On Tue, 2 Apr 2013, Riccardo (Jack) Lucchetti wrote:
On Tue, 2 Apr 2013, Sven Schreiber wrote:
> a colleague asked me about a non-convergence problem with his
> application of the SVAR package. I can replicate the error, which gives:
> Das Konvergenzkriterium wurde nicht getroffen
> *** error in function estC
> > catch scalar ll = NRmax(theta, "loglik(&theta, &dat, -1)",
> Sorry for the German, but this actually gave me a thought here: the
> error message ("convergence criterion not met", in German) comes from an
> internal function because it is translated. I assume it's from NRmax(),
> but then I wonder: isn't Jack's use of 'catch' supposed to suppress
> error message or at least suppress the interruption?
In theory, yes. Could you send me the data + a script, please?
I've just fixed a probably relevant bug in CVS. If "catch" is
used in a loop, and it applies to a genr expression which gets
compiled, then the "catchiness" was wearing off after one go:
on the second or subsequent execution of the given compiled
line the error would not be trapped.