Allows for merging of SparseDataFrames, and fixes array interface #19488

hexgnu · 2018-02-01T08:27:34Z

closes merge of sparse dataframe fails #13665 and SparseSeries.__array__ only returns non-fills #14167 and take_1d yields surprising results when working with SparseArray #19506
tests added / passed
passes git diff upstream/master -u -- "*.py" | flake8 --diff
whatsnew entry

This is still an in progress thing... since this is pretty complected code that I'm trying to unravel.

pep8speaks · 2018-02-01T08:27:40Z

Hello @hexgnu! Thanks for updating the PR.

Cheers ! There are no PEP8 issues in this Pull Request. 🍻

Comment last updated on February 12, 2018 at 05:07 Hours UTC

hexgnu · 2018-02-01T08:29:22Z

pandas/tests/reshape/merge/test_merge.py

+        dense_merge = dense_evens.merge(dense_threes, how=how, on='A')
+
+        # If you merge two dense frames together it tends to default to float64 not the original dtype
+        dense_merge['B_x'] = dense_merge['B_x'].astype(np.int64, errors='ignore')


This to me seems kind of bizarre and I couldn't find an issue for it but basically:

If you merge two dense frames together that I defined above the dtype goes from int64 to float64. I think I know where the code is that's doing that so I could fix it but didn't want to get too side tracked in this work.

That's since np.nan is the missing value indicator, which is a float. Doing the merge induces missing values for how=left/right/outer, so the ints are cast to floats.

hexgnu · 2018-02-01T08:30:47Z

pandas/core/sparse/frame.py

@@ -73,6 +75,9 @@ def __init__(self, data=None, index=None, columns=None, default_kind=None,
            if columns is None:
                raise Exception("cannot pass a series w/o a name or columns")
            data = {columns[0]: data}
+        elif isinstance(data, BlockManager):
+            if default_fill_value is None:
+                default_fill_value, _ = Counter([b.fill_value for b in data.blocks]).most_common(1)[0]


I don't know if this is kosher or not...

Basically I kept running into the issue of

If you create a SparseDataFrame with a bunch of SparseSeries / SparseArrays that have fill_values like == 1 or ==0 or something then it doesn't persist that to default_fill_value. That seems like an enhancement and I could take it out of this PR but it helped me test.

This seems like a recipe for surprise. I'd hate for df.reindex() to do something different, based on the types of blocks that SparseDataFrame happened to be initialized with. If a user explicitly sets default_fill_value that's one thing, but inferring it from the data seems problematic.

I can take it out it didn't add much tbh. was just making my testing easier ;)

TomAugspurger · 2018-02-01T15:03:48Z

doc/source/whatsnew/v0.23.0.txt

@@ -538,7 +538,7 @@ Reshaping
 - Bug in :func:`DataFrame.merge` in which merging using ``Index`` objects as vectors raised an Exception (:issue:`19038`)
 - Bug in :func:`DataFrame.stack`, :func:`DataFrame.unstack`, :func:`Series.unstack` which were not returning subclasses (:issue:`15563`)
 - Bug in timezone comparisons, manifesting as a conversion of the index to UTC in ``.concat()`` (:issue:`18523`)
-
+- Bug in :func:`SparseDataFrame.merge` which raises error (:issue:`13665`)


This can go under enhancements, saying "Merging sparse DataFrames" is now supported.

Oh it's an enhancement alrighty thanks for the guidance.

TomAugspurger · 2018-02-01T15:04:09Z

pandas/core/internals.py

    if klass is None:
        dtype = dtype or values.dtype
        klass = get_block_type(values, dtype)

    elif klass is DatetimeTZBlock and not is_datetimetz(values):
        return klass(values, ndim=ndim,
                     placement=placement, dtype=dtype)
-
+    


You might want to install a flake8 plugin for your editor ;)

ok ok installing now. I fully admit that I'm awful when it comes to linting failures and should just tool the fix. Might as well install Ctrl-P while I'm at it.

TomAugspurger · 2018-02-01T15:05:09Z

pandas/core/internals.py

        blocks.append(b)

    return BlockManager(blocks, axes)

+def is_sparse_join_units(join_units):


Docstring would be nice, at least noting that this is true if any blocks are sparse.

TomAugspurger · 2018-02-01T15:07:15Z

pandas/core/internals.py

@@ -5120,14 +5121,28 @@ def concatenate_block_managers(mgrs_indexers, axes, concat_axis, copy):
        elif is_uniform_join_units(join_units):
            b = join_units[0].block.concat_same_type(
                [ju.block for ju in join_units], placement=placement)
+        elif is_sparse_join_units(join_units):
+            values = concatenate_join_units(join_units, concat_axis, copy=copy)
+            values = values[0]


I don't understand this line... A comment maybe? or is it incorrect?

TomAugspurger · 2018-02-01T15:11:39Z

pandas/core/sparse/frame.py

@@ -73,6 +75,9 @@ def __init__(self, data=None, index=None, columns=None, default_kind=None,
            if columns is None:
                raise Exception("cannot pass a series w/o a name or columns")
            data = {columns[0]: data}
+        elif isinstance(data, BlockManager):
+            if default_fill_value is None:
+                default_fill_value, _ = Counter([b.fill_value for b in data.blocks]).most_common(1)[0]


This seems like a recipe for surprise. I'd hate for df.reindex() to do something different, based on the types of blocks that SparseDataFrame happened to be initialized with. If a user explicitly sets default_fill_value that's one thing, but inferring it from the data seems problematic.

TomAugspurger · 2018-02-01T15:13:47Z

pandas/tests/reshape/merge/test_merge.py

+        dense_merge = dense_evens.merge(dense_threes, how=how, on='A')
+
+        # If you merge two dense frames together it tends to default to float64 not the original dtype
+        dense_merge['B_x'] = dense_merge['B_x'].astype(np.int64, errors='ignore')


That's since np.nan is the missing value indicator, which is a float. Doing the merge induces missing values for how=left/right/outer, so the ints are cast to floats.

codecov · 2018-02-05T12:45:52Z

Codecov Report

Merging #19488 into master will decrease coverage by 0.19%.
The diff coverage is 96.87%.

@@            Coverage Diff            @@
##           master   #19488     +/-   ##
=========================================
- Coverage   91.82%   91.62%   -0.2%     
=========================================
  Files         152      150      -2     
  Lines       49248    48833    -415     
=========================================
- Hits        45222    44744    -478     
- Misses       4026     4089     +63

Flag	Coverage Δ
#multiple	`90% <96.87%> (-0.22%)`	⬇️
#single	`41.72% <28.12%> (-0.18%)`	⬇️

Impacted Files	Coverage Δ
pandas/core/indexes/base.py	`96.44% <100%> (-0.25%)`	⬇️
pandas/core/reshape/merge.py	`94.28% <100%> (+0.02%)`	⬆️
pandas/core/algorithms.py	`94.16% <100%> (-0.2%)`	⬇️
pandas/core/sparse/series.py	`95.31% <100%> (+0.1%)`	⬆️
pandas/core/sparse/frame.py	`95.28% <100%> (+0.46%)`	⬆️
pandas/core/sparse/array.py	`91.62% <100%> (+0.3%)`	⬆️
pandas/core/internals.py	`95.04% <92.85%> (-0.5%)`	⬇️
pandas/tseries/plotting.py	`0% <0%> (-100%)`	⬇️
pandas/core/dtypes/base.py	`47.61% <0%> (-44.28%)`	⬇️
pandas/plotting/_compat.py	`62% <0%> (-28.91%)`	⬇️
... and 69 more

Continue to review full report at Codecov.

Legend - Click here to learn more
Δ = absolute <relative> (impact), ø = not affected, ? = missing data
Powered by Codecov. Last update cdfce2b...353171b. Read the comment docs.

hexgnu · 2018-02-05T12:48:05Z

pandas/core/algorithms.py

@@ -1315,6 +1315,11 @@ def take_nd(arr, indexer, axis=0, out=None, fill_value=np.nan, mask_info=None,
        undefined if allow_fill == False and -1 is present in indexer.
    """

+    if is_sparse(arr):


So something that is a bit of a hotspot IMHO is that inside of algos it iterates through the memview and could cause some weird security / segfault issues. This fills that hole but really should fix that at a lower level.

Hmm, this seems sub-optimal. We shouldn't have to densify just to slice, right?

Can you take a look at SparseAray.take to see if any of that logic can be refactored / reused?

And since you're touching performance-related code, could you run the sparse-related ASVs?

thank you for pointing this out. exactly what I was looking for. Will run the ASVs

hexgnu · 2018-02-05T12:48:57Z

pandas/core/internals.py

+            if len(values.shape) == 2:
+                values = values[0]
+            else:
+                assert len(values).shape == 1


This never happens in my testing... so it's probably safe to pull out but it also would be odd if we got to this place.

sorry what I mean to say is... len(values).shape is always either 1 or 2 in my testing. I'm not really sure what to do if shape == 3 or something and it was a true ndarray that was sparse.

hexgnu · 2018-02-05T12:50:26Z

pandas/core/internals.py

@@ -5313,6 +5330,10 @@ def concatenate_block_managers(mgrs_indexers, axes, concat_axis, copy):
    return BlockManager(blocks, axes)


+def is_sparse_join_units(join_units):
+    return all(type(ju.block) is SparseBlock for ju in join_units)


This is very similar (and maybe could be DRYed up) like pd.concat([sparse, sparse])

hexgnu · 2018-02-05T12:51:24Z

pandas/core/internals.py

+
+            # You would think that you want self.block.fill_value here
+            # But in reality that will fill with a bunch of wrong values
+            fill_value = np.nan


this was surprising to me. If you don't pass in np.nan what ends up happening is that if you merge two sparse frames it will concat with a bunch of fill_value's which is most definitely not what I would expect as a user.

Yea this confused me a bunch.

evens = pd.DataFrame({'A': range(0, 20, 2), 'B': range(10)}) threes = pd.DataFrame({'A': range(0, 30, 3), 'B': range(10)}) threes.merge(evens, how="left", on="A")

Yields

A B_x B_y 0 0 0 0.0 1 3 1 NaN 2 6 2 3.0 3 9 3 NaN 4 12 4 6.0 5 15 5 NaN 6 18 6 9.0 7 21 7 NaN 8 24 8 NaN 9 27 9 NaN

As you'd expect for a dense dataframe

but if you sparsify it with a fill_value other than np.nan / None you get weird results

evens = pd.DataFrame({'A': range(0, 20, 2), 'B': range(10)}).to_sparse(fill_value=8675309) threes = pd.DataFrame({'A': range(0, 30, 3), 'B': range(10)}).to_sparse(fill_value=90210) threes.merge(evens, how="left", on="A")

Yields

A B_x B_y 0 0 0 0 1 3 1 8675309 2 6 2 3 3 9 3 8675309 4 12 4 6 5 15 5 8675309 6 18 6 9 7 21 7 8675309 8 24 8 8675309 9 27 9 8675309

Is that expected behavior?

hexgnu · 2018-02-05T12:53:08Z

pandas/core/reshape/merge.py

-                    key_col = Index(lvals).where(~mask, rvals)
+                    # Might need to be IntIndex not Index
+                    if isinstance(lvals, SparseArray):
+                        key_col = Index(lvals.get_values()).where(~mask, rvals)


I'm not sure if this has memory or performance issues. But this is the best solution I could come to with this. The other solution would be to look at using lvals.sp_index and implement a where on it that works.

One thing I have noticed is that IntIndex doesn't act quite like Index which makes for doing these masks tricky in sparse land.

It'd be nice to avoid get_values if possible.

hexgnu · 2018-02-05T12:53:33Z

pandas/core/sparse/array.py

@@ -292,9 +292,9 @@ def __setstate__(self, state):
        self._fill_value = fill_value

    def __len__(self):
-        try:
+        if hasattr(self, 'sp_index'):


This felt really icky to me to try and catch something... Feel free to push back on that change.

How was this failing before?

I have no clue! 😄 I couldn't figure out why the try / except was there in the first place tbh.

This isn't essential to this PR. I just don't understand why this code was here in the first place.

Meaning I can take it out to bring down the code changes

hexgnu · 2018-02-05T12:56:07Z

pandas/core/sparse/frame.py

@@ -73,6 +73,10 @@ def __init__(self, data=None, index=None, columns=None, default_kind=None,
            if columns is None:
                raise Exception("cannot pass a series w/o a name or columns")
            data = {columns[0]: data}
+        elif isinstance(data, BlockManager):
+            fill_value_size = len(set(b.fill_value for b in data.blocks))


@TomAugspurger @jreback

Ok I need feedback on this change. Basically what this does is: if every SparseSeries's fill_value is the same it sets the default_fill_value to that. This to me seems really intuitive, but I trust your judgement. It also makes testing a whole of a hell lot easier which is why I did it.

I would argue pretty heavily that as a user if I added sparse series that all had fill_value=0 then I would expect the sparse data frame to have default_fill_value = 0 as well.

yes this is the only reason to have a standalone SparseDataFrame, meaning that they have the same default fill values. But its not worth the cost generally.

hexgnu · 2018-02-05T12:56:36Z

pandas/core/sparse/series.py

@@ -175,7 +175,7 @@ def values(self):

    def __array__(self, result=None):
        """ the array interface, return my values """
-        return self.block.values
+        return self.block.values.values


This will return the dense version of SparseSeries when calling np.array(sp)

we need some more asv's for sparse I think

hexgnu · 2018-02-05T12:59:21Z

pandas/core/sparse/series.py

+        # Since we are returning a dense representation of
+        # SparseSeries sparse_index might not align when calling
+        # ufunc on the array. There doesn't seem to be a better way
+        # to do this unfortunately.


This wasn't intuitive to me but the array*_ functions defined in numpy don't work quite as expected in certain circumstances. np.abs(s) is the one that was throwing me for a loop.

s = SparseArray([1,-2,2,3], fill_value=-2) np.abs(s)

What would happen to the dense version of the array would be that it would have 3 npoints when really it should only have 2. By zeroing out the sparse_index it fixes the problem, because it relies on the dense index instead going around the problem.

these should simply operate on the sp_values, not the on the densified (or maybe its possible that some classes of ufuncs could operate on the dense, but not sure)

so the expected behavior is np.abs(SparseArray([1,-2,2,3])) == array([1,2,3]) ?

hexgnu · 2018-02-05T13:05:09Z

Also wanted to note that I would have liked to keep PR's separate (one for each bug) but honestly this code was all tied together.

hexgnu · 2018-02-06T09:44:10Z

This is ready for a review now that the tests pass. @TomAugspurger thanks :)

Do you want me to squash commits on here?

TomAugspurger · 2018-02-06T12:16:38Z

No need to squash commits.

Looks like there's a merge conflict in the whatsnew though, if you want to fix that and push again. Taking a look now.

TomAugspurger

Just partway through. Will revisit later.

TomAugspurger · 2018-02-06T12:17:04Z

doc/source/whatsnew/v0.23.0.txt

@@ -238,6 +238,7 @@ Other Enhancements

 - ``IntervalIndex.astype`` now supports conversions between subtypes when passed an ``IntervalDtype`` (:issue:`19197`)
 - :class:`IntervalIndex` and its associated constructor methods (``from_arrays``, ``from_breaks``, ``from_tuples``) have gained a ``dtype`` parameter (:issue:`19262`)
+- :func:`pandas.merge` now supports merging of :class:`SparseDataFrame` (:issue:`13665`)


Clarify: merging sparse to sparse? Sparse and dense?

TomAugspurger · 2018-02-06T12:17:18Z

doc/source/whatsnew/v0.23.0.txt

@@ -555,7 +556,7 @@ Sparse

 - Bug in which creating a ``SparseDataFrame`` from a dense ``Series`` or an unsupported type raised an uncontrolled exception (:issue:`19374`)
 - Bug in :class:`SparseDataFrame.to_csv` causing exception (:issue:`19384`)
-
+- Bug in :class:`SparseSeries.__array__` returning only non-faills (:issue:`13665`)


Typo in "faills"?

TomAugspurger · 2018-02-06T12:21:05Z

pandas/core/algorithms.py

@@ -1315,6 +1315,11 @@ def take_nd(arr, indexer, axis=0, out=None, fill_value=np.nan, mask_info=None,
        undefined if allow_fill == False and -1 is present in indexer.
    """

+    if is_sparse(arr):


Hmm, this seems sub-optimal. We shouldn't have to densify just to slice, right?

Can you take a look at SparseAray.take to see if any of that logic can be refactored / reused?

And since you're touching performance-related code, could you run the sparse-related ASVs?

TomAugspurger · 2018-02-06T12:23:59Z

pandas/core/internals.py

@@ -5304,6 +5305,22 @@ def concatenate_block_managers(mgrs_indexers, axes, concat_axis, copy):
        elif is_uniform_join_units(join_units):
            b = join_units[0].block.concat_same_type(
                [ju.block for ju in join_units], placement=placement)
+        elif is_sparse_join_units(join_units):


Do you ever go down the initial if with sparse arrays?

TomAugspurger · 2018-02-06T12:27:24Z

pandas/core/reshape/merge.py

-                    key_col = Index(lvals).where(~mask, rvals)
+                    # Might need to be IntIndex not Index
+                    if isinstance(lvals, SparseArray):
+                        key_col = Index(lvals.get_values()).where(~mask, rvals)


It'd be nice to avoid get_values if possible.

TomAugspurger · 2018-02-06T12:28:28Z

pandas/core/sparse/array.py

@@ -292,9 +292,9 @@ def __setstate__(self, state):
        self._fill_value = fill_value

    def __len__(self):
-        try:
+        if hasattr(self, 'sp_index'):


How was this failing before?

…parse_frames

jreback

I would wait on this for some more general machinery to exist, mainly the _ndarray_values in @TomAugspurger PR. This should be simpler over time. Further ok on adding some things e.g. fixing __array__ and/or .take which can be split from this.

jreback · 2018-02-12T12:08:11Z

pandas/core/algorithms.py

+                        fill_value=fill_value)
+
+    #    return take_nd(arr.get_values(), indexer, axis=axis, out=out,
+    #                   fill_value=fill_value, mask_info=mask_info,


remove this, make this an elif

jreback · 2018-02-12T12:09:17Z

pandas/core/indexes/base.py

@@ -618,6 +618,9 @@ def where(self, cond, other=None):
        if other is None:
            other = self._na_value

+        if isinstance(other, ABCSparseArray):


we shouldn't have Sparse specific things like this, use ._values

there isn't ._values on SparseArray though? Unless I'm missing something out of @TomAugspurger 's pull request...

jreback · 2018-02-12T12:10:08Z

pandas/core/internals.py

@@ -5304,6 +5305,22 @@ def concatenate_block_managers(mgrs_indexers, axes, concat_axis, copy):
        elif is_uniform_join_units(join_units):
            b = join_units[0].block.concat_same_type(
                [ju.block for ju in join_units], placement=placement)
+        elif is_sparse_join_units(join_units):
+            values = concatenate_join_units(join_units, concat_axis, copy=copy)


this is a mess

yes, yes it is. I'll work on cleaning this up tomorrow since there's too many branches in this code.

jreback · 2018-02-12T12:10:34Z

pandas/core/internals.py

+
+            # You would think that you want self.block.fill_value here
+            # But in reality that will fill with a bunch of wrong values
+            fill_value = np.nan


jreback · 2018-02-12T12:10:52Z

pandas/core/reshape/merge.py

@@ -731,7 +732,11 @@ def _maybe_add_join_keys(self, result, left_indexer, right_indexer):
                if mask.all():
                    key_col = rvals
                else:
-                    key_col = Index(lvals).where(~mask, rvals)
+                    # Might need to be IntIndex not Index


don't do this, use _values

jreback · 2018-02-12T12:11:11Z

pandas/core/sparse/array.py

@@ -292,9 +292,9 @@ def __setstate__(self, state):
        self._fill_value = fill_value

    def __len__(self):
-        try:
+        if hasattr(self, 'sp_index'):


jreback · 2018-02-12T12:12:07Z

pandas/core/sparse/frame.py

@@ -73,6 +73,10 @@ def __init__(self, data=None, index=None, columns=None, default_kind=None,
            if columns is None:
                raise Exception("cannot pass a series w/o a name or columns")
            data = {columns[0]: data}
+        elif isinstance(data, BlockManager):
+            fill_value_size = len(set(b.fill_value for b in data.blocks))


yes this is the only reason to have a standalone SparseDataFrame, meaning that they have the same default fill values. But its not worth the cost generally.

jreback · 2018-02-12T12:12:30Z

pandas/core/sparse/series.py

@@ -175,7 +175,7 @@ def values(self):

    def __array__(self, result=None):
        """ the array interface, return my values """
-        return self.block.values
+        return self.block.values.values


we need some more asv's for sparse I think

jreback · 2018-02-12T12:14:33Z

pandas/core/sparse/series.py

+        # Since we are returning a dense representation of
+        # SparseSeries sparse_index might not align when calling
+        # ufunc on the array. There doesn't seem to be a better way
+        # to do this unfortunately.


these should simply operate on the sp_values, not the on the densified (or maybe its possible that some classes of ufuncs could operate on the dense, but not sure)

hexgnu · 2018-02-12T13:51:40Z

pandas/tests/sparse/test_combine_concat.py

@@ -296,13 +287,13 @@ def test_concat_axis1(self):
        res = pd.concat([sparse, sparse3], axis=1)
        exp = pd.concat([self.dense1, self.dense3],
                        axis=1).to_sparse(fill_value=0)
-        exp._default_fill_value = np.nan
+        # exp._default_fill_value = np.nan


this needs to be taken out fully instead of just commented out

TomAugspurger · 2018-10-17T13:51:54Z

@hexgnu SparseArray.__array__ has been fixed to return a dense ndarray with all the values.

Is there anything remaining in this PR that would be useful?

hexgnu · 2018-10-17T15:24:46Z

@TomAugspurger I don't think there's anything else we should merge in. Let's just close this PR :) Thanks for all your work!

hexgnu added 2 commits January 26, 2018 10:34

Hrm I'm trying here but nothing seems to work

5b48613

First pass at fixing issues with SparseDataFrame merging

555fb91

hexgnu commented Feb 1, 2018

View reviewed changes

jreback added Reshaping Concat, Merge/Join, Stack/Unstack, Explode Sparse Sparse Data Type labels Feb 1, 2018

TomAugspurger reviewed Feb 1, 2018

View reviewed changes

hexgnu added 12 commits February 2, 2018 09:29

Fix linting errors

89677b0

One more linting error

97bd19b

Fixing more test failures

b04ac64

Add more tests and fix some existing ones

d24e464

Only allow for sparse merging if everything is sparse

a796280

Some more tests

be09289

Merge remote-tracking branch 'upstream/master' into dense_array_sparse

1aef901

Merge branch 'dense_array_sparse' into fix_for_merging_sparse_frames

42680ba

Fix problem with __array__ not showing dense values

2084ed3

Fix some linting errors

77c41b7

I think I fixed it

25fd08a

Don't assume that the dtype is int64

cd583f7

Get rid of WTF from code

45e7cd3

hexgnu commented Feb 5, 2018

View reviewed changes

Missed one point where assumed int64

6522d6b

hexgnu added 3 commits February 6, 2018 05:57

Typo fix should be len(values.shape)

029d37b

This will fix windows bug on appveyor

730c152

Linting error

171f5dd

hexgnu changed the title ~~First pass at fixing issues with SparseDataFrame merging~~ Allows for merging of SparseDataFrames, and fixes __array__ interface Feb 6, 2018

hexgnu added 2 commits February 6, 2018 16:49

Add whatsnew entry for everything

bde2588

Add docstring to is_sparse_join_units

bfe3065

TomAugspurger reviewed Feb 6, 2018

View reviewed changes

hexgnu added 4 commits February 7, 2018 10:01

WIP for merging sparse frames

d13daa7

Merge remote-tracking branch 'upstream/master' into fix_for_merging_s…

3ddba8b

…parse_frames

Allow for indexes to where on SparseArray

4fa7dec

Rely on ABCSparseArray over SparseArray

353171b

jreback requested changes Feb 12, 2018

View reviewed changes

hexgnu commented Feb 12, 2018

View reviewed changes

hexgnu closed this Oct 17, 2018

Allows for merging of SparseDataFrames, and fixes __array__ interface #19488

Allows for merging of SparseDataFrames, and fixes __array__ interface #19488

Conversation

hexgnu commented Feb 1, 2018 • edited Loading

pep8speaks commented Feb 1, 2018 • edited Loading

Comment last updated on February 12, 2018 at 05:07 Hours UTC

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

codecov bot commented Feb 5, 2018 • edited Loading

Codecov Report

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

hexgnu commented Feb 5, 2018

hexgnu commented Feb 6, 2018

TomAugspurger commented Feb 6, 2018

TomAugspurger left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jreback left a comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

TomAugspurger commented Oct 17, 2018

hexgnu commented Oct 17, 2018

Allows for merging of SparseDataFrames, and fixes array interface #19488

Allows for merging of SparseDataFrames, and fixes array interface #19488

hexgnu commented Feb 1, 2018 •

edited

Loading

pep8speaks commented Feb 1, 2018 •

edited

Loading

codecov bot commented Feb 5, 2018 •

edited

Loading