In eddelbuettel/rcppcnpy: Read-Write Support for 'NumPy' Files via 'Rcpp'

knitr::opts_chunk$set(python.reticulate = TRUE)
if (identical(Sys.info()[['sysname']], "Windows")) {
    knitr::opts_chunk$set(eval = FALSE)                                      
    msg <- paste("Note: Some examples in this vignette require Python", 
                 "but you are running this vignette on Windows where Python",
                 "is much less likely to be present, or even known to be",
                 "missing (i.e. win-builder) so examples will not be evaluated.")
    msg <- paste(strwrap(msg), collapse="\n")
    message(msg) 
}
options(width=50)

Motivation

The \pkg{RcppCNPy} package by \citet{JOSS:RcppCNPy} provides a simple and reliable access to \pkg{NumPy} files. It does not require \proglang{Python} as it relies on the \pkg{cnpy} library which is connected to \proglang{R} with the help of \pkg{Rcpp} \pkg{Rcpp} \shortcites{CRAN:Rcpp} \citep{JSS:Rcpp,Eddelbuettel:2013:Rcpp,CRAN:Rcpp}.

Now, thanks to the \pkg{reticulate} package by \citet{CRAN:reticulate}, we can consider an alternative which does not require \pkg{cnpy}--but which requires Python. We can (on a correctly set up machine, how to do that is beyond the scope of this note but described in the reticulate documentation) use Python to read \pkg{NumPy} data via \pkg{reticulate}.

file.copy("../tests/fmat.npy", ".")  
file.copy("../tests/fmat.npy.gz", ".")  
file.copy("../tests/fvec.npy", ".")  
file.copy("../tests/imat.npy", ".")  
file.copy("../tests/ivec.npy", ".")  
file.copy("../tests/arr.npy", ".")  

This short note reproduces all the examples in the primary \pkg{RcppCNPy} vignette, but using \pkg{reticulate} instead of \pkg{cnpy}.

Simple Examples

# load reticulate and use it to load numpy
library(reticulate)    
np <- import("numpy")

# data reading
mat <- np$load("fmat.npy")
mat

vec <- np$load("fvec.npy")
vec

Integer data can be read the same way:

imat <- np$load("imat.npy")
imat

Compressed Files

The gzip \proglang{Python} module allows us to access compressed files.

# use the gzip modules for compressed data  
gz <- import("gzip")  
# use it to create handle to uncompressed file 
mat2 <- np$load(gz$GzipFile("fmat.npy.gz","r"))
mat2

Saving Files

Similarly, files can be saved via reticulate access to \pkg{NumPy}.

tfile <- tempfile(fileext=".npy")
set.seed(42)
m <- matrix(sort(rnorm(6)), 3, 2)
m
np$save(tfile, m)

m2 <- np$load(tfile)
m2

all.equal(m, m2)

Savez Array Files

We can also access savez files.

First we save two vectors two different ways:

x <- seq(1, 10)
y <- sin(x)
np$savez("file1.npz", x, y)
np$savez("file2.npz", x=x, y=y)

We can access these files with and without names:

npz1 <- np$load("file1.npz")
npz1$files
npz1$f[["arr_0"]]
npz1$f[["arr_1"]]

Ditto for the second file:

npz2 <- np$load("file2.npz")
npz2$files
npz2$f[["x"]]
npz2$f[["y"]]

Three-dimentional Arrays

We can also import three-dimensional array from \pkg{NumPy} as the next example shows.

arr <- np$load("arr.npy")
arr

unlink("fmat.npy")  
unlink("fmat.npy.gz")  
unlink("fvec.npy")  
unlink("imat.npy")  
unlink("ivec.npy")  
unlink("file1.npz")  
unlink("file2.npz")  
unlink("arr.npy")

Summary

While the \pkg{RcppCNPy} package provides functions for the simple reading and writing of \pkg{NumPy} files, we can also use the \pkg{reticulate} package to access the \pkg{NumPy} functionality directly from \Rns.

\pkg{RcppCNPy} remains an attractive option for simple, direct and lighter-weight file imports whereas \pkg{reticulate} by by \citet{CRAN:reticulate} shines a more full-featured access to many more aspects of \proglang{Python}.

eddelbuettel/rcppcnpy documentation built on Jan. 4, 2024, 6:34 p.m.