ggplot2 hell with rpy2-2.0.7 + python 2.6 + r 2.11 (windows 7)

Question

i am using rpy2-2.0.7 (i need this to work with windows 7, and compiling the binaries for the newer rpy2 versions is a mess) to push a two-column dataframe into r, create a few layers in ggplot2, and output the image into a <.png>.

i have wasted countless hours fidgeting around with the syntax; i did manage to output the files i needed at one point, but (stupidly) did not notice and continued fidgeting around with my code ...

i would sincerely appreciate any help; below is a (trivial) example for demonstration. Thank you very much for your help!!! ~ Eric Butter

---

import rpy2.robjects as rob
from rpy2.robjects import r
import rpy2.rlike.container as rlc
from array import array

r.library("grDevices")    # import r graphics package with rpy2
r.library("lattice")
r.library("ggplot2")
r.library("reshape")

picpath = 'foo.png' 

d1 = ["cat","dog","mouse"]
d2 = array('f',[1.0,2.0,3.0])

nums = rob.RVector(d2)
name = rob.StrVector(d1)

tl = rlc.TaggedList([nums, name], tags = ('nums', 'name'))
dataf = rob.RDataFrame(tl)

## r['png'](file=picpath, width=300, height=300)
## r['ggplot'](data=dataf)+r['aes_string'](x='nums')+r['geom_bar'](fill='name')+r['stat_bin'](binwidth=0.1)
r['ggplot'](data=dataf)
r['aes_string'](x='nums')
r['geom_bar'](fill='name')
r['stat_bin'](binwidth=0.1)
r['ggsave']()
## r['dev.off']()

*The output is just a blank image (181 b).

---

here are a couple common errors R itself throws as I fiddle around in ggplot2:

r['png'](file=picpath, width=300, height=300)
r['ggplot']()
r['layer'](dataf, x=nums, fill=name, geom="bar")
r['geom_histogram']()
r['stat_bin'](binwidth=0.1)
r['ggsave'](file=picpath)
r['dev.off']()

*RRuntimeError: Error: No layers in plot

r['png'](file=picpath, width=300, height=300)
r['ggplot'](data=dataf)
r['aes'](geom="bar")
r['geom_bar'](x=nums, fill=name)
r['stat_bin'](binwidth=0.1)
r['ggsave'](file=picpath)
r['dev.off']()

*RRuntimeError: Error: When setting aesthetics, they may only take one value. Problems: fill,x

Answer 1

I use rpy2 solely through Nathaniel Smith's brilliant little module called rnumpy (see the "API" link at the rnumpy home page). With this you can do:

from rnumpy import *

r.library("ggplot2")

picpath = 'foo.png' 
name = ["cat","dog","mouse"]
nums = [1.0,2.0,3.0]

r["dataf"] = r.data_frame(name=name, nums=nums)
r("p <- ggplot(dataf, aes(name, nums, fill=name)) + geom_bar(stat='identity')")
r.ggsave(picpath)

(I'm guessing a little about how you want the plot to look, but you get the idea.)

Another great convenience is entering "R mode" from Python with the ipy_rnumpy module. (See the "IPython integration" link at the rnumpy home page).

For complicated stuff, I usually prototype in R until I have the plotting commands worked out. Error reporting in rpy2 or rnumpy can get quite messy.

For instance, the result of an assignment (or other computation) is sometimes printed even when it should be invisible. This is annoying e.g. when assigning to large data frames. A quick workaround is to end the offending line with a trailing statement that evaluates to something short. For instance:

In [59] R> long <- 1:20
Out[59] R>
  [1]   1   2   3   4   5   6   7   8   9  10  11  12  13  14  15  16  17  18
 [19]  19  20

In [60] R> long <- 1:100; 0
Out[60] R> [1] 0

(To silence some recurrent warnings in rnumpy, I've edited rnumpy.py to add 'from warnings import warn' and replace 'print "error in process_revents: ignored"' with 'warn("error in process_revents: ignored")'. That way, I only see the warning once per session.)

Answer 2

You have to engage the dev() before you shut it off, which means that you have to print() (like JD guesses above) prior to throwing dev.off().

from rpy2 import robjects                          
r = robjects.r                                                                                    
r.library("ggplot2")
robjects.r('p = ggplot(diamonds, aes(clarity, fill=cut)) + geom_bar()') 
r.ggsave('/stackBar.jpeg') 
robjects.r('print(p)')
r['dev.off']()

Answer 3

To make it slightly more easy when you have to draw more complex plots:

from rpy2 import robjects
from rpy2.robjects.packages import importr
import rpy2.robjects.lib.ggplot2 as ggplot2
r = robjects.r
grdevices = importr('grDevices')
p = r('''
  library(ggplot2)

  p <- ggplot(diamonds, aes(clarity, fill=cut)) + geom_bar()
  p <- p + opts(title = "{0}")
    # add more R code if necessary e.g. p <- p + layer(..)
  p'''.format("stackbar")) 
  # you can use format to transfer variables into R
  # use var.r_repr() in case it involves a robject like a vector or data.frame
p.plot()
# grdevices.dev_off()