Converting numpy to R array using rpy2 low level interface

Question

There are a lot of related answers on SO, but all of them use the high-level robjects interface provided by rpy2. But, if I wanted to use the low-level interface, how do I go about it? Following is a MCVE (provided you have the package copula installed in R):

Alternative 1, using just the low-level interface as in the docs

import numpy as np
from rpy2 import rinterface as ri

ri.initr()

def rimport(packname):
    as_environment = ri.baseenv['as.environment']
    require = ri.baseenv['require']
    require(ri.StrSexpVector([packname]),
            quiet = ri.BoolSexpVector((True, )))
    packname = ri.StrSexpVector(['package:' + str(packname)])
    pack_env = as_environment(packname)
    return pack_env

Copula = rimport('copula')

# The ri.SexpVector line causes the problem, but this is how the docs has it.
gc = Copula['gofCopula'](copula=Copula['gumbelCopula'](dim=5),
                         x=ri.SexpVector(np.random.randn(100,5), ri.REALSXP),
                         N=ri.IntSexpVector((1000,)),
                         simulation=ri.StrSexpVector(('mult',)))

Dies with:

RRuntimeError                             Traceback (most recent call last)
<ipython-input-3-63487c32d528> in <module>()
      2                          x=ri.SexpVector(np.random.randn(100,5), ri.REALSXP),
      3                          N=ri.IntSexpVector((1000,)),
----> 4                          simulation=ri.StrSexpVector(('mult',)))
      5 
      6 gc

RRuntimeError: Error: (d <- ncol(x)) > 1 is not TRUE

Alternative 2, using `numpy2ri`, as shown in all answers on SO

import numpy as np
from rpy2 import rinterface as ri
from rpy2.robjects import numpy2ri

ri.initr()
numpy2ri.activate()

def rimport(packname):
    as_environment = ri.baseenv['as.environment']
    require = ri.baseenv['require']
    require(ri.StrSexpVector([packname]),
            quiet = ri.BoolSexpVector((True, )))
    packname = ri.StrSexpVector(['package:' + str(packname)])
    pack_env = as_environment(packname)
    return pack_env

Copula = rimport('copula')

# Automatic conversion does not happen!
gc = Copula['gofCopula'](copula=Copula['gumbelCopula'](dim=5),
                         x=np.random.randn(100,5), # Hoping for automatic conversion
                         N=ri.IntSexpVector((1000,)),
                         simulation=ri.StrSexpVector(('mult',)))

Dies with:

ValueError                                Traceback (most recent call last)
<ipython-input-1-17b2b5105f01> in <module>()
     24                          x=np.random.randn(100,5),
     25                          N=ri.IntSexpVector((1000,)),
---> 26                          simulation=ri.StrSexpVector(('mult',)))
     27 
     28 gc

ValueError: All parameters must be of type Sexp_Type,or Python int/long, float, bool, or None

Alternative 3, using both together

Dies just like Alternative 1.

Additional Notes

1)

list(ri.SexpVector(np.random.randn(100,5), ri.REALSXP))

Is a list that contains NA_real_, exactly 100 of them.

2)

If I ditch the low-level interface, and just use the high-level interface, everything works fine. But that is not what I am looking for.

import numpy as np
from rpy2.robjects.packages import importr
from rpy2.robjects import numpy2ri

numpy2ri.activate()

Copula = importr('copula')
gc = Copula.gofCopula(copula=partial(Copula.gumbelCopula, dim=5)(),
                      x=np.random.randn(100,5),
                      N=1000,
                      simulation='mult')

lgautier · Accepted Answer · 2017-07-29 14:03:12Z

1

Thanks for your praise of the high-level interface and the work that went into it, although an implicit one: it is indeed designed to "just work".

The low-level interface is quite close to R's C-level API and its use may require more knowledge about it that would deem reasonable to expect from a Python audience. Note that the documentation published is unfortunately not complete because the docstrings are not included (issue with readthedocs).

I'd encourage you to use the high-level interface unless a specific reason not to, but since the source is open what is happening in the converter can be checked easily (it is here).

answered Jul 29, 2017 at 14:03

lgautier

11.6k31 silver badges43 bronze badges

Sign up to request clarification or add additional context in comments.

2 Comments

Kartik Over a year ago

Hello again, Igautier! I hope you've been doing great. In the discussion following your answer before, we found some limitations with the high-level interface and multiprocessing. I usually work with large datasets, and, unfortunately, a lot of great statistical packages have no equivalent in Python (copula being one of them). So I must use R. It takes about 14 hours to fit a copula model to one subset of my data, of which I have 28. So, parallel is the way to go. Hence, I am looking to use the low-level interface which works with multiprocessing

Kartik Over a year ago

The real question I have is, how to convert numpy array to R matrix, while preserving the shape of the array, using the low-level rinterface. Although the high-level api is amazing, and I really appreciate all the great work that has gone into making it, it does not solve my specific problem. Thank you for pointing out the exact location for numpy2ri in the source. That was very helpful to replicate in my scripts.

Collectives™ on Stack Overflow

Converting numpy to R array using rpy2 low level interface

Alternative 1, using just the low-level interface as in the docs

Alternative 2, using `numpy2ri`, as shown in all answers on SO

Alternative 3, using both together

Additional Notes

1 Answer 1

2 Comments

Your Answer

Linked

Hot Network Questions

Collectives™ on Stack Overflow

Alternative 1, using just the low-level interface as in the docs

Alternative 2, using numpy2ri, as shown in all answers on SO

Alternative 3, using both together

Additional Notes

1 Answer 1

2 Comments

Your Answer

Sign up or log in

Post as a guest

Linked

Related

Alternative 2, using `numpy2ri`, as shown in all answers on SO