Rolling window endpoints inclusion #15795

carlosdanielcsantos · 2017-03-23T22:29:19Z

closes Exact matches in time-based .rolling() #13965
time-based window tests added / passed
fixed window tests added / passed
passes git diff upstream/master --name-only -- '*.py' | flake8 --diff
whatsnew entry

carlosdanielcsantos · 2017-03-23T22:37:43Z

pandas/core/window.pyx


 cdef _roll_min_max(ndarray[numeric] input, int64_t win, int64_t minp,
- object index, bint is_max):
+ object index, str closed, bint is_max):


Not sure about the behavior of str here (cdef)

carlosdanielcsantos · 2017-03-23T22:44:06Z

Here is a quick example of what this feature looks like

df = pd.DataFrame({'x': [1]*5}, ....: index = [pd.Timestamp('20130101 09:00:01'), ....: pd.Timestamp('20130101 09:00:02'), ....: pd.Timestamp('20130101 09:00:03'), ....: pd.Timestamp('20130101 09:00:04'), ....: pd.Timestamp('20130101 09:00:06')]) df["right"] = df.rolling('2s', closed='right').x.sum() # current behavior df["both"] = df.rolling('2s', closed='both').x.sum() df["left"] = df.rolling('2s', closed='left').x.sum() df["open"] = df.rolling('2s', closed='neither').x.sum() print(df) x right both left open 2013-01-01 09:00:01 1 1.0 1.0 NaN NaN 2013-01-01 09:00:02 1 2.0 2.0 1.0 1.0 2013-01-01 09:00:03 1 2.0 3.0 2.0 1.0 2013-01-01 09:00:04 1 2.0 3.0 2.0 1.0 2013-01-01 09:00:06 1 1.0 2.0 1.0 NaN

jreback

mostly minor points

add a whatsnew entry (Other Enhancements) in whatsnew (0.20.0), with a link to the computation.rst docs (where you have a new section showing the use of closed; you can use your test examples, e.g. showing all 4 cases).

jreback · 2017-03-23T22:39:11Z

pandas/core/window.py

 if self.min_periods is not None and not \
 is_integer(self.min_periods):
 raise ValueError("min_periods must be an integer")
+ if self.closed not in ['right', 'both', 'left', 'neither']:


testing this?

jreback · 2017-03-23T22:40:13Z

pandas/core/window.pyx

 def __init__(self, ndarray input, int64_t win, int64_t minp,
- object index=None, object floor=None):
+ object index=None, object floor=None,
+ bint l_closed=False, bint r_closed=True):


if you have a chance, can you add a mini-doc strings to things.

jreback · 2017-03-23T22:41:18Z

pandas/core/window.pyx

 if we are a fixed indexer, return a mock indexer
 instead of the FixedWindow Indexer. This is a type
 compat Indexer that allows us to use a standard
 code path with all of the indexers.


can you augment the doc-string here (with closed)

jreback · 2017-03-23T22:41:55Z

pandas/core/window.pyx

+ bint l_closed = False
+ bint r_closed = False
+
+ if closed not in ['right', 'left', 'both', 'neither']:


you can just do an assert here (this is an internal routine that has certain guarantees)

jreback · 2017-03-23T22:42:20Z

pandas/core/window.pyx


 def roll_count(ndarray[double_t] input, int64_t win, int64_t minp,
- object index):
+ object index, closed='right'):


just make this required

object closed

jreback · 2017-03-23T22:42:47Z

pandas/core/window.pyx


 def roll_max(ndarray[numeric] input, int64_t win, int64_t minp,
- object index):
+ object index, str closed):


same for the rest of these: str-> object

jreback · 2017-03-23T22:42:57Z

pandas/core/window.pyx


 cdef _roll_min_max(ndarray[numeric] input, int64_t win, int64_t minp,
- object index, bint is_max):
+ object index, str closed, bint is_max):


jreback · 2017-03-23T22:43:09Z

pandas/tests/test_window.py

 tm.assert_frame_equal(result, expected)

+ def test_closed(self):
+ df = DataFrame({'A': [1]*5},


add this issue number as a comment

jreback · 2017-03-23T22:43:18Z

pandas/tests/test_window.py

+ df = DataFrame({'A': [1]*5},
+ index = [pd.Timestamp('20130101 09:00:01'),
+ pd.Timestamp('20130101 09:00:02'),
+ pd.Timestamp('20130101 09:00:03'),


test passing invalid closed

jreback · 2017-03-23T22:43:49Z

pandas/tests/test_window.py

+ expected["A"] = [np.nan, 1.0, 2, 2, 1]
+ result = df.rolling('2s', closed='left').sum()
+ tm.assert_frame_equal(result, expected)
+


@chrisaycock how do these tests look? what corners do we need to test here

What happens if the DataFrame is empty?

If the DataFrame is empty, the behavior is exactly the same as in the current implementation. For time-based windows, it throws "ValueError: window must be an integer". Note that this feature only affects the window indexer's start and end vectors. In fixed windows, empty DataFrames always return empty (the result is constructed as np.empty(N) where N is the length of the input)

separately I think [4] should just return an empty frame for consistency.

In [4]: DataFrame().rolling('1s').sum() ValueError: window must be an integer In [5]: DataFrame().rolling(1).sum() Out[5]: Empty DataFrame Columns: [] Index: []

@chrisaycock ?

jreback · 2017-03-23T22:46:26Z

@carlosdanielcsantos #15795 (comment) are basically the docs (with a bit of prose) and nice formatting. Laying out the possibilities (in a sub-section of computation.rst). Note you also might want to have a 'real-world' uses case sentence for some of these.

jreback · 2017-03-23T22:54:19Z

you can see pep errors: http://pandas-docs.github.io/pandas-docs-travis/contributing.html#python-pep8

use git diff master | flake8 --diff

codecov · 2017-03-27T12:56:42Z

Codecov Report

Merging #15795 into master will decrease coverage by 0.02%.
The diff coverage is 92.85%.

@@ Coverage Diff @@ ## master #15795 +/- ## ========================================== - Coverage 91.01% 90.99% -0.03%  ========================================== Files 143 143 Lines 49400 49404 +4 ========================================== - Hits 44963 44956 -7  - Misses 4437 4448 +11

Impacted Files	Coverage Δ
pandas/core/generic.py	`96.24% <100%> (ø)`	⬆️
pandas/core/window.py	`96.08% <92.3%> (-0.11%)`	⬇️
pandas/io/gbq.py	`25% <0%> (-58.34%)`	⬇️
pandas/core/common.py	`90.96% <0%> (-0.34%)`	⬇️
pandas/core/frame.py	`97.86% <0%> (-0.1%)`	⬇️

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 9d3554c...8533421. Read the comment docs.

codecov · 2017-03-27T12:56:43Z

Codecov Report

Merging #15795 into master will increase coverage by 0.02%.
The diff coverage is 100%.

@@ Coverage Diff @@ ## master #15795 +/- ## ========================================== + Coverage 91% 91.03% +0.02%  ========================================== Files 145 145 Lines 49581 49592 +11 ========================================== + Hits 45122 45146 +24  + Misses 4459 4446 -13

Flag	Coverage Δ
#multiple	`88.8% <100%> (+0.02%)`	⬆️
#single	`40.56% <0%> (-0.01%)`	⬇️

Impacted Files	Coverage Δ
pandas/core/generic.py	`96.26% <ø> (ø)`	⬆️
pandas/core/window.py	`96.24% <100%> (+0.02%)`	⬆️
pandas/core/common.py	`90.68% <0%> (-0.35%)`	⬇️
pandas/util/testing.py	`81.13% <0%> (+0.18%)`	⬆️
pandas/tools/plotting.py	`72.47% <0%> (+0.71%)`	⬆️

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 838e09c...aad97dc. Read the comment docs.

carlosdanielcsantos · 2017-03-27T15:02:54Z

@jreback How should I add this to computations.rst? A new section inside Window Functions or just add the examples inside the existing Window Functions -> Rolling Windows?

jreback · 2017-03-27T15:03:39Z

doc/source/whatsnew/v0.20.0.txt

 Other enhancements
 ^^^^^^^^^^^^^^^^^^

+- ``DataFrame.rolling()`` now accepts the parameter ``closed='right'|'left'|'both'|'neither'`` to choose the rolling window endpoint closedness. (:issue:`13965`)


is that a word?

There is some discussion going on about that online, but its a useful word to refer the property of being closed. But I am open to suggestions, no pun intended :)

hahah yep that is fine

maybe closed-ness? (though I don't think that is actually a word)

also add a :ref: to the new sub-section that you are creating in computation.rst

jreback · 2017-03-27T15:04:05Z

pandas/core/window.py

 on : string, optional
 For a DataFrame, column on which to calculate
 the rolling window, rather than the index
+ closed : 'right', 'left', 'both', 'neither'


add the default setting (right)

I would do this with a bulleted list

jreback · 2017-03-27T15:05:13Z

pandas/core/window.pyx

 floor: optional
 unit for flooring
+ l_closed: bint
+ left endpoint closedness


I would write these out l_closed -> left_closed easier to read

jreback · 2017-03-27T15:05:54Z

pandas/core/window.pyx

 index: ndarray
 index of the input
+ l_closed: bint
+ left endpoint closedness


can add a bit of explication here (1 line), just for future readers

jreback · 2017-03-27T15:06:32Z

pandas/core/window.pyx


 start[0] = 0
- end[0] = 1
+


put the comments above the code

if r_closed: # ..... end[0] = 1 else: # .... end[0] 0

jreback · 2017-03-27T15:06:38Z

pandas/core/window.pyx

 end_bound = index[i]
 start_bound = index[i] - win

+ if l_closed: # left endpoint is closed


of can do (I use this if only an if statement)

# left endpoint .... if left_closed: ....

jreback · 2017-03-27T15:07:14Z

pandas/core/window.pyx

 else:
 end[i] = end[i - 1]

+ # right endpoint is open


yep excatly like this one

jreback · 2017-03-27T15:07:50Z

pandas/core/window.pyx

 minp: integer, minimum periods
 index: 1d ndarray, optional
 index to the input array
+ closed: 'right', 'left', 'both', 'neither'


closed: string, default 'right' {'right',......}

jreback · 2017-03-27T15:08:24Z

@carlosdanielcsantos lgtm. some more doc-string nitpicks.

jreback · 2017-03-27T15:08:46Z

ping when all green.

carlosdanielcsantos · 2017-03-27T15:17:34Z

@jreback I was going to start writing tests for the FixedWindowIndexer but there are no explicit rolling computing test going on in TestRolling (they are done in TestMoments, I think). Where should I add these new tests?

jreback · 2017-03-27T15:22:44Z

@jreback How should I add this to computations.rst? A new section inside Window Functions or just add the examples inside the existing Window Functions -> Rolling Windows?

exactly, a new sub-section that pretty much shows the example with all 4 closed. maybe right before Centering Windows?

jreback · 2017-03-27T15:24:15Z

@jreback I was going to start writing tests for the FixedWindowIndexer but there are no explicit rolling computing test going on in TestRolling (they are done in TestMoments, I think). Where should I add these new tests?

We don't have explicit tests for the cython routines, these are covered by the integration tests (top-level routines). Though certainly could add them if you want / can.

You could add a new class in test_windows that explicity exercises these.

carlosdanielcsantos · 2017-03-28T12:47:23Z

@jreback @chrisaycock
Thinking deeper about the fixed window case, it seems that, intuitively, the default behavior right now better resembles the closed='both' case. Notice that when you use rolling(3).apply(f), the result for index i will be the result of f applied to [i-2, i-1, i], not ]i-2, i-1, i] (ISO 31-11 math set notation). One could argue that this is equivalent to having closed='right' with f applied to ]i-3, i-2, i-1, i]. Indeed, but this probably adulterates the meaning of the window size.

jreback · 2017-03-28T12:52:31Z

@jreback @chrisaycock
Thinking deeper about the fixed window case, it seems that, intuitively, the default behavior right now better resembles the closed='both' case. Notice that when you use rolling(3).apply(f), the result for index i will be the result of f applied to [i-2, i-1, i], not ]i-2, i-1, i] (ISO 31-11 math set notation). One could argue that this is equivalent to having closed='right' with f applied to ]i-3, i-2, i-1, i]. Indeed, but this probably adulterates the meaning of the window size.

@carlosdanielcsantos can you show a comparison of these 2, to show more what you mean?

carlosdanielcsantos · 2017-03-28T13:21:08Z

@jreback

pd.DataFrame([1] * 5).rolling(3, min_periods=1).sum() 0 0 1.0 1 2.0 2 3.0 3 3.0 4 3.0

Look how these results come from the application of sum() to every element inside the valid window. This resembles closed='both'. closed=right would look like [1.0, 1.0, 2.0, 2.0, 2.0]. Maybe the confusion arises from the fact that in fixed windows we are using the size of the window and not an interval as in the variable window, so maybe the closed parameter doesn't quite make sense here.

Nonetheless, I think that the possibility of ignoring the current index in computations is valuable (that was my initial problem, after all). Some solutions for this are:

Use closed='both' as default in Fixed windows and allow closed = 'right', 'left', 'neither' as well.
Drop the closed argument in Fixed windows but allow an argument to ignore the current index (same behavior as closed='left' in variable windows). The other cases equivalent to closed='right', 'neither' in variable windows can be obtained by varying window size: for example: rolling(3, closed='right') is the same as rolling(2) and rolling(3, closed='neither') is the same as rolling(2, ignore_own_index=True)

jreback · 2017-03-28T17:00:05Z

Nonetheless, I think that the possibility of ignoring the current index in computations is valuable (that was my initial problem, after all). Some solutions for this are:

This a bit orthogonal though. We always want to actually use the index. Since using a fixed window size, it looks like we are ignoring the index :>.

So a possiblity is to actually check the index whether its equiv to a RangeIndex. e.g.

In [5]: df1 = pd.DataFrame([1] * 5) In [6]: df1.rolling(3, min_periods=1).sum() Out[6]: 0 0 1.0 1 2.0 2 3.0 3 3.0 4 3.0 In [8]: df2 = pd.DataFrame([1] * 5, index=[0, 1, 3, 4, 5]) In [9]: df2.rolling(3, min_periods=1).sum() Out[9]: 0 0 1.0 1 2.0 3 3.0 4 3.0 5 3.0

In [11]: (np.diff(df1.index)==1).all() Out[11]: True In [12]: (np.diff(df2.index)==1).all() Out[12]: False

Let's deal with this in a separate issue.

A way to disambiguate on this ATM is to allow closed=None to be the default parameter. Then it cannot be anything but that for a FixedWindow. For Variable, if we see closed=None -> closed='right' IOW the default is the same. And then allow the rest of the closed intervals.

I think we should fix this for Fixed intervals, but I agree this is a problem because we essentially have closed='both' as the default now (its implicit though) (as you stated above).

jreback · 2017-03-29T19:37:36Z

@carlosdanielcsantos if you can update would be great (docs and such), ignoring the issue about fixed windows for the moment.

jreback · 2017-04-03T15:13:58Z

@carlosdanielcsantos can you update

carlosdanielcsantos · 2017-04-03T16:18:18Z

@jreback I'll do that in the next hours. Should I just leave the fixed-window as it is?

jreback · 2017-04-03T16:29:19Z

I think make the closed=None the default, and document the possibilities (and default) for the variable sized window (and validate left,right,both,neither). For the fixed window document as well (you can say its closed='both', but ATM only allow None). and then open a new issue where we can deal with this.

We have to fix this, but this PR merely opens up the possibilities of change here. So I think ok to merge this.

jreback · 2017-04-09T14:01:12Z

@carlosdanielcsantos can you update

carlosdanielcsantos · 2017-04-10T00:30:39Z

@jreback made some corrections but circleci is not passing in test_append_index - pandas.tests.test_multilevel.TestMultiLevel, but passing locally. Any thoughts?

jreback · 2017-04-10T00:52:59Z

try pushing again (make a modification)
looks like a fluke

carlosdanielcsantos · 2017-04-11T08:27:10Z

@jreback Still not passing. Could I have broken this with the changes in window?

…ndexer

‘Closed’ feature is working as expected in time-based windows. Sum tests are passing.

Adding test of assert for closed parameter Adding assert for closed parameter in get_window_indexer

Also, adding whatsnew entry

Adding computation.rst section (still not written)

Not allowing closed parameter for fixed windows Passing tests again Updating docs

Style corrections

jreback · 2017-04-11T10:22:14Z

@carlosdanielcsantos I rebased you against current master. This should pass cleanly. If you'd make those other changes and ping.

carlosdanielcsantos · 2017-04-12T08:23:24Z

@jreback what changes? I believe I updated everything that you requested. Can you take a look at AppVeyor build that failed?

jreback

minor comments.

jreback · 2017-04-09T22:58:43Z

doc/source/computation.rst

+.. _stats.rolling_window.endpoints:
+
+Rolling window endpoint inclusion
+~~~~~~~~~~~~~~~~~~


this underline needs to be the same length as the text

make this Rolling Window Endpoints

jreback · 2017-04-12T10:36:04Z

doc/source/computation.rst

+The inclusion of the interval endpoints in rolling window calculations can be specified with the ``closed``
+parameter:
+
+ - ``right`` : close right endpoint (default for time-based windows)


can you make this a csv-table (3 columns: closed, description, default) or something like that

jreback · 2017-04-12T10:36:53Z

doc/source/computation.rst

+
+ df
+
+Currently, this feature is only implemented for time-based windows.


expand this a touch to say that only closed='both' is accepted for fixed windows.

I am only accepting closed=None for fixed windows, as previously requested by you

For the fixed window document as well (you can say its closed='both', but ATM only allow None)

that's what i mean just make sure the docs are clear on that ; e.g. None -> 'both' for fixed (and assert that as well )

jreback · 2017-04-12T10:37:38Z

doc/source/whatsnew/v0.20.0.txt

 Other Enhancements
 ^^^^^^^^^^^^^^^^^^

+- ``DataFrame.rolling()`` now accepts the parameter ``closed='right'|'left'|'both'|'neither'`` to choose the rolling window endpoint closedness. See :ref:`documentation <stats.rolling_window.endpoints>` (:issue:`13965`)


See :ref:`the documentation ....

jreback · 2017-04-12T10:38:40Z

pandas/core/window.py

 raise ValueError("min_periods must be an integer")
+ if self.closed is not None and self.closed not in \
+ ['right', 'both', 'left', 'neither']:
+ raise ValueError("closed must be 'right', 'left', 'both' or "


is there validation for a fixed window when closed is NOT both?

Again, there is a validation for when closed is not None in Rolling.validate()

carlosdanielcsantos · 2017-04-13T09:29:32Z

pandas/tests/test_window.py

 tm.assertRaisesRegexp(UnsupportedFunctionCall, msg,
 getattr(r, func), dtype=np.float64)

+ def test_closed(self):


@jreback here is the piece that tests that closed must be None for fixed-window. Well, actually it only tests that it should not be 'neither', but I do not think replication is necessary.

carlosdanielcsantos · 2017-04-13T09:31:35Z

pandas/core/window.py

 elif self.window < 0:
 raise ValueError("window must be non-negative")

+ if not self.is_datetimelike and self.closed is not None:


@jreback here we enforce that closed must be None for windows that are not datetimelike

jreback · 2017-04-13T11:39:51Z

thanks @carlosdanielcsantos nice PR. as you can see, the code changes are the 'easy' part, its the docs & testing that are the details.!

keem em coming!

carlosdanielcsantos commented Mar 23, 2017

View reviewed changes

jreback mentioned this pull request Mar 23, 2017

ENH: allow exact matches in time-based .rolling() #13968

Closed

4 tasks

jreback added API Design Reshaping Concat, Merge/Join, Stack/Unstack, Explode Datetime Datetime data dtype labels Mar 23, 2017

jreback requested changes Mar 23, 2017

View reviewed changes

jreback reviewed Mar 27, 2017

View reviewed changes

pandas/core/window.pyx

else:

end[i] = end[i - 1]

# right endpoint is open

Copy link

Contributor

jreback Mar 27, 2017

Choose a reason for hiding this comment

The reason will be displayed to describe this comment to others. Learn more.

yep excatly like this one

jreback reviewed Mar 27, 2017

View reviewed changes

jreback approved these changes Mar 27, 2017

View reviewed changes

jreback added this to the 0.20.0 milestone Mar 27, 2017

carlosdanielcsantos and others added 12 commits April 11, 2017 06:14

Adding window slicing endpoint inclusion selection to VariableWindowI…

34f1309

…ndexer

Commiting progress

da034bf

Time-based windows working

2cf6804

‘Closed’ feature is working as expected in time-based windows. Sum tests are passing.

str closed -> object closed

5eaf3b4

Adding test of assert for closed parameter Adding assert for closed parameter in get_window_indexer

Adding doc-strings and PEP8 corrections

0e8e65c

Also, adding whatsnew entry

Changing l_closed and r_closed variable names

ec4bbc7

Commiting progress on default=None. Still not tested

306b9f7

Adding computation.rst section (still not written)

Almost there

8bd336a

Not allowing closed parameter for fixed windows Passing tests again Updating docs

Correcting bug in window validation

90dfb0c

Fixing test of assertion of closed parameter in fixed windows

c18a31b

Style corrections

Fixing style

037b84e

Innocuous change to rerun tests

568c12f

jreback force-pushed the rwindow-endpoints-inclusion branch from 6ca4949 to 568c12f Compare April 11, 2017 10:19

jreback approved these changes Apr 12, 2017

View reviewed changes

Updating docs

aad97dc

carlosdanielcsantos commented Apr 13, 2017

View reviewed changes

jreback closed this in 7322239 Apr 13, 2017


		df

		Currently, this feature is only implemented for time-based windows.

Uh oh!

Rolling window endpoints inclusion #15795

Rolling window endpoints inclusion #15795

Uh oh!

Conversation

carlosdanielcsantos commented Mar 23, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Choose a reason for hiding this comment

Choose a reason for hiding this comment

carlosdanielcsantos commented Mar 23, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

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

Choose a reason for hiding this comment

jreback commented Mar 23, 2017

jreback commented Mar 23, 2017

codecov bot commented Mar 27, 2017

Codecov Report

codecov bot commented Mar 27, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

Codecov Report

carlosdanielcsantos commented Mar 27, 2017

Choose a reason for hiding this comment

carlosdanielcsantos Mar 27, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

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

jreback commented Mar 27, 2017

jreback commented Mar 27, 2017

carlosdanielcsantos commented Mar 27, 2017

jreback commented Mar 27, 2017

jreback commented Mar 27, 2017

carlosdanielcsantos commented Mar 28, 2017

jreback commented Mar 28, 2017

carlosdanielcsantos commented Mar 28, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

jreback commented Mar 28, 2017

jreback commented Mar 29, 2017

jreback commented Apr 3, 2017

carlosdanielcsantos commented Apr 3, 2017 • edited Loading Uh oh! There was an error while loading. Please reload this page.

Uh oh!

jreback commented Apr 3, 2017

jreback commented Apr 9, 2017

carlosdanielcsantos commented Apr 10, 2017

jreback commented Apr 10, 2017

carlosdanielcsantos commented Apr 11, 2017

jreback commented Apr 11, 2017

carlosdanielcsantos commented Apr 12, 2017

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

carlosdanielcsantos commented Mar 23, 2017 •

edited

Loading

carlosdanielcsantos commented Mar 23, 2017 •

edited

Loading

codecov bot commented Mar 27, 2017 •

edited

Loading

carlosdanielcsantos Mar 27, 2017 •

edited

Loading

carlosdanielcsantos commented Mar 28, 2017 •

edited

Loading

carlosdanielcsantos commented Apr 3, 2017 •

edited

Loading