Skip to content
Back to formatted view

Raw Message

Message-ID: <20110605210558.GA5658@praha1.ff.cuni.cz>
Date: 2011-06-05T21:05:58Z
From: Petr Savicky
Subject: reproduction archives
In-Reply-To: <4DEBE210.3040906@ufl.edu>

On Sun, Jun 05, 2011 at 03:07:44PM -0500, Alexander Shenkin wrote:
> Hello Folks,
> 
> As some of my old code broke when an updated package changed its
> interface, I started thinking about reproduction of analyses.  It's not
> good enough to save our code - we have to save the package versions
> those analyses used as well as the R-core.  I saw a couple references to
> "reproduction archives" around, but nothing specific.  Is there any good
> way to package up code along with the relevant packages and R version in
> order to guarantee that we can reproduce our results in the future?  I
> suppose that one could make note of all package versions, etc, but
> automating the process would mean that it would be followed more often. 
> (also, tracking down those packages and installing them all just to
> reproduce a result could be an undertaking).

Hello Allie:

Reproducibility of the results is an important question. Personally,
i do not believe much to automated solutions, although i may be wrong.
When i store a result of a simulation for future reference as
an ".rda" file, then the script includes into it an object "identification",
which contains seeds, a critical function used in the simulation
and also a package and R version. The last two things may be obtained
for example as

  R.version.string
  library(help=mvtnorm)[[3]][[1]][3]

Up to now, i did not really needed these things, so i also do not
know, whether they are sufficient.

Petr Savicky.