RcppCNPy

NumPy is a popular package for scientific / numeric computing with Python. It uses an internal file to store vectors and matrices.

Sometimes we want to read (or write) these files with languages other than Python. The small CNPy library on GitHub permits this from C++, and by relying on Rcpp we can easily bring this to R.

Example

The following Python code

>>> import numpy as np
>>> fm = np.arange(12).reshape(3,4) * 1.1
>>> fm
array([[  0. ,   1.1,   2.2,   3.3],
       [  4.4,   5.5,   6.6,   7.7],
       [  8.8,   9.9,  11. ,  12.1]])
>>> np.save("fmat.npy", fm)
>>>
>>> im = np.arange(12).reshape(3,4)
>>> im
array([[ 0,  1,  2,  3],
       [ 4,  5,  6,  7],
       [ 8,  9, 10, 11]])
>>> np.save("imat.npy", im)
>>> 

saves two matrices in floating-point and integer representation.

With this R code we can read and assign the files:

R> library(RcppCNPy)
R> fmat <- npyLoad("fmat.npy")
R> fmat
     [,1] [,2] [,3] [,4]
[1,]  0.0  1.1  2.2  3.3
[2,]  4.4  5.5  6.6  7.7
[3,]  8.8  9.9 11.0 12.1
R> 
R> imat <- npyLoad("imat.npy", "integer")
R> imat
     [,1] [,2] [,3] [,4]
[1,]    0    1    2    3
[2,]    4    5    6    7
[3,]    8    9   10   11
R> 

Going the opposite way by saving in R and reading in Python works equally well. An extension not present in CNPy allows reading and writing of gzip-compressed files.

The package has been tested and used on several architecture, and copes correctly with little-vs-big endian switches.

More details are available in the package vignette.

Where

The package is hosted in this GitHub repository and available via CRAN as well.

License

GPL (>= 2)

Status

The package is fairly mature.

Author

Dirk Eddelbuettel for this package.
Carl Rogers for CNPy.

Initially created: Tue Jan 6 21:24:50 CST 2015
Last modified: Thu Sep 22 05:20:21 CDT 2016