10:12 p.m.
Hello everybody,
I am surprised that it is not possible to save models within the loop
function. I really thought that this worked some versions ago, but I
might e wrong.
I've got a list C with 10 variables and for each variable I want to run
a regression:
loop foreach i C
model_$i <- ols $i const Y
endloop
Is there a reason why it is not activated? I mean if it is possible to
save models manually by "model <- ols y const x" it should also be
possible to do the same within the loop command; at least for consistency.
Cheers,
Artur
12:45 a.m.
On Thu, 22 Oct 2009, Artur T. wrote:
I am surprised that it is not possible to save models within the
loop function. I really thought that this worked some versions
ago, but I might e wrong.
I'm not 100% certain, but pretty sure this has never worked. The
difficulty is that commands within a loop are handled "deep
inside" libgretl, as opposed to commands registered via GUI
point-and-click or the gui console. (The processing of commands
within a loop is shared between the gretl GUI program and
gretlcli, which has no concept of "named models".)
So to get this sort of thing to work we need to exploit a
"callback function" from libgretl to the GUI program, whereby
libgretl says, "Hey, the user wants to save this model under a
given name, can you handle that?".
As it happens, I'm working on that right now (I started earlier
today). What I'm aiming to implement is:
* the ability to save a named model within a loop;
* the ability to add the last model to the "model table" within a
loop; plus
* the corresponding functionality for graphs within loops (save
named graph; add graph to the GUI "graph page").
So this is something that, hopefully, should be available before
too long.
Allin
7:23 a.m.
On Thu, 22 Oct 2009, Allin Cottrell wrote:
On Thu, 22 Oct 2009, Artur T. wrote:
> I am surprised that it is not possible to save models within the
> loop function. I really thought that this worked some versions
> ago, but I might e wrong.
I'm not 100% certain, but pretty sure this has never worked. The
difficulty is that commands within a loop are handled "deep
inside" libgretl, as opposed to commands registered via GUI
point-and-click or the gui console. (The processing of commands
within a loop is shared between the gretl GUI program and
gretlcli, which has no concept of "named models".)
So to get this sort of thing to work we need to exploit a
"callback function" from libgretl to the GUI program, whereby
libgretl says, "Hey, the user wants to save this model under a
given name, can you handle that?".
As it happens, I'm working on that right now (I started earlier
today). What I'm aiming to implement is:
* the ability to save a named model within a loop;
* the ability to add the last model to the "model table" within a
loop; plus
* the corresponding functionality for graphs within loops (save
named graph; add graph to the GUI "graph page").
So this is something that, hopefully, should be available before
too long.
FWIW, I'm not entirely enthusiastic about this kind of thing. Sure, the
more you can do (especially in a script), the better. But my spider sense
is tingling --- there are many many things that can go wrong with this
apparatus.
By the way, what's the point of storing models in a loop? The only
situation I can think of is when you have to apply the same estimation
method over and over again and need to compare the results at the end.
Except when you have very few of them, you probably won't want to compare
them by clicking on each model in turn and having a gazillion windows open
on your desktop. What you probably want to do is store the relevant
quantities (coefficients, residuals, information criteria) somewhere (in a
matrix, a text file, whatever) and start from there.
That said, I would love to be wrong!
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
9:16 a.m.
Riccardo (Jack) Lucchetti schrieb:
On Thu, 22 Oct 2009, Allin Cottrell wrote:
>
> On Thu, 22 Oct 2009, Artur T. wrote:
>
>> I am surprised that it is not possible to save models within the
>> loop function. I really thought that this worked some versions
>> ago, but I might e wrong.
>
> I'm not 100% certain, but pretty sure this has never worked. The
> difficulty is that commands within a loop are handled "deep
> inside" libgretl, as opposed to commands registered via GUI
> point-and-click or the gui console. (The processing of commands
> within a loop is shared between the gretl GUI program and
> gretlcli, which has no concept of "named models".)
>
> So to get this sort of thing to work we need to exploit a
> "callback function" from libgretl to the GUI program, whereby
> libgretl says, "Hey, the user wants to save this model under a
> given name, can you handle that?".
>
> As it happens, I'm working on that right now (I started earlier
> today). What I'm aiming to implement is:
>
> * the ability to save a named model within a loop;
>
> * the ability to add the last model to the "model table" within a
> loop; plus
>
> * the corresponding functionality for graphs within loops (save
> named graph; add graph to the GUI "graph page").
>
> So this is something that, hopefully, should be available before
> too long.
FWIW, I'm not entirely enthusiastic about this kind of thing. Sure, the
more you can do (especially in a script), the better. But my spider sense
is tingling --- there are many many things that can go wrong with this
apparatus.
By the way, what's the point of storing models in a loop? The only
situation I can think of is when you have to apply the same estimation
method over and over again and need to compare the results at the end.
Except when you have very few of them, you probably won't want to
compare them by clicking on each model in turn and having a gazillion
windows open on your desktop. What you probably want to do is store
the relevant quantities (coefficients, residuals, information
criteria) somewhere (in a matrix, a text file, whatever) and start
from there.
Yes, I thought it is a convenient way to compare the models. But
actually you might be right. If the implementation is accompanied by
possible problems "with the apparatus" than the negative effect of it
might be higher than the positive one.
But at least the saving of graphs (using different options) should be
possible. ;)
That said, I would love to be wrong!
Riccardo (Jack) Lucchetti
Dipartimento di Economia
Università Politecnica delle Marche
r.lucchetti(a)univpm.it
http://www.econ.univpm.it/lucchetti
------------------------------------------------------------------------
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
12:46 p.m.
On Fri, 23 Oct 2009, Riccardo (Jack) Lucchetti wrote:
On Thu, 22 Oct 2009, Allin Cottrell wrote:
> What I'm aiming to implement is:
>
> * the ability to save a named model within a loop;
>
> * the ability to add the last model to the "model table" within a
> loop; plus
>
> * the corresponding functionality for graphs within loops (save
> named graph; add graph to the GUI "graph page").
>
> So this is something that, hopefully, should be available before
> too long.
FWIW, I'm not entirely enthusiastic about this kind of thing. Sure, the
more you can do (especially in a script), the better. But my spider sense
is tingling --- there are many many things that can go wrong with this
apparatus.
You may well be right, but what sort of going-wrong do you have in
mind?
By the way, what's the point of storing models in a loop? The
only
situation I can think of is when you have to apply the same estimation
method over and over again and need to compare the results at the end.
Except when you have very few of them, you probably won't want to compare
them by clicking on each model in turn and having a gazillion windows open
on your desktop. What you probably want to do is store the relevant
quantities (coefficients, residuals, information criteria) somewhere (in a
matrix, a text file, whatever) and start from there.
I agree, that will often be a smarter thing to do. I was thinking
of the case where you're estimating relatively few models (maybe
the same model on different samples), and you'd like to be able to
compare the models side by side.
Of course in such a case it shouldn't be too burdensome to
"unroll" the loop, yet in some contexts there may be a fair amount
of setup to be repeated each time, so that a loop might be
noticeably more convenient.
Allin.
5544
days inactive
5545
days old
4 comments
3 participants
participants (3)
-
Allin Cottrell -
Artur T. -
Riccardo (Jack) Lucchetti