Am 06.05.2021 um 07:42 schrieb atecon: > Hi all, > > I just would like to ask whether this behavior is actually expected: > > <hansl> > matrix m = {1, NA; 2, 3} > print m > eval cum(m) > > ? print m > m (2 x 2) > >     1   nan >     2     3 > > ? eval cum(m) >     1   nan >     3   nan > </hansl> > > I expected the entry (2,2) to be 3 as this is the first valid value of > the 2nd column. I'd say it's expected because nans propagate. Basically you don't know what 3 + nan is. To get what you want, I was going to say "do cum(misszero(m))", but that doesn't work on a matrix. Maybe it should? (But remember the old fact that a NaN and a missing value is not always the same thing.)

Right, but I'm wondering whether this isn't inefficient because the ok values are overwritten with themselves. Or aren't they internally?

On Thu, 6 May 2021, Sven Schreiber wrote: > Am 06.05.2021 um 10:05 schrieb Sven Schreiber: > >> >> Right, but I'm wondering whether this isn't inefficient because the ok >> values are overwritten with themselves. Or aren't they internally? > > BTW, shouldn't this work? > > <hansl> > matrix m = {1, NA} > eval misszero(m[2]) > </hansl> > > I'm getting an error "is matrix not series", but m[2] should -I believe- > be auto-converted to scalar type here. Actually no. The policy for some time now has been that a 1x1 submatrix remains a matrix in the course of calculation, and is auto-converted to a scalar only on assignment. This seemed to be the way to minimize unexpected results.