PythonInR

More documenation can be found at https://bitbucket.org/Floooo/pythoninr.

Dependencies

Python >= 2.7.0 R >= 2.15.0

R-packages:
- pack

Linux

Python headers

On Debian and Debian-based Linux distributions (including Ubuntu and other derivatives) the “Python Development Headers” can be installed by typing the following into the terminal.

    apt-get install python-dev

For installation on Red Hat Enterprise Linux , Fedora, and other Red Hat Linux-based distributions, use the following:

    yum install python-devel

Windows

There are no additional dependencies on Windows. (One obviously needs to have R and Python installed.)

Installation

    install.packages("PythonInR")
#   or via devtools
    require(devtools)
    install_bitbucket("Floooo/PythonInR")

Windows Setup

Since the Windows version of PythonInR uses explicit linkage one can switch between different Python versions without recompiling the package. This flexibility comes at the price of additional configuration at the startup. Which results in a different behavior for the static (Linux default) and the explicit linked (Windows default) version. Where as the static linked version automatically connects, when the package get’s loaded, the explicitly linked version needs to be connected manually.

To enable automatic connection for the explicitly linked version the environment variable PYTHON_EXE has to be set. You can put your Python path into your .Renviron or .Rprofile file (Setting up a .Renviron file).

NOTES

Python 3

Due to api changes in Python 3 the function execfile is no longer available. The PythonInR package provides a execfile function following the typical workaround.

def execfile(filename):
    exec(compile(open(filename, 'rb').read(), filename, 'exec'), globals())

Type Casting

R to Python (pySet)

To allow a nearly one to one conversion from R to Python, PythonInR provides Python classes for vectors, matrices and data.frames which allow an easy conversion from R to Python and back. The names of the classes are PrVector, PrMatrix and PrDataFrame.

Default Conversion

Change the predefined conversion of pySet

PythonInR is designed in way that the conversion of types can easily be added or changed. This is done by utilizing polymorphism: if pySet is called, pySet calls pySetPoly which can be easily modified by the user. The following example shows how pySetPoly can be used to modify the behavior of pySet on the example of integer vectors.

The predefined type casting for integer vectors at an R level looks like the following:

setMethod("pySetPoly", signature(key="character", value = "integer"),
          function(key, value){
    success <- pySetSimple(key, list(vector=unname(value), names=names(value), rClass=class(value)))
    cmd <- sprintf("%s = PythonInR.prVector(%s['vector'], %s['names'], %s['rClass'])", 
                   key, key, key, key)
    pyExec(cmd)
})

To change the predefined behavior one can simply use setMethod again.

pySetPoly <- PythonInR:::pySetPoly
showMethods("pySetPoly")

pySet("x", 1:3)
pyPrint(x)
pyType("x")

setMethod("pySetPoly",
          signature(key="character", value = "integer"),
          function(key, value){
    PythonInR:::pySetSimple(key, value)
})

pySet("x", 1:3)
pyPrint(x)
pyType("x")

NOTE PythonInR:::pySetSimple
The functions pySetSimple and pySetPoly shouldn’t be used outside the function pySet since they do not check if R is connected to Python. If R is not connected to Python this can yield to segfault !

NOTE (named lists):
When executing pySet("x", list(b=3, a=2)) and pyGet("x") the order of the elements in x will change. This is not a special behavior of PythonInR but the default behavior of Python for dictionaries.

NOTE (matrix):
Matrices are either transformed to an object of the class PrMatrix or to an numpy array (if the option useNumpy is set to TRUE).

NOTE (data.frame):
Data frames are either transformed to an object of the class PrDataFrame
or to a pandas DataFrame (if the option usePandas is set to TRUE).

R to Python (pyGet)

Change the predefined conversion of pyGet

Similar to pySet the behavior of pyGet can be changed by utilizing pyGetPoly. The predefined version of pyGetPoly for an object of class PrMatrix looks like the following:

setMethod("pyGetPoly", signature(key="character", autoTypecast = "logical", simplify = "logical", pyClass = "PrMatrix"),
          function(key, autoTypecast, simplify, pyClass){
    x <- pyExecg(sprintf("x = %s.toDict()", key), autoTypecast = autoTypecast, simplify = simplify)[['x']]
    M <- do.call(rbind, x[['matrix']])
    rownames(M) <- x[['rownames']]
    colnames(M) <- x[['colnames']]
    return(M)
})

For objects of type “type” no conversion is defined. Therefore, PythonInR doesn’t know how to transform it into an R object so it will return a PythonInR_Object. This is kind of a nice example since the return value of type(x) is a function therefore PythonInR will return an object of type pyFunction.

pyGet("type(list())")

One can define a new function to get elements of type “type” as follows.

pyGetPoly <- PythonInR:::pyGetPoly
setClass("type")
setMethod("pyGetPoly", signature(key="character", autoTypecast = "logical", simplify = "logical", pyClass = "type"),
          function(key, autoTypecast, simplify, pyClass){
    pyExecg(sprintf("x = %s.__name__", key))[['x']]
})
pyGet("type(list())")

NOTE pyGetPoly
The functions pyGetPoly should not be used outside the function pyGet since it does not check if R is connected to Python. If R is not connected to Python this will yield to segfault !

NOTE (bytes):
In short, in Python 3 the data type string was replaced by the data type bytes.

Cheat Sheet

Usage Examples

Dynamic Documents

• PythonInR and KnitR Example

Data and Text Mining

• PythonInR and Glove Example
  GloVe is an unsupervised learning algorithm for obtaining vector representations for words. More information can be found here.
  Jeffrey Pennington, Richard Socher, and Christopher D. Manning. “Glove: Global vectors for word representation.” Proceedings of the Empiricial Methods in Natural Language Processing (EMNLP 2014) 12 (2014): 1532-1543.

• PythonInR and TensorFlow Example
  TensorFlow is an open source software library for numerical computation using data flow graphs. More information can be found here.
  Martín Abadi, Ashish Agarwal, Paul Barham, Eugene Brevdo, Zhifeng Chen, Craig Citro, Greg S. Corrado, Andy Davis, Jeffrey Dean, Matthieu Devin, Sanjay Ghemawat, Ian Goodfellow, Andrew Harp, Geoffrey Irving, Michael Isard, Rafal Jozefowicz, Yangqing Jia, Lukasz Kaiser, Manjunath Kudlur, Josh Levenberg, Dan Mané, Mike Schuster, Rajat Monga, Sherry Moore, Derek Murray, Chris Olah, Jonathon Shlens, Benoit Steiner, Ilya Sutskever, Kunal Talwar, Paul Tucker, Vincent Vanhoucke, Vijay Vasudevan, Fernanda Viégas, Oriol Vinyals, Pete Warden, Martin Wattenberg, Martin Wicke, Yuan Yu, and Xiaoqiang Zheng. “TensorFlow: Large-scale machine learning on heterogeneous systems.” (2015).