How to plot two different graphs on same axes

[Richard Stanton]

I have two different dataframes that I'd like to plot simultaneously (they share an x-axis).

df1 I want to plot as a stacked bar chart, with labels for the different lines
df2 I want to plot as a line chart on the same axes, but using a different y-scale.

I can do either of these alone, but haven't been successful at getting them to plot simultaneously. Does anyone know how to do this?

Thanks a lot.

Richard Stanton

[Chang She]

--

What do you get you when you try plotting them simultaneously? You're most likely running into this bug: https://github.com/pydata/pandas/issues/1800

I'll put in a fix for it on master this week. Please stay tuned.

-Chang

[Richard Stanton]

>
>
>Richard Stanton <sta...@haas.berkeley.edu>
> Oct 09 02:05PM -0700

>
>
>
>
>I have two different dataframes that I'd like to plot simultaneously
>(they share an x-axis).
>
>
>
>
>
>df1 I want to plot as a stacked bar chart, with labels for the different
>lines
>
>
>df2 I want to plot as a line chart on the same axes, but using a
>different y-scale.
>
>
>
>
>
>
>I can do either of these alone, but haven't been successful at getting
>them to plot simultaneously. Does anyone know how to do this?
>
>
>
>
>Thanks a lot.
>
>
>
>
>
>Richard Stanton
>
>
>
>
>

>Chang She <ch...@lambdafoundry.com>
> Oct 09 05:08PM -0400

>
>
>
>
>
>
>> Thanks a lot.
>
>
>
>
>
>> Richard Stanton
>
>
>
>
>
>> --
>
>
>
>
>
>What do you get you when you try plotting them simultaneously? You're
>most likely running into this bug:
>
>https://github.com/pydata/pandas/issues/1800
>

It certainly sounds relevant, but I don't necessarily get an error
message. For example, when I try

df1 = DataFrame({'Series1' : range(9)}, index = range(2000,2009))
df2 = DataFrame({'Series2' : 5}, index = range(2000,2009))
df1.plot(kind='bar')
df2.plot(secondary_y=True)
draw()


I get two plots, one after the other. The second (line) plot has an x
range of 0 - 2500, so the entire graph is really compressed!

If I make the index range(9), now the axes look better (still two separate
graphs, though), but even if they were on the same axes they wouldn't
quite work as the spacing of the x-axis labels is different. On the bar
chart, there's a gap between the left/right hand side of the box and the
leftmost/rightmost tick. On the line graph, there's no such gap.

Richard

[Richard Stanton]

>
>Wouter Overmeire <lod...@gmail.com>
> Oct 09 08:41PM +0200
>This will give you 145 different colors.
>
>
>
>
>
>In [45]: from matplotlib.colors import cnames
>In [46]: df.plot(style=cnames.keys())

Thanks, Wouter.
>
>

[Wouter Overmeire]

2012/10/10 Richard Stanton <sta...@haas.berkeley.edu>

Sometimes i combine a boxplot and lineplot which has by default the same issue. Boxplots are plotted at positions 1, 2, 3, ...

matplotlib boxplot offers the argument positions, if used the boxes are created at the desired positions (i use the frame index values) instead of default 1, 2, 3, ...  When adding the lineplot, it nicely overlay`s with the boxes.

matplotlib bar does not offer the positions option :-( 

Anyway this seems an interesting use case for pandas to fix.

Note that if you want to add a horizontal line you can do this with axhline().

In [57]: ax = df1.plot(kind='bar')

In [58]: ax.axhline(5)

Out[58]: <matplotlib.lines.Line2D at 0xbd2e22c>

In [59]: draw()

[Alvaro Tejero Cantero]

Hi,

FWIW I just tried style=cnames.keys() with df.plot(kind='barh',
stacked=True); cycling colors does not work there

-á.

> --
>
>

[Wouter Overmeire]

2012/10/19 Alvaro Tejero Cantero <alv...@minin.es>

--

Can you show the exact command you used?

[Alvaro Tejero Cantero]

from matplotlib.colors import cnames
blocks.apply(equip_units, axis=1).fillna(0).plot(kind='barh',
stacked=True, style=cnames.keys()).set_xlabel('number'); plt.show()


-á.

> --
>
>

[Chang She]

If you do `color=cnames.keys()` it works for barh but `style=cnames.keys()` for line.

I made a bug report on github, we'll make it work for both parameters: https://github.com/pydata/pandas/issues/2082

--

[Alvaro Tejero Cantero]

Fantastic, thank you Chang!

color did the trick. As a note, since there's so much of automatic
plotting happening in Pandas, one might want to predefine subsets of
cnames (discrete/categorical colormaps?) that are a bit more
orthogonal. I tried a few 6-element slices of cnames and always found
two nearly identical colors among the 6.


-á.

> --
>
>