Re: [Gretl-users] saving session vs saving data
On Fri, 11 Sep 2009, Summers, Peter wrote:
Allin & Sven, many thanks for your prompt responses and clear
explanations. I hope you don't mind another two cents' worth on
this.
Not at all.
I like the provisional fix -- in fact I was typing a suggestion
along those lines when I got the message below. You really do
rock, Allin!
;-) thanks.
I like the idea of preserving an 'in-session' version of the
data set, but I can also see confusion arising if, say, I end up
with two slightly different versions of what I think are the
same data (ie, 'my data.gdt' and 'my data' that's saved in 'my
session.gretl'). Maybe there could also be a prompt or option to
update the .gdt file from within the session?
This is tricky. There are various possible scenarios.
1) I load a data file, do some econometrics, and save the session.
The original data file is still on my hard drive, and I decide to
update it at some point. The version in my session file is now
out of sync. Having some sort of sync option could be nice, but
it could also be fiendishly complicated. Suppose the "update"
involved deleting some variables, but those variables were used in
models saved in the session file. Or suppose the update revised
the values of some series that were used in the models. Now the
estimates are out of sync, plus some tests on the saved models
could now go awry.
2) Since the session file was saved, the original data file has
been deleted or moved. Perhaps an unrelated data file has been
created, with the same name as the one that was originally loaded
into the session!
3) I email my session file to someone else. The other party
happens to have an unrelated gretl data file of the same name as
the one I used.
And so on. This could get really ugly. I think the session-file
has to be a self-contained snapshot. But this should be explained
properly in the manual.
On the other hand, if you have a script that carries out the
econometrics, there's no problem in rerunning that script on
a new version of the dataset. (Well, it might involve some
revisions to the script, but that's easy.)
Allin.