On Wed, 5 Nov 2014, Sven Schreiber wrote:
> I'm wondering whether some of the function packages are still needed,
> because sometimes it looks as if the same functionality is already built
> into gretl. Here's a quick list in random order:
>
> 1) "clustered_ols" by Claudia Pigini. In what way is this different from
> the built-in "ols ... --cluster=xyz"?
>
> 2) "cnumber" by Monsueto, S.E -- Here I don't think it's redundant but
> I'm not getting the same results as with other code that should (?) be
> equivalent. Consider this example and please explain why the results are
> different:
> <hansl>
> include cnumber.gfn
> open greene12_1.gdt
> matrix X = {age, income, expend}
> cnumber (X)
> l = eigensym(X'X, null) # lines taken from Allin's email
> CN = maxc(l)/minc(l) #
> printf "CN = %g, (square root = %g)\n", CN, sqrt(CN)
> </hansl>
>
> 3) "fgls" by Yi-Nung Yang: the help just says feasible GLS which is very
> vague and generic; in what way is this different from gretl's "hsk" or
> "wls" commands? If it is different, I would suggest to make the help
> text more informative.
>
> 4) "GHegy" by Ignacio Díaz-Emparanza and "HEGY_test" by Jack Lucchetti.
> Could you guys perhaps consolidate these?
>
> 5) "HoltWinters" by Ignacio Díaz-Emparanza. This sounds a lot like it's
> also done by the "fcModels" package by Yi-Nung Yang. Is this true?
>
> 6) "JB" by Yi-Nung Yang. This does a Jarque-Bera test and even the help
> text mentions that it is built into gretl (as "normtest ... --jbera").
> So it's not obvious to me what the purpose of the package is.
>
> 7) "MWU" and "mwu" by Yi-Nung Yang. AFAICS the only difference is
> whether you pass the arguments as series or as vectors (matrices). This
> may be a matter of taste, but personally I would suggest that you choose
> which is the interface of your function, i.e. series or matrices, and
> then the user/caller will have to follow that rule, and the other
> package with the other interface would be gone.
> And with respect to "mwu_dummy", this is even more of a personal taste
> thing, but perhaps you could consolidate all this into one function, for
> example by introducing a third function parameter with a default value:
> "function void mwu(series x, series y, int dummy[0])"
> So to do what "mwu_dummy()" does now you would pass your 'group'
> variable as y, and call the function with dummy=1.
I won't go into all the specifics -- although I think you're right
-- but rather I'll address the general issue here.
We could really do with some editoral control over the function
packages. Up till now we have been totally "laissez faire", but if
the packages on the server are supposed to be a showcase for adding
to gretl's built-in functionality they need to be pruned and
consolidated. I wonder if we could assemble a committee of 2 or 3
members to work on this.
One other thought: some of the current packages that are strictly
redundant might nonetheless be preserved in some form as nice
examples of scripting. For example, Claudia Pigini's
"clustered_ols". Maybe there should be an "example scripts" area
separate from the function package area?
Allin
