Re: [Gretl-devel] imaxc/imaxr
Am 21.04.2015 um 13:25 schrieb Riccardo (Jack) Lucchetti:
Folks, consider this:
<hansl>
x = imaxr({1,2,1,2,1,2})
</hansl>
At present, this returns the scalar 2, but I wonder if it shouldn't
instead return a row vector holding {2,4,6}. But then, generalisation to
the case when the argument to imaxr is a matrix with more than 1 row
becomes awkward. Thoughts?
A Jack-style answer might be: You could use
<hansl>
matrix m1 = {1,2,1,2,1,2}
matrix check = selifc( seq(1, cols(m1)), (m1 .>= max(m1)) )
print check
</hansl>
to get the vector output, but I know that that's not the point of the
question and you knew that already.
FWIW, (Python's) Numpy's argmax() and argmin() functions explicitly note
that:
"In case of multiple occurrences of the maximum values, the indices
corresponding to the first occurrence are returned."
I'd say that's fine, and for other use cases I'd tend to recommend
workarounds such as the above. Alternatively, perhaps an optional second
argument to the iminr/iminc/imaxc/imaxr functions to request a check for
all occurrences?
(BTW, it seems suboptimal to me to have four different function names
that essentially do the same thing. But probably too late now.)
cheers,
sven