paper/joss/2016-09-21-reply.md

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

Reply to Review Comments on RcppCNPy: Read-Write Support for NumPy Files in R

Thank you for your very detailed review and comments.

The package and paper have been updated updated to reflect these suggestions, and the additional documentation is very beneficial.

This reply contains two major section. The first is 'open issues' and addresess items which were brought up as required improvements or changes. The second is 'comments' and either responds to some minor items, or provides other clarification.

Open Issues

Functionality: Installation

R packages need to be loaded, so library("RcppCNPy") (which is case-sensitive) is indeed required.

We do have a section in the README.md showing this as floating point (shown below) and also integer:

R> library(RcppCNPy)
R> fmat <- npyLoad("fmat.npy")

The package contains a demo file (which can be executed via demo("timings", package="RcppCNPy") showing this, and the vignette also shows it.

To be even more explicit we added this also to the 'Examples' sections of the help pages.

That provides four different instances showing how to load the package and perform a few first steps which should be sufficient for users of the package.

Documentation: Automated Tests

The package already contains tests which are executed each time R CMD check runs. The standard R mechanism of using files in the top-level directory tests/ is used.

There are two levels of tests.

Files ending in .Rout.save provide (persistent) reference output which is compared character-by-character to freshly created output during a test. Any differences are reported (but are not fatal).

Additionally, the file tests/loadFiles.R contains five stopifnot() tests which would abort unless the tested condition (of comparing read content to expected content) are met exactly. That provides stringent unit testing.

It is unclear whether an additional markdown or README file is required as the package conforms to standard R and CRAN practice.

Documentation: Community Guidelines

The DESCRIPTION file lists the standard GitHub facilities:

BugReports: https://github.com/eddelbuettel/rcppcnpy/issues

By being on GitHub and listing the repository, the standard and well-known participation venues offered by GitHub are available.

Following this suggestion, we made this more explicit by adding a short section to the README.md as well.

Software Paper: References

We added the DOIs (as supplied by the publisher) for one of the two references. The second reference does not yet have a DOI.

We also added the DOI for this submission as a 'badge' to the main README.md.

Comments

General Checks: Repository

The installation command for R is case-sensitive so the most recent released version can be installed via `install.packages("RcppCNPy").

Should this be desired, R also allows installation directly from the GitHub repository by supplying eddelbuettel/rcppcnpy as the (unique) 'author/repo' token. But as RcppCNPy is a CRAN package so we prefer the more common first approach using install.packages().

General Check: License

Line 17 of DESCRIPTION clearly states License: GPL (>= 2). This exact form is described in the corresponding R manuals, and e.g. expands on the CRAN webpage for the package to 'GPL-2 | GPL-3'. Both licenses are OSI approved, and used by R itself in the same form.

The GNU statement was already present at src/cnpyMod.cpp, the main C++ intgegration script. Files src/cnpy.cpp and src/cnpy.h carry both a statement for the MIT license of the underlying cnpy library by Rogers, and the a short sentence stating GPL for the additional integration provided by this package.

The remaining source file R/cnpy.R is a one-liner so that adding a copyright header seemed excessive.

Functionality: Performance

In general, code from a vignette can always be extracted via Stangle(). Here, we chose to keep the result fixed and static to not endure the (still short) simulation on each latex-compilation of the vignette. However, the package contains a demo detailing this which can be seen via demo("timings", package="RcppCNPy"). It contains the same benchmark. This permits each user to validate performance locally.

Documentation: Statement of Need

The package provides an input/ouput facility so that R users can read and/or write files in the NumPy format. As such, it the target audience is not restricted to any class or set of users, members of discipline, or methodological school.

The key focus is stated is stated in the package description, in the first full sentence in the README.md, and in the first line of the abstract of the vignette.

Documentation: Installation Instructions

As an R package, there should be rather a limited need for the Python specifics particularly as every NumPy version appears to provide the identical output. As such, we have not seen any need to provide more detail.

Documentation: Functionality Documentation

As an R package, primary documentation is provided via the help pages, and then the vignette. Both detail how to use the package.

Recommendation To The Author

Adding .npz files has long been noted as possible extensions, this is also explicitly stated in the vignette. Development of the package was needs-driven: we needed to process certain files, and the npz was not a format we used. Pull-requests by users with the need to process such files are therefore strongly encouraged.

The need to explicitly request integer format is a limitation of the format and the underlying library. There appears to be no meta-data signalling the content.

Recommendation To JOSS

I second this recommendation.

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