Re: [Gretl-devel] some bugs and issues
On Fri, 20 Jan 2012, Talha Yalta wrote:
I would like to report a few problems or issues:
1)- In the "Restrict based on criterion" window, if I use comma as
the decimal sign I get a syntax error.
That's expected, and will not change: when you specify a criterion
for restricting the sample "based on a criterion" you're accessing
the "genr" mechanism, and that mechanism doesn't/can't support the
decimal comma, for one thing because in that context ',' is the
separator for function arguments.
2)- Why is it that many of the tests are not available when WLS (and
heteroscedasticity corrected OLS) is used? I am asking this because I
can easily obtain the same results with whighting the model manually,
in which case I can also use the White test etc.(and see H0 is
rejected).
Maybe this should be relaxed. First, however, I'd want to be clearer
in my own mind about what is the "right" thing to do in regard to
the various residual-based tests after WLS estimation.
3)- When I add observations using Data>Add observations, the
forecasts
window still says "There are no observations available for forecasting
out of sample...". (after Cochrane-Orcutt, HL, PW etc.)
If your model includes exogenous stochastic regressors, as one would
usually expect with Cochrane-Orcutt and so on, then gretl cannot
extend these automatically, and so cannot forecast out of sample
unless you manually fill in the extra values of the relevant series.
But if the models you're talking about have only simple
deterministic terms on the right-hand side then out-of-sample
forecasting should work once you've requested that observations be
added. So if that's not working, please give details.
4)- When I import a data file with labels, gretl gives the message:
"first row label "xxx", last label "zzz" label strings can't
be
consistent dates" This is not an error message right? Because it
sounds like one. I think the "label strings can't be consistent dates"
part is unnecessary and confusing. What do you think?
No, it's not an error message as such. But in current CVS and
snapshots that message should only be printed on stderr, and not
shown to the user in a dialog box. Please let us know if that's not
the case.
5)- IMO the data menu (along with the variable menu) is really too
long. How about considering the following changes to make it look less
intimidating:
a)- Removing the refresh window item. (I don't know if it is
really needed. Besides it doesn't even look like it belongs to the
data menu.)
This item _ought_ to be redundant, since the main data window is
supposed to refresh automatically. Has anyone seen a case lately
where the automatic refresh does not occur? If not, we can probably
get rid of it.
b)- Making a default named list called "all variables"
and
removing the "select all" item. Also, could you provide a means to
delete such lists.
"Select all" is a standard GUI option which admittedly may be useful
only rarely in gretl, but which -- it seems to me -- is not
equivalent to defining a list containing all variables. On your
second point, yes, I guess deleting lists via the GUI should
probably be supported.
c)- Having a single entry for compact data and expand data
That would reduce the menu slightly, but complicate considerably the
dialog box that the menu item calls up.
Allin Cottrell