ENH/VIS: Area plot is now supported by kind='area'. #6656

sinhrks · 2014-03-17T13:38:17Z

Area plot is added to plotting method. The AreaPlot class is created as a subclass of LinePlot, thus it works also in time series.

By default, area plot is being stacked. When area plot is not stacked (stacked=False), alpha value is set to 0.5 to show overlapped area if not configured specifically. As a side benefit, line plot also can be stacked by specifying stacked=True (disabled by default). Different from stacked bar plot, I don't know good visualization for positive/negative mixed data. Thus, input must be all positive or all negative when stacked=True. I'll try to implement it if there is a good way. Also, area plot doesn't support logy and loglog plot because filling area starts from 0.

Note: Area plot's legend is implemented based on the answer described in:
http://stackoverflow.com/questions/14534130/legend-not-showing-up-in-matplotlib-stacked-area-plot

Example:

TomAugspurger · 2014-03-23T14:35:26Z

pandas/tools/plotting.py

+
+    def _get_plot_function(self):
+        if self.logy or self.loglog:
+            raise ValueError("Log-scales are not supported in area plot")


Log-x scales are supported right? You could change this to "Log-y scales are not supported in area plot".

When I try a logx=True I get a segfault actually (on the matplotlib side).

Modified the error message. logx=True should work, and I've added a test case.

TomAugspurger · 2014-03-23T14:46:49Z

You can put an example in doc/source/visualization.rst if you want to show off your work.

One thing case that your implementation doesn't allow is stacked with mixed positive and negatives, where every value in a column are either all positive or all negative. I think this would make sense visually. We can leave that for a future issue if you want.

sinhrks · 2014-03-29T09:50:35Z

Thanks. I've added examples in visualization.rst, and answered above comments.

jorisvandenbossche · 2014-03-29T10:41:34Z

doc/source/visualization.rst

+Area plot
+~~~~~~~~~~~~~~~~~~~
+
+*New in .14* You can create area plots with ``Series.plot`` and ``DataFrame.plot`` by passing ``kind='area'``. Area plots are stacked by default. To produce stacked area plot, all the dataframe values are either positive or negative.


Can you change the New in .14 to .. versionadded:: 0.14 on a seperate line?

sinhrks · 2014-03-29T11:18:37Z

Sure. Done it.

There are other version descriptions which uses italic in 455, 468, 487th lines. Shall I modify it as the same manner?

jorisvandenbossche · 2014-03-29T11:44:54Z

doc/source/visualization.rst

+~~~~~~~~~~~~~~~~~~~
+
+.. versionadded:: 0.14
+You can create area plots with ``Series.plot`` and ``DataFrame.plot`` by passing ``kind='area'``. Area plots are stacked by default. To produce stacked area plot, all the dataframe values are either positive or negative.


Small detail, there also has to be a blank line here between versionadded and the necxt paragraph

Thank you, fixed.

sinhrks · 2014-03-29T12:25:38Z

Failed in test_frame_groupby_plot_boxplot which is not modified. Maybe the same issue as #6670? I'll hook Travis again.

jreback · 2014-04-05T23:55:56Z

@sinhrks can you rebase

@TomAugspurger @jorisvandenbossche ok with this?

sinhrks · 2014-04-06T02:11:59Z

Rebased. Could you check #6678 first, because I'll apply the same fix to area plot.

TomAugspurger · 2014-04-22T00:17:36Z

pandas/tools/plotting.py

@@ -1933,6 +2036,8 @@ def plot_frame(frame=None, x=None, y=None, subplots=False, sharex=True,
        klass = ScatterPlot
    elif kind == 'hexbin':
        klass = HexBinPlot
+    elif kind == 'area':


Could you add 'area' to the set of valid values for kind in the docstring for plot_frame and plot_series?

TomAugspurger · 2014-04-22T00:58:32Z

The way you've done it makes gives some strange results if someone does multiple area plots on the same axis (I can't think of a reason why anyone would want to do this, but who knows).

In [64]: df
Out[64]: 
     0   1   2
0    0   1   3
1    1   2   7
2    1   6   8
3    5  10   8
4    5  11   8
5    9  15  11
6    9  19  15
7   12  20  17
8   15  20  19
9   16  25  20
10  17  27  21
11  21  30  25
12  21  30  28
13  25  31  30
14  25  32  34
15  28  34  35
16  29  38  38
17  30  38  42
18  34  38  44
19  38  38  48

[20 rows x 3 columns]

In [65]: df2 = df + 10

In [66]: ax = df.plot(kind='area')

In [67]: df2.plot(kind='area', ax=ax)
Out[67]: <matplotlib.axes.AxesSubplot at 0x10f603250>

The original shading is removed:

Honestly, I think I'm OK with this. Unless someone can think of any reason where multiple, independently stacked datasets on the same axes makes sense.

TomAugspurger · 2014-04-22T01:00:31Z

doc/source/release.rst

@@ -150,6 +150,7 @@ API Changes

 - ``DataFrame.sort`` now places NaNs at the beginning or end of the sort according to the ``na_position`` parameter. (:issue:`3917`)

+<<<<<<< HEAD


Merge errer here. You can remove while rebasing.

TomAugspurger · 2014-04-22T01:10:38Z

That's all my comments. The biggest one is deciding how to handle NaNs. Thoughts? line leaves gaps (which doesn't really make sense for area, does it?), scatter and hexbin excludes those rows with NaNs (which is an option), bar essentially draws a bar of length 0 (which isn't an option)

sinhrks · 2014-04-26T05:34:15Z

Thanks for all your comments.

Multiple area plots on single axes:

Actually it is not deleted. When staked=True, alpha value 1.0 is used because it was considered each area will not be overlapped. This makes previous area hidden when plotting new area on the same axes. One option is always use alpha=0.5 in area plot unless otherwise specified.

If you pass alpha explicitly, you can see all the area are drawn.

import numpy as np
import matplotlib.pyplot as plt
import pandas.util.testing as tm
t_df = DataFrame(np.random.rand(6, 3),
               index=tm.makeDateIndex(k=6), columns=['a', 'b', 'c'])
t_df2 = DataFrame(np.random.rand(6, 3),
               index=tm.makeDateIndex(k=6), columns=['x', 'y', 'z'])

fig, axes = plt.subplots(1, 1, figsize=(14, 8))
plt.subplots_adjust(top=0.9, bottom=0.2, left=0.15, right=0.9, hspace=0.5)
t_df.plot(kind='area', ax=axes, alpha=0.5)
t_df2.plot(kind='area', ax=axes, alpha=0.5)

Handling NaN:

It looks MPL stackplot doesn't care NaN (right figure). I think automatically filling by 0 is preferable (left figure). How about the following spec?:

Line plot without stacking: skip NaN and split lines (no change).
Line plot with stacking: fill NaN with 0 (for stacking purpose)
Area plot: fill NaN with 0 (for stacking/filling purpose)

import matplotlib.pyplot as plt
fig, axes = plt.subplots(1, 2)

x = range(6)
y11 = [1, 1, 0, 1, 1, 1]
y12 = [2, 2, 2, 0, 2, 2]
axes[0].stackplot(x, y11, y12)

y21 = [1, 1, np.nan, 1, 1, 1]
y22 = [2, 2, 2, np.nan, 2, 2]
axes[1].stackplot(x, y21, y22)

sinhrks · 2014-04-26T05:51:21Z

Also, I noticed that initial description was incorrect. Each line/area can be stacked on positive/negative direction separately if each columns are all either positive or negative. No need to all the data is either positive or negative. I've changed the logic, and will add some tests/docs for this.

Different from stacked bar plot, I don't know good visualization for positive/negative mixed data. Thus, input must be all positive or all negative when stacked=True.

import numpy as np
import pandas.util.testing as tm

df = pd.DataFrame({'a': np.random.rand(6),
                'b': np.random.rand(6),
                'c': - np.random.rand(6),
                'd': - np.random.rand(6)},
               index=tm.makeDateIndex(k=6))
df.plot(kind='area', stacked=True)

sinhrks · 2014-04-26T15:58:33Z

I've added tests for fixed NaN handling and stacking, and modified doc. I think remainings are:

What the default alpha should be.
Whether NaN handling described above looks OK.

TomAugspurger · 2014-04-28T13:55:56Z

doc/source/visualization.rst

+
+.. versionadded:: 0.14
+
+You can create area plots with ``Series.plot`` and ``DataFrame.plot`` by passing ``kind='area'``. Area plots are stacked by default. To produce stacked area plot, each columns must be either all positive or negative values.


typo: columns -> column. Also maybe change "must be either" to "must have either" and "all positive or negative" to "all positive or all negative"

TomAugspurger · 2014-04-28T14:03:26Z

I think a default alpha of 1 when stacked and .5 when not stacked is good.

Your NaN handling looks good too, especially since it's consistent with how matplotlib does it.

Good catch on the all positive / all negative checks by column.

I'm reviewing now. Should be able to merge this today.

TomAugspurger · 2014-04-28T14:10:29Z

doc/source/visualization.rst

+.. ipython:: python
+
+   @savefig area_plot_unstacked.png
+   df.plot(kind='area', stacked=False);


Could you make a note here that NaNs are filled to zero by default, and that if you don't want that you should fill or drop the NaNs before plotting. (No need for an example with nans)

TomAugspurger · 2014-04-28T16:28:23Z

pandas/tests/test_graphics.py

@@ -1256,6 +1406,14 @@ def test_partially_invalid_plot_data(self):
                with tm.assertRaises(TypeError):
                    df.plot(kind=kind)

+            # area plot doesn't support positive/negative mixed data
+            kinds = ['area']
+            df = DataFrame(rand(10, 2), dtype=object)


Same thing here with using a seed known to produce a mixed frame.

TomAugspurger · 2014-04-28T16:29:07Z

@sinhrks ping me when you rebase.

…rea plot can be stacked by stacked=True.

jreback · 2014-05-01T14:14:21Z

@sinhrks @TomAugspurger

TomAugspurger · 2014-05-01T14:16:57Z

Looks like there was a Yahoo failure, I restarted the build.

jreback · 2014-05-01T15:14:39Z

this is fine, I already fixed master

ENH/VIS: Area plot is now supported by kind='area'.

sinhrks mentioned this pull request Mar 23, 2014

DOC: Mention date and time formatting available throughout MPL #6660

Merged

TomAugspurger reviewed Mar 23, 2014
View reviewed changes

jorisvandenbossche reviewed Mar 29, 2014
View reviewed changes

jreback added Bug labels Apr 6, 2014

jreback added this to the 0.14.0 milestone Apr 6, 2014

TomAugspurger reviewed Apr 22, 2014
View reviewed changes

TomAugspurger reviewed Apr 28, 2014
View reviewed changes

TomAugspurger mentioned this pull request Apr 28, 2014

BUG: enabling subplots works unexpectedly #6951

Closed

ENH: Area plot is now supported by kind='area'. Also, line plot and a…

b63a699

…rea plot can be stacked by stacked=True.

jreback added a commit that referenced this pull request May 1, 2014

Merge pull request #6656 from sinhrks/area_pr2

614e273

ENH/VIS: Area plot is now supported by kind='area'.

jreback merged commit 614e273 into pandas-dev:master May 1, 2014

sinhrks deleted the area_pr2 branch May 2, 2014 11:57

Provide feedback

Saved searches

Use saved searches to filter your results more quickly

ENH/VIS: Area plot is now supported by kind='area'. #6656

ENH/VIS: Area plot is now supported by kind='area'. #6656

sinhrks commented Mar 17, 2014

TomAugspurger Mar 23, 2014

sinhrks Mar 29, 2014

TomAugspurger commented Mar 23, 2014

sinhrks commented Mar 29, 2014

jorisvandenbossche Mar 29, 2014

sinhrks commented Mar 29, 2014

jorisvandenbossche Mar 29, 2014

sinhrks Mar 29, 2014

sinhrks commented Mar 29, 2014

jreback commented Apr 5, 2014

sinhrks commented Apr 6, 2014

TomAugspurger Apr 22, 2014

TomAugspurger commented Apr 22, 2014

TomAugspurger Apr 22, 2014

TomAugspurger commented Apr 22, 2014

sinhrks commented Apr 26, 2014

sinhrks commented Apr 26, 2014

sinhrks commented Apr 26, 2014

TomAugspurger Apr 28, 2014

TomAugspurger commented Apr 28, 2014

TomAugspurger Apr 28, 2014

TomAugspurger Apr 28, 2014

TomAugspurger commented Apr 28, 2014

jreback commented May 1, 2014

TomAugspurger commented May 1, 2014

jreback commented May 1, 2014

		@@ -150,6 +150,7 @@ API Changes

		- ``DataFrame.sort`` now places NaNs at the beginning or end of the sort according to the ``na_position`` parameter. (:issue:`3917`)

		<<<<<<< HEAD


		.. versionadded:: 0.14

		You can create area plots with ``Series.plot`` and ``DataFrame.plot`` by passing ``kind='area'``. Area plots are stacked by default. To produce stacked area plot, each columns must be either all positive or negative values.

ENH/VIS: Area plot is now supported by kind='area'. #6656

ENH/VIS: Area plot is now supported by kind='area'. #6656

Conversation

sinhrks commented Mar 17, 2014

Choose a reason for hiding this comment

Choose a reason for hiding this comment

TomAugspurger commented Mar 23, 2014

sinhrks commented Mar 29, 2014

Choose a reason for hiding this comment

sinhrks commented Mar 29, 2014

Choose a reason for hiding this comment

Choose a reason for hiding this comment

sinhrks commented Mar 29, 2014

jreback commented Apr 5, 2014

sinhrks commented Apr 6, 2014

Choose a reason for hiding this comment

TomAugspurger commented Apr 22, 2014

Choose a reason for hiding this comment

TomAugspurger commented Apr 22, 2014

sinhrks commented Apr 26, 2014

Multiple area plots on single axes:

Handling NaN:

sinhrks commented Apr 26, 2014

sinhrks commented Apr 26, 2014

Choose a reason for hiding this comment

TomAugspurger commented Apr 28, 2014

Choose a reason for hiding this comment

Choose a reason for hiding this comment

TomAugspurger commented Apr 28, 2014

jreback commented May 1, 2014

TomAugspurger commented May 1, 2014

jreback commented May 1, 2014