On Fri, 2 Oct 2009, Sven Schreiber wrote:
I'm using this opportunity to point out the feature requests on
tracker that are more than one year old. Here's a list with some
thoughts: (I just closed the VECM-related feature request, see my
* Improvement of logit regression modelling in gretl - ID: 1886339
E.g. multinomial logit -- this is of course a big thing and should just
wait I guess.
As Jack said, we now do multinomial logit. That has been
available since gretl 1.8.1 (May, 2009). I'm not sure it's worth
implementing the remaining points: if the user wants a reference
category other than the default (i.e. the lowest value), it's easy
enough to redefine the dependent variable accordingly, as in
alt_y = (y == 5)? 0 : y
to make (y = 5) the reference category (assuming there isn't
already a value 0 in the y series).
* access to the ranking() command from the GUI - ID: 1906097
IMHO there is no reason not to decide on what to do with it now, no? Is
there a plan to implement it (probably in the vicinity of 'add logs'
etc.) or should it be discarded?
I'm not very enthusiastic. If we add rankings to the GUI menu
we'd presumably have to add sorted series also, and my feeling is
that this would clutter the menu.
* Saving rankings from difftest () to the data set - ID: 1906103
Not sure what the use case is. Would it be enough to implement the
previous feature? Actually if the previous feature is implemented I
guess it would be relatively easy to write a function package that
covers this request. So in that sense I would tend to close it, adding
an appropriate comment suggesting such a package.
I like the idea of hiving this off to a package.
BTW, in the test output I found some strings that are not yet marked
string: (zero differences: 0, non-zero ties: 0)
string: Two-tailed p-value = 0.2
strings: value rank group
OK, I'll take a look at that.
* gnuplot --loess-fit - ID: 2000792
more precisely, this is about getting date labels on the x-axis in
combination with --loess-fit. Here I would also tend to outsource it to
someone with sufficient gnuplot knowledge who could make a function
package for it. Or is this 'solution' too lazy and it really should be
done in gretl proper?
The thing is that we don't offer the --loess-fit option (or in
fact, any fit option) for time-series plots. You can get such a
plot by doing
plot y time --loess
But then it's quite right that the x-axis values are 1, 2, 3, ...
because those are the values of the variable called "time".
So I suppose this one is really a request to implement the gnuplot
"fit" options for time-series plots.
* $hausman should also access random vs. fixed effects test - ID:
well this is a request by myself, so of course I think it should be
Yes, that one should be implemented -- and now it is, in CVS.