On Thu, July 6, 2006 11:58, Sven Schreiber wrote:
Unfortunately I don't have time in the next couple of days, but I
thinking that maybe a more formalized testing period would be useful to
make sure that no existing things were broken by the internal
restructuring. I have in mind things like the histogram bug (continuous
vs. discrete) that appeared in 1.5.1. (No offense intended, Jack, it's
just an example ;-)
Oh, no, you're 100% right: that was a last-minute addition. Exactly what I was
As far as I'm concerned, testing can start NOW. Apart from fixes that go in at
the last minute of extra time (pun intended, Sven :-)), we don't foresee any
code change before release.
So maybe put out a 1.6 release candidate for two or three weeks? It
would also help if Allin and Jack could tell us what areas have been
rewritten and that are therefore most likely to contain new bugs.
Well, arima for a start. The new default is now to estimate with full ML
through the Kalman filter. We ran as many benchmarks as we could find, and I'm
now convinced the arma code is pretty good, but I may be wrong. In some cases,
you have a few discrepancies against x-12; in all those cases, reasonably
accurate scrutiny convinced me that native estimates are better (this occurs
especially when the MA part is nearly non-invertible).
Panel data: test the new mechanism for importing data; the statistics should
be ok (again, we ran quite a few benchmarks, but it's never enough). This is
especially important as we plan to implement Arellano-Bond after 1.6.0.
The new function setup: you may have noticed, there's a new mechanism in place
for accessing user-written functions from the GUI. This needs to be tested.
Besides, writing functions is one of the best way to uncover bugs.
The manual has been restructured, with some bit rewritten in large part (arma
& panel), and a few bits added: for example, there's a new chapter entitled
"Discrete variables". The features that we talk about in there have not been
tested a lot, so please have a go at it.
Multiple precision: Talha has already beaten the cr*p out of it, so it should
be in good shape, but have a look at it if you can.
Matrices: a few corner cases were fixed, but there may be still bugs lurking
around. People with decent gauss/matlab/ox experience may want to convert some
of their programs to our own matrix syntax to check if it works ok.
You'll also notice random bits here and there, like multiple time-series
graphs and a partially chnged menu structure: comments on usability are very
Btw, I'm very happy with gretl's progress!
Glad to hear that. Thanks!
Riccardo "Jack" Lucchetti
Dipartimento di Economia
Facoltà di Economia "G. Fuà"