSHModel(LSHModel, BucketedRandomProjectionLSHParams, JavaMLReadable,
class ImputerParams(JavaParams, HasInputCols, HasOutputCols):
class Imputer(JavaEstimator, ImputerParams, JavaMLReadable, JavaMLWritable):
class ImputerModel(JavaModel, ImputerParams, JavaMLReadable, JavaMLWritable):
class MaxAbsScalerParams(JavaParams, HasInputCol, HasOutputCol):
class MaxAbsScaler(JavaEstimator, MaxAbsScalerParams, JavaMLReadable, JavaMLWritable):
class MaxAbsScalerModel(JavaModel, MaxAbsScalerParams, JavaMLReadable, JavaMLWritable):
class MinMaxScalerParams(JavaParams, HasInputCol, HasOutputCol):
class MinMaxScaler(JavaEstimator, MinMaxScalerParams, JavaMLReadable, JavaMLWritable):
class MinMaxScalerModel(JavaModel, MinMaxScalerParams, JavaMLReadable, JavaMLWritable):
class OneHotEncoderParams(JavaParams, HasInputCols, HasOutputCols, HasHandleInvalid):
class OneHotEncoder(JavaEstimator, OneHotEncoderParams, JavaMLReadable, JavaMLWritable):
class OneHotEncoderModel(JavaModel, OneHotEncoderParams, JavaMLReadable, JavaMLWritable):
class RobustScalerParams(JavaParams, HasInputCol, HasOutputCol):
class RobustScaler(JavaEstimator, RobustScalerParams, JavaMLReadable, JavaMLWritable):
class RobustScalerModel(JavaModel, RobustScalerParams, JavaMLReadable, JavaMLWritable):
class StandardScalerParams(JavaParams, HasInputCol, HasOutputCol):
class StandardScaler(JavaEstimator, StandardScalerParams, JavaMLReadable, JavaMLWritable):
class StandardScalerModel(JavaModel, StandardScalerParams, JavaMLReadable, JavaMLWritable):
class VectorIndexerParams(JavaParams, HasInputCol, HasOutputCol, HasHandleInvalid):
class VectorIndexer(JavaEstimator, VectorIndexerParams, JavaMLReadable, JavaMLWritable):
class VectorIndexerModel(JavaModel, VectorIndexerParams, JavaMLReadable, JavaMLWritable):
class Word2VecParams(JavaParams, HasStepSize, HasMaxIter, HasSeed, HasInputCol, HasOutputCol):
class Word2Vec(JavaEstimator, Word2VecParams, JavaMLReadable, JavaMLWritable):
class Word2VecModel(JavaModel, Word2VecParams, JavaMLReadable, JavaMLWritable):
class PCAParams(JavaParams, HasInputCol, HasOutputCol):
class PCA(JavaEstimator, PCAParams, JavaMLReadable, JavaMLWritable):
class PCAModel(JavaModel, PCAParams, JavaMLReadable, JavaMLWritable):
class RFormulaParams(JavaParams, HasFeaturesCol, HasLabelCol, HasHandleInvalid):
class RFormula(JavaEstimator, RFormulaParams, JavaMLReadable, JavaMLWritable):
class RFormulaModel(JavaModel, RFormulaParams, JavaMLReadable, JavaMLWritable):
class ChiSqSelectorParams(JavaParams, HasFeaturesCol, HasOutputCol, HasLabelCol):
class ChiSqSelector(JavaEstimator, ChiSqSelectorParams, JavaMLReadable, JavaMLWritable):
class ChiSqSelectorModel(JavaModel, ChiSqSelectorParams, JavaMLReadable, JavaMLWritable):

zhengruifeng · 2019-09-24T07:58:27Z

python/pyspark/ml/feature.py



-class LSHParams(Params):
+class LSHParams(JavaParams, HasInputCol, HasOutputCol):


Do we neet to extend JavaParam here and in other places?

No. I will remove.

zhengruifeng · 2019-09-24T08:00:53Z

python/pyspark/ml/feature.py

    >>> model = maScaler.fit(df)
+    >>> model.getOutputCol()
+    'scaled'
    >>> model.transform(df).show()


what about making sure that the setters really works?

zhengruifeng · 2019-09-24T08:09:00Z

python/pyspark/ml/feature.py

        return self._call_java("numDocs")


+class ImputerParams(JavaParams, HasInputCols, HasOutputCols):


Here by extending HasOutputCols, we do not need to add var outputCols by hand

yes, I didn't have outputCols as member variable

zhengruifeng · 2019-09-24T08:11:02Z

python/pyspark/ml/feature.py

    >>> mmScaler = MinMaxScaler(inputCol="a", outputCol="scaled")
    >>> model = mmScaler.fit(df)
+    >>> model.setOutputCol("scaledOutput")
+    MinMaxScaler...


Why not MinMaxScalerModel...?

it is MinMaxScaler_2cca32620254

huaxingao · 2019-09-24T18:16:29Z

_CountVectorizerParams and _StringIndexerParams have _ . Shall we follow the same convention? @zhengruifeng

SparkQA · 2019-09-24T18:38:09Z

Test build #111304 has finished for PR 25908 at commit 74de7a4.

This patch passes all tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
class LSHParams(HasInputCol, HasOutputCol):
class BucketedRandomProjectionLSHParams():
class IDFParams(HasInputCol, HasOutputCol):
class IDF(JavaEstimator, IDFParams, JavaMLReadable, JavaMLWritable):
class IDFModel(JavaModel, IDFParams, JavaMLReadable, JavaMLWritable):
class ImputerParams(HasInputCols, HasOutputCols):
class MaxAbsScalerParams(HasInputCol, HasOutputCol):
class MinMaxScalerParams(HasInputCol, HasOutputCol):
class OneHotEncoderParams(HasInputCols, HasOutputCols, HasHandleInvalid):
class RobustScalerParams(HasInputCol, HasOutputCol):
class StandardScalerParams(HasInputCol, HasOutputCol):
class VectorIndexerParams(HasInputCol, HasOutputCol, HasHandleInvalid):
class Word2VecParams(HasStepSize, HasMaxIter, HasSeed, HasInputCol, HasOutputCol):
class PCAParams(HasInputCol, HasOutputCol):
class RFormulaParams(HasFeaturesCol, HasLabelCol, HasHandleInvalid):
class ChiSqSelectorParams(HasFeaturesCol, HasOutputCol, HasLabelCol):

srowen

Pretty much same comments as #25859 ; does this change any API or just refactor?

huaxingao · 2019-09-25T17:51:31Z

Similar to the other PR, this PR adds setters/getters to feature models.

zhengruifeng · 2019-09-26T03:50:07Z

@huaxingao

Shall we follow the same convention?

I prefer to follow it in the following PRs.

SparkQA · 2019-09-26T17:33:43Z

Test build #111437 has finished for PR 25908 at commit 425959e.

This patch passes all tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
class _LSHParams(HasInputCol, HasOutputCol):
class LSH(JavaEstimator, _LSHParams, JavaMLReadable, JavaMLWritable):
class LSHModel(JavaModel, _LSHParams):
class _BucketedRandomProjectionLSHParams():
class BucketedRandomProjectionLSH(LSH, _BucketedRandomProjectionLSHParams,
class BucketedRandomProjectionLSHModel(LSHModel, _BucketedRandomProjectionLSHParams, JavaMLReadable,
class _IDFParams(HasInputCol, HasOutputCol):
class IDF(JavaEstimator, _IDFParams, JavaMLReadable, JavaMLWritable):
class IDFModel(JavaModel, _IDFParams, JavaMLReadable, JavaMLWritable):
class _ImputerParams(HasInputCols, HasOutputCols):
class Imputer(JavaEstimator, _ImputerParams, JavaMLReadable, JavaMLWritable):
class ImputerModel(JavaModel, _ImputerParams, JavaMLReadable, JavaMLWritable):
class _MaxAbsScalerParams(HasInputCol, HasOutputCol):
class MaxAbsScaler(JavaEstimator, _MaxAbsScalerParams, JavaMLReadable, JavaMLWritable):
class MaxAbsScalerModel(JavaModel, _MaxAbsScalerParams, JavaMLReadable, JavaMLWritable):
class MinHashLSH(JavaEstimator, _LSHParams, HasInputCol, HasOutputCol, HasSeed,
class _MinMaxScalerParams(HasInputCol, HasOutputCol):
class MinMaxScaler(JavaEstimator, _MinMaxScalerParams, JavaMLReadable, JavaMLWritable):
class MinMaxScalerModel(JavaModel, _MinMaxScalerParams, JavaMLReadable, JavaMLWritable):
class _OneHotEncoderParams(HasInputCols, HasOutputCols, HasHandleInvalid):
class OneHotEncoder(JavaEstimator, _OneHotEncoderParams, JavaMLReadable, JavaMLWritable):
class OneHotEncoderModel(JavaModel, _OneHotEncoderParams, JavaMLReadable, JavaMLWritable):
class _RobustScalerParams(HasInputCol, HasOutputCol):
class RobustScaler(JavaEstimator, _RobustScalerParams, JavaMLReadable, JavaMLWritable):
class RobustScalerModel(JavaModel, _RobustScalerParams, JavaMLReadable, JavaMLWritable):
class _StandardScalerParams(HasInputCol, HasOutputCol):
class StandardScaler(JavaEstimator, _StandardScalerParams, JavaMLReadable, JavaMLWritable):
class StandardScalerModel(JavaModel, _StandardScalerParams, JavaMLReadable, JavaMLWritable):
class _VectorIndexerParams(HasInputCol, HasOutputCol, HasHandleInvalid):
class VectorIndexer(JavaEstimator, _VectorIndexerParams, JavaMLReadable, JavaMLWritable):
class VectorIndexerModel(JavaModel, _VectorIndexerParams, JavaMLReadable, JavaMLWritable):
class _Word2VecParams(HasStepSize, HasMaxIter, HasSeed, HasInputCol, HasOutputCol):
class Word2Vec(JavaEstimator, _Word2VecParams, JavaMLReadable, JavaMLWritable):
class Word2VecModel(JavaModel, _Word2VecParams, JavaMLReadable, JavaMLWritable):
class _PCAParams(HasInputCol, HasOutputCol):
class PCA(JavaEstimator, _PCAParams, JavaMLReadable, JavaMLWritable):
class PCAModel(JavaModel, _PCAParams, JavaMLReadable, JavaMLWritable):
class _RFormulaParams(HasFeaturesCol, HasLabelCol, HasHandleInvalid):
class RFormula(JavaEstimator, _RFormulaParams, JavaMLReadable, JavaMLWritable):
class RFormulaModel(JavaModel, _RFormulaParams, JavaMLReadable, JavaMLWritable):
class _ChiSqSelectorParams(HasFeaturesCol, HasOutputCol, HasLabelCol):
class ChiSqSelector(JavaEstimator, _ChiSqSelectorParams, JavaMLReadable, JavaMLWritable):
class ChiSqSelectorModel(JavaModel, _ChiSqSelectorParams, JavaMLReadable, JavaMLWritable):

zhengruifeng · 2019-09-27T08:52:41Z

python/pyspark/ml/feature.py


 @inherit_doc
-class MinHashLSH(JavaEstimator, LSHParams, HasInputCol, HasOutputCol, HasSeed,
+class MinHashLSH(JavaEstimator, _LSHParams, HasInputCol, HasOutputCol, HasSeed,


This place seems a bit different from scala side
MinHashLSH(override val uid: String) extends LSH[MinHashLSHModel] with HasSeed

Will change.

zhengruifeng · 2019-09-27T09:05:41Z

python/pyspark/ml/tests/test_param.py

                if not name.endswith('Model') and not name.endswith('Params') \
                        and issubclass(cls, JavaParams) and not inspect.isabstract(cls) \
-                        and not name.startswith('Java'):
+                        and not name.startswith('Java') and name != 'LSH':


Why need to filter out LSH here?

The original code doesn't have python class LSH

class LSH(JavaEstimator, _LSHParams, JavaMLReadable, JavaMLWritable): """ Mixin for Locality Sensitive Hashing (LSH). """ def setNumHashTables(self, value): """ Sets the value of :py:attr:`numHashTables`. """ return self._set(numHashTables=value)

I add this class so I can have a place for setNumHashTables. I don't have a __init__ in LSH because scala LSH is an abstract class and no constructor. Since no __init__, this self._java_obj is never set and test_param will throw Exception for LSH

Traceback (most recent call last): File "/Users/hgao/spark092119/spark/python/pyspark/ml/tests/test_param.py", line 358, in test_java_params check_params(self, cls(), check_params_exist=False) File "/Users/hgao/spark092119/spark/python/pyspark/testing/mlutils.py", line 40, in check_params java_stage = py_stage._to_java() File "/Users/hgao/spark092119/spark/python/pyspark/ml/wrapper.py", line 222, in _to_java self._transfer_params_to_java() File "/Users/hgao/spark092119/spark/python/pyspark/ml/wrapper.py", line 145, in _transfer_params_to_java pair = self._make_java_param_pair(param, self._defaultParamMap[param]) File "/Users/hgao/spark092119/spark/python/pyspark/ml/wrapper.py", line 131, in _make_java_param_pair java_param = self._java_obj.getParam(param.name) AttributeError: 'NoneType' object has no attribute 'getParam'

SparkQA · 2019-09-27T18:41:07Z

Test build #111493 has finished for PR 25908 at commit 50e4467.

This patch passes all tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
class MinHashLSH(LSH, HasInputCol, HasOutputCol, HasSeed, JavaMLReadable, JavaMLWritable):

zero323 · 2019-09-28T11:07:27Z

@srowen @zhengruifeng @huaxingao Could you explain the rationale behind 425959e? Leading underscore doesn't really have the same semantics as Scala package private (or private modifier in general).

However, when used in context which have direct impact on user facing API, it is quite annoying, as it requires manually segregation from truly internal components. Even if the goal is simple API parity it makes sense to take such things into account. Otherwise the whole thing could be just code generated. Just saying...

srowen · 2019-09-28T15:26:22Z

If the general question is, why hide things? just the usual software design idea, to only expose what you intend to support as a user or developer API. I'm not making a specific judgment about whether these should be exposed or not, just don't know yet.

If the question is, can we make this more extensible, I also don't have a strong opinion but am not against it, but think that can be considered separately.

There are also going to be a few follow-on PRs to this one to further remove some setters.

huaxingao · 2019-09-28T15:49:15Z

@zero323
If your question is specific for this leading underscore I just added, there are two reasons:

To keep consistent with the previous implementation in _CountVectorizerParams and _StringIndexerParams
Since Python doesn't really have private modifier, seems to me the closest one is to have leading underscore specified in PEP 8 convention. https://www.python.org/dev/peps/pep-0008/
_single_leading_underscore: weak "internal use" indicator. E.g. from M import * does not import objects whose names start with an underscore.

I am neural on this leading underscore. Since I am following Bryan's convention in _CountVectorizerParams, I would like to check with him to see if there are any other reasons for this leading underscore. Thanks in advance @BryanCutler

zero323 · 2019-09-28T16:45:24Z

If the general question is, why hide things? just the usual software design idea, to only expose what you intend to support as a user or developer API. I'm not making a specific judgment about whether these should be exposed or not, just don't know yet.

Not at all @srowen. It is a specific question. These mixins are used only to augment public API, and as such directly affect user's code (for example method resolution order), so it is a bit fuzzy why indicate them as internal.

From one hand their usage is limited to specific Estimator / Model pair, so there is really little value in exposing these to the end user. From the other, hand limited scope suggests little potential for change, especially one that doesn't affect public API, hence there seems to be no little harm in exposing these (the problem is similar to SPARK-7146).

In contrast to SPARK-29212, I am indifferent here and just trying to get a better feeling how the API evolves and what drives certain decisions.

zero323 · 2019-09-28T17:20:10Z

@huaxingao. I understand leading underscore usage and I've seen Bryan's work, but for now it was an exception in pyspark.ml. We have a bunch of fairly isolated *Params but nothing indicates these are considered internal. Hence I was curious why the sudden change.

SparkQA · 2019-10-03T21:43:23Z

Test build #111754 has finished for PR 25908 at commit bc25799.

This patch passes all tests.
This patch merges cleanly.
This patch adds no public classes.

zero323 · 2019-10-03T21:50:28Z

python/pyspark/ml/feature.py

+        return self.getOrDefault(self.numHashTables)
+
+
+class LSH(JavaEstimator, _LSHParams, JavaMLReadable, JavaMLWritable):


I guess this should be _LSH as well, shouldn't it? Scala counterpart is private[ml].

zero323 · 2019-10-03T21:51:07Z

python/pyspark/ml/feature.py

-

-class LSHModel(JavaModel):
+class LSHModel(JavaModel, _LSHParams):


Ditto. Maybe it should be _LSHModel?

zero323 · 2019-10-03T22:02:32Z

python/pyspark/ml/feature.py

+    """
+    Params for :py:class:`BucketedRandomProjectionLSH` and
+    :py:class:`BucketedRandomProjectionLSHModel`.
+    .. versionadded:: 3.0.0


There should be an empty line between the description and .. versionadded:: ... (or any directive)

""" Params for :py:class:`BucketedRandomProjectionLSH` and :py:class:`BucketedRandomProjectionLSHModel`. .. versionadded:: 3.0.0

Otherwise such elements are not recognized as directives and incorrectly appended to the body.

zero323 · 2019-10-03T22:02:56Z

python/pyspark/ml/feature.py

+class _IDFParams(HasInputCol, HasOutputCol):
+    """
+    Params for :py:class:`IDF` and :py:class:`IDFModel`.
+    .. versionadded:: 3.0.0


zero323 · 2019-10-03T22:03:07Z

python/pyspark/ml/feature.py

+class _ImputerParams(HasInputCols, HasOutputCols):
+    """
+    Params for :py:class:`Imputer` and :py:class:`ImputerModel`.
+    .. versionadded:: 3.0.0


zero323 · 2019-10-03T22:03:17Z

python/pyspark/ml/feature.py

+class _MaxAbsScalerParams(HasInputCol, HasOutputCol):
+    """
+    Params for :py:class:`MaxAbsScaler` and :py:class:`MaxAbsScalerModel`.
+    .. versionadded:: 3.0.0


zero323 · 2019-10-03T22:03:25Z

python/pyspark/ml/feature.py

+class _MinMaxScalerParams(HasInputCol, HasOutputCol):
+    """
+    Params for :py:class:`MinMaxScaler` and :py:class:`MinMaxScalerModel`.
+    .. versionadded:: 3.0.0


zero323 · 2019-10-03T22:03:35Z

python/pyspark/ml/feature.py

+class _OneHotEncoderParams(HasInputCols, HasOutputCols, HasHandleInvalid):
+    """
+    Params for :py:class:`OneHotEncoder` and :py:class:`OneHotEncoderModel`.
+    .. versionadded:: 3.0.0


zero323 · 2019-10-03T22:03:46Z

python/pyspark/ml/feature.py

+class _RobustScalerParams(HasInputCol, HasOutputCol):
+    """
+    Params for :py:class:`RobustScaler` and :py:class:`RobustScalerModel`.
+    .. versionadded:: 3.0.0


zero323 · 2019-10-03T22:04:02Z

python/pyspark/ml/feature.py

+class _StandardScalerParams(HasInputCol, HasOutputCol):
+    """
+    Params for :py:class:`StandardScaler` and :py:class:`StandardScalerModel`.
+    .. versionadded:: 3.0.0


Ditto. And the remaining ones as well.

SparkQA · 2019-10-04T05:15:31Z

Test build #111767 has finished for PR 25908 at commit c305a43.

This patch fails Python style tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
class _LSH(JavaEstimator, _LSHParams, JavaMLReadable, JavaMLWritable):
class _LSHModel(JavaModel, _LSHParams):
class BucketedRandomProjectionLSH(_LSH, _BucketedRandomProjectionLSHParams,
class BucketedRandomProjectionLSHModel(_LSHModel, _BucketedRandomProjectionLSHParams, JavaMLReadable,
class MinHashLSH(_LSH, HasInputCol, HasOutputCol, HasSeed, JavaMLReadable, JavaMLWritable):
class MinHashLSHModel(_LSHModel, JavaMLReadable, JavaMLWritable):

SparkQA · 2019-10-04T06:37:47Z

Test build #111768 has finished for PR 25908 at commit 64fca95.

This patch passes all tests.
This patch merges cleanly.
This patch adds the following public classes (experimental):
class BucketedRandomProjectionLSHModel(_LSHModel, _BucketedRandomProjectionLSHParams,

srowen · 2019-10-07T15:55:59Z

Merged to master

huaxingao · 2019-10-07T16:54:12Z

Thanks all for your help!

[SPARK-29143][PYTHON][ML] Pyspark feature models support column sette…

2ced6ea

…rs/getters

dongjoon-hyun added ML PYSPARK labels Sep 24, 2019

zhengruifeng reviewed Sep 24, 2019

View reviewed changes

address comments

74de7a4

srowen reviewed Sep 25, 2019

View reviewed changes

zhengruifeng mentioned this pull request Sep 26, 2019

[SPARK-29142][PYTHON][ML] Pyspark clustering models support column setters/getters/predict #25859

Closed

add _ in front of xxxParams to indicate internal use (PEP8)

425959e

zhengruifeng reviewed Sep 27, 2019

View reviewed changes

address comments

50e4467

change :py:attr: to :py:class for classes

bc25799

zero323 reviewed Oct 3, 2019

View reviewed changes

address comments

c305a43

fix pycodestyle error

64fca95

zhengruifeng approved these changes Oct 4, 2019

View reviewed changes

srowen closed this in 2399134 Oct 7, 2019

zero323 mentioned this pull request Oct 7, 2019

Sync with changes merged after 6378d4bc06cd1bb1a209bd5fb63d10ef52d75eb4 zero323/pyspark-stubs#230

Closed

47 tasks

huaxingao deleted the spark-29143 branch October 7, 2019 16:54



		class LSHParams(Params):
		class LSHParams(JavaParams, HasInputCol, HasOutputCol):

		return self._call_java("numDocs")


		class ImputerParams(JavaParams, HasInputCols, HasOutputCols):

		return self.getOrDefault(self.numHashTables)


		class LSH(JavaEstimator, _LSHParams, JavaMLReadable, JavaMLWritable):



		class LSHModel(JavaModel):
		class LSHModel(JavaModel, _LSHParams):

[SPARK-29143][PYTHON][ML] Pyspark feature models support column setters/getters #25908

[SPARK-29143][PYTHON][ML] Pyspark feature models support column setters/getters #25908

Uh oh!

Conversation

huaxingao commented Sep 24, 2019

What changes were proposed in this pull request?

Why are the changes needed?

Does this PR introduce any user-facing change?

How was this patch tested?

Uh oh!

SparkQA commented Sep 24, 2019

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

huaxingao commented Sep 24, 2019

Uh oh!

SparkQA commented Sep 24, 2019

Uh oh!

srowen left a comment

Choose a reason for hiding this comment

Uh oh!

huaxingao commented Sep 25, 2019

Uh oh!

zhengruifeng commented Sep 26, 2019 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Uh oh!

SparkQA commented Sep 26, 2019

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

SparkQA commented Sep 27, 2019

Uh oh!

zero323 commented Sep 28, 2019

Uh oh!

srowen commented Sep 28, 2019

Uh oh!

huaxingao commented Sep 28, 2019

Uh oh!

zero323 commented Sep 28, 2019

Uh oh!

zero323 commented Sep 28, 2019

Uh oh!

SparkQA commented Oct 3, 2019

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

Choose a reason for hiding this comment

Uh oh!

zhengruifeng commented Sep 26, 2019 •

edited

Loading

zero323 Oct 3, 2019 •

edited

Loading