BUG: fix construction of Series from dict with nested lists #18626

toobaz · 2017-12-04T10:40:22Z

closes Initializing Series with dict, list of lists values and explicit dtype raises ValueError #18625
tests added / passed
passes git diff upstream/master -u -- "*.py" | flake8 --diff
whatsnew entry

jreback · 2017-12-04T11:11:12Z

pandas/core/series.py

@@ -3208,7 +3208,11 @@ def _try_cast(arr, take_fast_path):
        return subarr

    elif isinstance(data, (list, tuple)) and len(data) > 0:
-        if dtype is not None:
+        if all(is_list_like(item) for item in data):


not sure this is the appropriate place for this.

(let me try to persuade you that other things were out of place!)

For performance, can't we first try the _try_cast and only if that fails / does not give an appropriate sized subarr do this check of all lists?

Not sure this would be simple to do... but anyway, notice that

if data is "sane" (e.g. first element is not a list), then that line has a very small performance cost, since it is lazily evaluated

while if data is a list of lists, then going through _try_cast could have a performance cost (but most importantly from my point of view, behavior would be more complex to analyze)

Just tested:

In [79]: %timeit np.array(data) 6.17 ms ± 53.3 µs per loop (mean ± std. dev. of 7 runs, 100 loops each) In [80]: %timeit all(pd.api.types.is_list_like(item) for item in data) 2.2 µs ± 24.3 ns per loop (mean ± std. dev. of 7 runs, 100000 loops each)

and the time for such checking is actually negligible compared to the time that the actual conversion to array takes. So indeed not needed for performance reasons.

(if it would not have been the case, I think keeping 'normal' list -> series fast would be higher in priority than the possible slowdown in converting list of lists)

(if it would not have been the case, I think keeping 'normal' list -> series fast would be higher in priority than the possible slowdown in converting list of lists)

I totally agree

By the way: if/when there will be an ndmax argument to np.array, it will allow us to speed up further the array creation.

I would like to move this loggic to _init_listlike or something similar.

closes pandas-dev#18626

codecov · 2017-12-04T13:52:16Z

Codecov Report

Merging #18626 into master will decrease coverage by 0.17%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #18626      +/-   ##
==========================================
- Coverage   91.59%   91.42%   -0.18%     
==========================================
  Files         155      157       +2     
  Lines       51255    51457     +202     
==========================================
+ Hits        46949    47046      +97     
- Misses       4306     4411     +105

Flag	Coverage Δ
#multiple	`89.29% <100%> (-0.15%)`	⬇️
#single	`40.6% <75%> (-0.18%)`	⬇️

Impacted Files	Coverage Δ
pandas/core/series.py	`94.82% <100%> (+0.01%)`	⬆️
pandas/io/gbq.py	`25% <0%> (-58.34%)`	⬇️
pandas/io/excel.py	`80.39% <0%> (-9.74%)`	⬇️
pandas/plotting/_converter.py	`63.44% <0%> (-1.82%)`	⬇️
pandas/compat/__init__.py	`58.18% <0%> (-0.59%)`	⬇️
pandas/core/frame.py	`97.81% <0%> (-0.1%)`	⬇️
pandas/core/sparse/api.py	`100% <0%> (ø)`	⬆️
pandas/compat/openpyxl_compat.py	`90.9% <0%> (ø)`
pandas/core/sparse/list.py	`97.1% <0%> (ø)`
pandas/core/tools/datetimes.py	`84.48% <0%> (+0.05%)`	⬆️
... and 3 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 2c903d5...c5015c1. Read the comment docs.

codecov · 2017-12-04T13:52:27Z

Codecov Report

Merging #18626 into master will increase coverage by 0.03%.
The diff coverage is 100%.

@@            Coverage Diff             @@
##           master   #18626      +/-   ##
==========================================
+ Coverage   91.82%   91.85%   +0.03%     
==========================================
  Files         152      152              
  Lines       49255    49252       -3     
==========================================
+ Hits        45226    45239      +13     
+ Misses       4029     4013      -16

Flag	Coverage Δ
#multiple	`90.23% <100%> (+0.03%)`	⬆️
#single	`41.91% <90%> (ø)`	⬆️

Impacted Files	Coverage Δ
pandas/core/series.py	`94.07% <100%> (+0.17%)`	⬆️
pandas/util/testing.py	`84.73% <0%> (+0.2%)`	⬆️
pandas/plotting/_converter.py	`66.81% <0%> (+1.73%)`	⬆️

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 4efb39f...14f0443. Read the comment docs.

toobaz · 2017-12-04T23:48:44Z

The new version scans nested types once rather than twice, but there is room for improvements, which I will push tomorrow.

jreback · 2017-12-05T00:18:32Z

pandas/core/series.py

@@ -3177,7 +3177,8 @@ def _try_cast(arr, take_fast_path):

    # GH #846
    if isinstance(data, (np.ndarray, Index, Series)):
-
+        if data.ndim > 1:


actually should error on data.ndim != 1 (e.g. a numpy scalar of ndim==0 I think hits this path maybe)

can you add tests to cover this case

Shouldn't we allow numpy scalars? (just like normal scalars work, eg like Series(1))

Sure we do! And this is widely tested (as opposed to the initialization with length 1 list-likes, which was working by accident, accidentally used in the tests, and I now disabled).

closes pandas-dev#18626

toobaz · 2017-12-05T14:18:17Z

pandas/tests/reshape/test_reshape.py

@@ -462,7 +462,7 @@ def test_preserve_categorical_dtype(self):
            cidx = pd.CategoricalIndex(list("xyz"), ordered=ordered)
            midx = pd.MultiIndex(levels=[['a'], cidx],
                                 labels=[[0, 0], [0, 1]])
-            df = DataFrame([[10, 11]], index=midx)
+            df = DataFrame([[10, 11]] * 2, index=midx, columns=[0, 1])


This columns=[0,1] is a typo. Waiting for any comment on the rest, then I will remove it.

jreback · 2017-12-06T01:00:11Z

pandas/core/series.py

@@ -3208,7 +3208,11 @@ def _try_cast(arr, take_fast_path):
        return subarr

    elif isinstance(data, (list, tuple)) and len(data) > 0:
-        if dtype is not None:
+        if all(is_list_like(item) for item in data):


I would like to move this loggic to _init_listlike or something similar.

jreback · 2017-12-06T01:00:59Z

pandas/core/series.py

-            value = data
+    if subarr.ndim == 0 or is_scalar(data):
+        if index is None:
+            return subarr.item()


this is very odd, are you actually returning a scalar on construction?

I have no idea of what could possibly hit this (i.g. when index is None), and it does look broken (e.g. BlockManager won't accept it), so I removed it, but if you want to stay on the safe side just discard the last commit I just pushed.

jreback · 2017-12-06T01:01:21Z

pandas/core/series.py


+        if subarr.ndim == 1:
+            # a scalar upcasted to 1-dimensional by maybe_cast_to_datetime()


so would much rather fix this case inside maybe_cast_to_datetime

My (limited) understanding is that it is impossible, basically because

def broken_datetimeindex_roundtrip(dtix): scalar = dtix.values[0, ...] return pd.DatetimeIndex(np.array([scalar], dtype=scalar.dtype))

looses tz and freq.
This in turn is probably not trivial at all to fix (or more honestly: I have no idea).

looses tz and freq.

Uhm... it probably makes perfect sense that it looses freq (but not tz).

jreback · 2017-12-06T01:01:40Z

pandas/tests/indexes/test_multi.py

@@ -3066,8 +3066,7 @@ def test_nan_stays_float(self):
        assert pd.isna(idx0.get_level_values(1)).all()
        # the following failed in 0.14.1
        assert pd.isna(idxm.get_level_values(1)[:-1]).all()
-
-        df0 = pd.DataFrame([[1, 2]], index=idx0)


what is the issue here?

the (outer) list has a different length than the index

jreback · 2017-12-06T01:04:13Z

pandas/tests/series/test_constructors.py

@@ -651,6 +656,17 @@ def test_constructor_dict(self):
        expected.iloc[1] = 1
        assert_series_equal(result, expected)

+    @pytest.mark.parametrize('input_class', [list, tuple])


I believe we also accept a duck-typed list class that is not a generator here? (IOW an iterable).

closes pandas-dev#18626

toobaz · 2017-12-06T10:15:35Z

pandas/core/common.py

@@ -381,25 +401,17 @@ def _asarray_tuplesafe(values, dtype=None):
        return values.values

    if isinstance(values, list) and dtype in [np.object_, object]:
-        return lib.list_to_object_array(values)
+        return _as_1d_array(values)


Notice this is more efficient than the cython version.

jreback · 2017-12-06T10:49:32Z

can u run the series construction asv for this
we may not have enough coverage of the various cases as well

closes pandas-dev#18626

toobaz · 2017-12-06T14:03:27Z

can u run the series construction asv for this

       before           after         ratio
     [fdba1333]       [b5225ee5]
-         263±4ms        235±0.6ms     0.89  series_methods.series_constructor_no_data_datetime_index.time_series_constructor_no_data_datetime_index
-     1.67±0.02ms      1.41±0.02ms     0.84  series_methods.series_value_counts.time_value_counts_int64

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.

(but they are borderline, they don't appear on every run)

jreback · 2017-12-06T14:27:05Z

can u show the series constructions ones
i don’t thing we cover other than the basic cases

toobaz · 2017-12-06T16:53:11Z

can u show the series constructions ones

Sorry, I had missed ctors.py, here is the result:

     [fdba1333]       [b5225ee5]
-     13.8±0.07μs      12.4±0.08μs     0.90  ctors.Constructors.time_series_from_ndarray
-     11.0±0.03μs      9.92±0.05μs     0.90  ctors.Constructors.time_dtindex_from_series
-        25.2±3μs      22.3±0.05μs     0.88  ctors.Constructors.time_dtindex_from_index_with_series

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.

These are from frame_ctor.py instead:

       before           after         ratio
     [fdba1333]       [b5225ee5]
-     3.98±0.02ms      3.61±0.01ms     0.91  frame_ctor.FromDicts.time_series_ctor_from_dict
-      76.0±0.4ms       68.3±0.2ms     0.90  frame_ctor.FromDicts.time_frame_ctor_nested_dict_int64
-      33.6±0.2ms       30.1±0.1ms     0.90  frame_ctor.FromDictwithTimestampOffsets.time_frame_ctor('QuarterBegin', 1)
-      30.5±0.4ms       27.3±0.2ms     0.90  frame_ctor.FromDictwithTimestampOffsets.time_frame_ctor('CDay', 2)
-      30.4±0.2ms      27.1±0.08ms     0.89  frame_ctor.FromDictwithTimestampOffsets.time_frame_ctor('CustomBusinessDay', 2)
-      30.8±0.1ms       27.4±0.1ms     0.89  frame_ctor.FromDictwithTimestampOffsets.time_frame_ctor('BusinessDay', 1)
-      30.6±0.3ms      27.1±0.07ms     0.89  frame_ctor.FromDictwithTimestampOffsets.time_frame_ctor('CDay', 1)
-      30.7±0.2ms      27.0±0.07ms     0.88  frame_ctor.FromDictwithTimestampOffsets.time_frame_ctor('BDay', 2)
-      31.3±0.2ms       27.5±0.3ms     0.88  frame_ctor.FromDictwithTimestampOffsets.time_frame_ctor('Day', 2)
-      32.0±0.4ms       28.0±0.5ms     0.87  frame_ctor.FromDictwithTimestampOffsets.time_frame_ctor('CBMonthEnd', 1)
-      32.2±0.5ms       27.8±0.3ms     0.86  frame_ctor.FromDictwithTimestampOffsets.time_frame_ctor('Day', 1)

SOME BENCHMARKS HAVE CHANGED SIGNIFICANTLY.

jreback · 2017-12-07T01:08:35Z

those benchmarks are mostly not on point. e.g. to we have ones which test list-of-list, dict-of-list, dict-of-array, dict-of-series etc. for frame

list, tuple, dict, array, Index, Series for series

jreback · 2017-12-07T01:10:28Z

pandas/core/common.py

@@ -372,6 +372,26 @@ def intersection(*seqs):
    return type(seqs[0])(list(result))


+def _as_1d_array(values, dtype='object'):


move to pandas/core/dtypes/cast.py (at the bottom); rename in that style as well.

can move _asarray_tuplesafe as well (here or future PR)

jreback · 2017-12-07T01:10:51Z

pandas/core/common.py

+    -------
+    1-dimensional numpy array
+    """
+    # Making a 1D array that safely contains list-likes is a bit tricky


explain why if you are leaving a comment like this

jreback · 2017-12-07T01:13:15Z

pandas/core/common.py


    result = np.asarray(values, dtype=dtype)

    if issubclass(result.dtype.type, compat.string_types):
        result = np.asarray(values, dtype=object)

    if result.ndim == 2:
-        if isinstance(values, list):
-            return lib.list_to_object_array(values)


so is seems like list basically replaces lib.list_to_object_array if that is so, then we should simply do that (it can live in pandas.core.dtypes.cast as well).

if hasattr(values, '__array__'): values = [tuple(x) for x in values result = _as_1d_array(values)

Yes, _as_1d_array can entirely replace lib.list_to_object_array. My only doubt is that the latter is used in some cython files... I have no idea if in those cases using a cython function has advantages.

Sorry, I had missed the second part of your comment.And I don't understand it (isn't it what I'm doing?)

jreback · 2017-12-07T01:13:33Z

pandas/core/series.py

@@ -3212,7 +3214,9 @@ def _try_cast(arr, take_fast_path):
        return subarr

    elif isinstance(data, (list, tuple)) and len(data) > 0:
-        if dtype is not None:
+        if all(is_list_like(item) for item in data):
+            subarr = _as_1d_array(data)


see comment above

jreback · 2017-12-07T01:14:24Z

pandas/core/series.py

+    if subarr.ndim == 0 or is_scalar(data):
+        if subarr.ndim == 1:
+            # a scalar upcasted to 1-dimensional by maybe_cast_to_datetime()
+            value = subarr[0]


how is this branch ever hit?

pd.Series(pd.Timestamp('2015-10-10'), index=range(3))

toobaz · 2017-12-07T07:23:36Z

those benchmarks are mostly not on point. e.g. to we have ones which test

For frames there's the entire frame_ctor.py, but for Series all I could find was series_constructor_dict_data_datetime_index in series_methods.py and time_series_from_ndarray in ctors.py. They just didn't exhibit any significant change.

jorisvandenbossche · 2017-12-07T14:58:09Z

For the change of list of one element (previously broadcasted, now error), I would do this as deprecation instead of a hard breaking change. Thoughts?
(it is quite easy to do for us I think, and if even Tom writes code using that .. :-))

toobaz · 2017-12-07T15:18:37Z

(it is quite easy to do for us I think, and if even Tom writes code using that .. :-))

You mean the tests? They look to me as accidental uses of this "feature" - after all, just passing the scalar is simpler.

Anyway, I'm a bit against deprecation in this case because the docs are quite clear, but sure, it can be done.

jorisvandenbossche · 2017-12-07T15:21:03Z

They look to me as accidental uses of this "feature" - after all, just passing the scalar is simpler.

Yes, accidental, but many people can have done it "accidental", and since it works perfectly there was no incentive to fix your code.

jreback · 2017-12-08T02:05:29Z

For frames there's the entire frame_ctor.py, but for Series all I could find was series_constructor_dict_data_datetime_index in series_methods.py and time_series_from_ndarray in ctors.py. They just didn't exhibit any significant change.

ok, as indicated above, pls add some.

jreback · 2017-12-08T02:06:25Z

pandas/core/dtypes/cast.py

@@ -1162,3 +1162,23 @@ def construct_1d_arraylike_from_scalar(value, length, dtype):
        subarr.fill(value)

    return subarr
+
+
+def cast_to_1d_array(values, dtype='object'):


rename like the above, e.g.

construct_1d_array_from_listlike

jreback · 2017-12-08T02:06:44Z

pandas/core/dtypes/cast.py

+
+    Returns
+    -------
+    1-dimensional numpy array


of object type

jreback · 2017-12-08T02:07:47Z

pandas/core/dtypes/cast.py

+    1-dimensional numpy array
+    """
+    # numpy will try to interpret nested lists as further dimensions, hence
+    # making a 1D array that contains list-likes is a bit tricky:


can you add some tests for this, e.g. cases would be iterables, including a list of strings with one or more np.nan elements, and throw in some Nones

jreback · 2018-01-21T18:24:06Z

rebase and will have a look

jorisvandenbossche · 2018-02-09T15:13:00Z

@toobaz status of this?

toobaz · 2018-02-09T23:42:56Z

@toobaz status of this?

I'll rebase on #18600 when merged

toobaz · 2018-04-02T21:22:59Z

Rebased.

We need to decide what to do about the "length 1 = scalar" thing. I thought we could just break it since it was never documented, is conceptually wrong, and supporting it complicates our code. Joris suggests a deprecation cycle instead. Could a compromise make sense, in which we remove support but also insert (for a couple of releases) an informative warning saying "broadcasting [...] was never officially supported, but it worked accidentally, support was dropped in version 0.23, replace EXAMPLE_CODE with FIXED_EXAMPLE")?

jreback · 2018-04-02T21:45:42Z

we already changed the length 1 = scalar iirc

are you suggesting we should revert and deprecate instead?

toobaz · 2018-04-02T22:38:26Z

we already changed the length 1 = scalar iirc

Sorry, I got confused. True, we have disabled pd.Series([1], index=range(3)). What I should have referred to is pd.DataFrame([[1]], index=range(3)), which still works, and this PR disables.

However, the thing is slightly more complicated than I thought. My idea was that (in analogy with the Series case) pd.DataFrame([[1]], index=range(3)) should have been replaced with pd.DataFrame([1], index=range(3)) (as in "broadcast if you have lower dimensional value"). However, that won't work, because [1] gets aligned vertically. And by the way, this "feature" is also broken, since pd.DataFrame([1, 2, 3], index=range(3)) works (but could be simply be written as pd.DataFrame([[1], [2], [3]], index=range(3))), while pd.DataFrame([1, 2, 3], columns=20) (which would be the useful case) doesn't.

So, ideal world: if a lower dim object obj is passed as data, it is interpreted as [obj] * len(index), both in Series and in DataFrame.

Acceptable world: Series remain as they are, but a 1d list obj passed as data for a DataFrame is interpreted as [[i]*len(columns) for i in obj].

If the ideal world is too disruptive (backward incompatible), we can at least aim for the acceptable one. In this case, I'm no more sure we want to drop support for pd.DataFrame([[1]], index=range(3)).

toobaz · 2018-04-27T06:32:47Z

However, the thing is slightly more complicated than I thought.

Answered myself in #20837 . Will update this PR accordingly (not in time for 0.23.0).

jreback · 2018-09-25T16:25:31Z

can you rebase & move note to 0.24

jreback · 2018-11-23T03:31:32Z

can you rebase

jreback · 2018-12-14T15:52:16Z

closing as stale, but pls reopen if can continue

jreback requested changes Dec 4, 2017

View reviewed changes

toobaz force-pushed the series_of_lists_of_lists_18625 branch from a45ce6c to c5015c1 Compare December 4, 2017 13:52

toobaz added a commit to toobaz/pandas that referenced this pull request Dec 4, 2017

BUG: fix construction of Series from dict with nested lists

c5015c1

closes pandas-dev#18626

jreback added Bug Dtype Conversions Unexpected or buggy dtype conversions Reshaping Concat, Merge/Join, Stack/Unstack, Explode labels Dec 5, 2017

jreback requested changes Dec 5, 2017

View reviewed changes

toobaz force-pushed the series_of_lists_of_lists_18625 branch from c5015c1 to d81ce3a Compare December 5, 2017 10:55

toobaz added a commit to toobaz/pandas that referenced this pull request Dec 5, 2017

BUG: fix construction of Series from dict with nested lists

4952a9b

closes pandas-dev#18626

toobaz force-pushed the series_of_lists_of_lists_18625 branch from d81ce3a to 16e21fd Compare December 5, 2017 10:59

toobaz added a commit to toobaz/pandas that referenced this pull request Dec 5, 2017

BUG: fix construction of Series from dict with nested lists

66caa72

closes pandas-dev#18626

toobaz commented Dec 5, 2017

View reviewed changes

jreback requested changes Dec 6, 2017

View reviewed changes

toobaz force-pushed the series_of_lists_of_lists_18625 branch from 16e21fd to 707abf2 Compare December 6, 2017 10:12

toobaz added a commit to toobaz/pandas that referenced this pull request Dec 6, 2017

BUG: fix construction of Series from dict with nested lists

40d26e4

closes pandas-dev#18626

toobaz commented Dec 6, 2017

View reviewed changes

toobaz force-pushed the series_of_lists_of_lists_18625 branch from 707abf2 to b5225ee Compare December 6, 2017 13:59

toobaz added a commit to toobaz/pandas that referenced this pull request Dec 6, 2017

BUG: fix construction of Series from dict with nested lists

b90bd9f

closes pandas-dev#18626

toobaz mentioned this pull request Dec 6, 2017

Broken roundrip DatetimeIndex -> CategoricalIndex -> DatetimeIndex #18664

Closed

jreback requested changes Dec 7, 2017

View reviewed changes

toobaz force-pushed the series_of_lists_of_lists_18625 branch from b5225ee to f56134b Compare December 7, 2017 07:23

jreback requested changes Dec 8, 2017

View reviewed changes

This was referenced Dec 13, 2017

Construct 1d array from listlike #18769

Merged

len-1 scalar is accepted as valid input for len >1 Series #18819

Closed

BUG: avoid unnecessary casting when unstacking index with unused levels #18460

Merged

toobaz added 4 commits April 2, 2018 11:26

BUG: fix construction of Series from dict with nested lists

9ef97f7

TST: fixed invalid DataFrame() initializations in tests

f8ac3e6

TST: invalid construction of Series() from list of len != len(index)

c6aa204

CLN: remove impossible "Index is None" case

14f0443

toobaz force-pushed the series_of_lists_of_lists_18625 branch from f56134b to 14f0443 Compare April 2, 2018 10:02

toobaz mentioned this pull request Apr 27, 2018

Allow broadcasting vertically a 1-dim input to pd.DataFrame(), - and document #20837

Open

jreback closed this Dec 14, 2018


		if subarr.ndim == 1:
		# a scalar upcasted to 1-dimensional by maybe_cast_to_datetime()

		@@ -372,6 +372,26 @@ def intersection(*seqs):
		return type(seqs[0])(list(result))


		def _as_1d_array(values, dtype='object'):

BUG: fix construction of Series from dict with nested lists #18626

BUG: fix construction of Series from dict with nested lists #18626

Conversation

toobaz commented Dec 4, 2017

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 Dec 4, 2017

Codecov Report

codecov bot commented Dec 4, 2017 • edited Loading

Codecov Report

toobaz commented Dec 4, 2017

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

toobaz Dec 5, 2017 • edited Loading

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

jreback commented Dec 6, 2017

toobaz commented Dec 6, 2017

jreback commented Dec 6, 2017

toobaz commented Dec 6, 2017

jreback commented Dec 7, 2017 • edited Loading

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

toobaz commented Dec 7, 2017 • edited Loading

jorisvandenbossche commented Dec 7, 2017

toobaz commented Dec 7, 2017 • edited Loading

jorisvandenbossche commented Dec 7, 2017

jreback commented Dec 8, 2017

Choose a reason for hiding this comment

Choose a reason for hiding this comment

Choose a reason for hiding this comment

jreback commented Jan 21, 2018

jorisvandenbossche commented Feb 9, 2018

toobaz commented Feb 9, 2018

toobaz commented Apr 2, 2018

jreback commented Apr 2, 2018

toobaz commented Apr 2, 2018

toobaz commented Apr 27, 2018

jreback commented Sep 25, 2018

jreback commented Nov 23, 2018

jreback commented Dec 14, 2018

codecov bot commented Dec 4, 2017 •

edited

Loading

toobaz Dec 5, 2017 •

edited

Loading

jreback commented Dec 7, 2017 •

edited

Loading

toobaz commented Dec 7, 2017 •

edited

Loading

toobaz commented Dec 7, 2017 •

edited

Loading