[Gretl-devel] Re: The any() function
Hi
Am 01.05.21 um 01:53 schrieb Allin Cottrell:
On Thu, 29 Apr 2021, Riccardo (Jack) Lucchetti wrote:
> Folks,
>
> from time to time I need to check if a certain object belongs or not
> to a set of integers. For example, suppose you have a panel and you
> want to select only certain units, whose id you list in a vector.
>
> This can of course be generalised, so I wrote a little hansl function
> (I called it any() because of my lack of imagination: alternative
> proposals are very welcome):
Below I added another case which is, however, not supported, as matrices
are not supported by the meta-type 'numeric' (yet?): array of matrices.
This is also a case supported by MATLAB:
https://www.mathworks.com/help/matlab/ref/any.html
> <hansl>
> set verbose off
> set seed 999
>
> function numeric any(numeric X, const matrix A)
> # checks if X belongs to the set A
> if typeof(X) == 1
> # scalar
> scalar ret = max(A .= X)
> elif typeof(X) == 2
> # series
> series valid = ok(X)
> series ret = NA
> smpl valid --dummy
> matrix tmp = { X } .= vec(A)'
> ret = sumr(tmp) .> 0
> smpl full
> elif typeof(X) == 3
> # matrix
> scalar r = rows(X)
> scalar c = cols(X)
> matrix ret = maxr(vec(X) .= vec(A)')
> ret = mshape(ret, r, c)
elif typeof(X) == 6 #
untested
matrix ret = array(nelem(X))
loop i=1..nelem(X)
matrix m = X[i]
# matrix
scalar r = rows(m)
scalar c = cols(m)
matrix rtmp = maxr(vec(m) .= vec(A)')
ret[i] = mshape(rtmp, r, c)
endloop
> endif
> return ret
> end function
>
> ###
> ### usage example
> ###
# array of matrices
foo = {1, 3, 11}
matrix Z1 = mrandgen(i, 1, 6, 6, 3)
matrix Z2 = mrandgen(i, 1, 6, 6, 3)
matrices ZA = defarray(Z1, Z2)
eval any(ZA, foo)
> </hansl>
>
> If this is useful, we could either (a) add this to the extra package
> or (b) make this a native libgretl function. Comments?
Good idea; I think this could be useful in a number of contexts.
I'm reluctant to mess with git master when we're about to release, but
in a local branch I've tried adding a libgretl function to this effect
(diff against master attached).
Two small comments:
(1) I've named my built-in function inset() rather than any().
I guess that Jack had Matlab's function name in mind:
https://www.mathworks.com/help/matlab/ref/any.html
(2) I've switched the order of the arguments: it seems to me a
little
more intuitive to call
inset(S, whatever)
where S is the "set" and "whatever" is the object to be assessed in
term
of membership of the set (scalar, series or matrix).
One further point: hansl has no concept of "set" as such; the doc for
such a function would have to make clear that the "S" argument is a
matrix standing in for a set (with numeric elements only). Ideally,
therefore, S would be a vector with no repeated elements.
But my inset(), like your any(), is permissive on that point. An
arbitrary matrix will be accepted: the row/column structure is ignored
and repeated elements are tolerated but will just slow things down.
True, the term 'set' is a bit misleading here. Suppose at some point in
future, gretl supports a set, the naming of this inset() function may
confuse people as it's not really about a set of values.
We already have the following functions:
- inlist()
- inbundle()
- instring()
- instrings()
Following this list, a candidate name may be innumeric() or innum().
Best,
Artur