Students interested in working on the
akima replacement proposal should send me off-list answers to this test:
Set up and document a benchmarking test to be able to see whether the future replacement package for akima performs in the same ways as the package being replaced. This benchmarking test is essential so that when implementing the new package, we'll know to what extent other packages can depend on it. The test may be very simple, but it should span the use cases adequately.
Detailed points:
Review and analyse the use of the functions in the akima package in
a sample of packages in CRAN, the comprehensive R archive network. Use
of awk/sed on files in the man/ directories of the packages using akima
may be a good way in - spotting and extracting uses of functions in
akima; the source packages are all on CRAN in predictable locations for a
given mirror.
I attach a text file to load into R giving the
names of packages that depend on, import or suggest akima (March 18, 2013). Reading the R
admin. manual may show how the package system works; the pkgDepTools
package from Bioconductor was used here to analyse the dependency tree
of packages and generate the two character (string) vectors of names,
which may not be unique.
Some uses are trivial, not altering the
defaults used in the akima package, but others permit the user to modify
arguments, and these will be the best test of a drop-in replacement.
The test will be to set up an effective benchmark, so that we can know
to what extent the drop-in replacement gives users the same or similar
results, and the same or similar control over the results.
Of
course, the examples in the help files in the akima package itself are a
good place to start, but the other packages using akima may exercise it
ways not anticipated by the package authors.
Roger Bivand