1:05 p.m.
Hi all,
I obtain an error running the following script + panel dataset. The
error message I receive is
*** Error in `/usr/bin/gretl_x11': corrupted double-linked list:
0x0a2a82c0 ***
and gretl stops working. I am using current cvs on ubuntu.
Best,
Artur
1:59 p.m.
On Tue, 10 Sep 2013, Artur Tarassow wrote:
I obtain an error running the following script + panel dataset. The
error message I receive is
*** Error in `/usr/bin/gretl_x11': corrupted double-linked list:
0x0a2a82c0 ***
and gretl stops working. I am using current cvs on ubuntu.
Thanks for the report. I'll work on the problem, but in the meantime note
that you can work around it by simplifying your "outlier" function. You
have:
<hansl>
function series outlier (series Y)
smpl Y > quantile(Y,0.01) && Y < quantile(Y,0.99) --restrict --replace
series Ytmp = Y
smpl full
Y = Ytmp
return Y
end function
</hansl>
and the segfault is coming from incorrect handling of the panel
sub-sampling and restoration inside the function, which needs to be fixed.
But you don't need to mess with the sample; this should have the same
effect:
<hansl>
function series outlier (series Y)
scalar q01 = quantile(Y,0.01)
scalar q99 = quantile(Y,0.99)
return Y > q01 && Y < q99 ? Y : NA
end function
</hansl>
Allin
7:56 p.m.
On Tue, 10 Sep 2013, Artur Tarassow wrote:
I obtain an error running the following script + panel dataset. The
error message I receive is
*** Error in `/usr/bin/gretl_x11': corrupted double-linked list:
0x0a2a82c0 ***
and gretl stops working. I am using current cvs on ubuntu.
I finally found the place in gretl_panel.c where it says:
/* FIXME sub-sampled dataset (needs to be disallowed?) */
The thing that I thought might have to be disallowed (and I now see it
does have to be!) is the setting of panel structure as in your
setobs ID year --panel
after sub-sampling the dataset. That now generates an error message.
However, there are other ways of doing what you want.
1) Just do "setobs ID year --panel" once, right after opening your
dataset. Your subsequent sample restrictions do not destroy the panel
structure, since they just involve removing (all countries for certain
years) or (all years for certain countries).
2) If you want a "really" smaller dataset, then
i) Apply all your restrictions.
ii) Save the reduced dataset ("store")
iii) Open the reduced dataset and do setobs ID year --panel
Allin
7:42 a.m.
Am 10.09.2013 21:56, schrieb Allin Cottrell:
On Tue, 10 Sep 2013, Artur Tarassow wrote:
> I obtain an error running the following script + panel dataset. The
> error message I receive is
>
> *** Error in `/usr/bin/gretl_x11': corrupted double-linked list:
> 0x0a2a82c0 ***
>
> and gretl stops working. I am using current cvs on ubuntu.
I finally found the place in gretl_panel.c where it says:
/* FIXME sub-sampled dataset (needs to be disallowed?) */
The thing that I thought might have to be disallowed (and I now see it
does have to be!) is the setting of panel structure as in your
setobs ID year --panel
after sub-sampling the dataset. That now generates an error message.
However, there are other ways of doing what you want.
1) Just do "setobs ID year --panel" once, right after opening your
dataset. Your subsequent sample restrictions do not destroy the panel
structure, since they just involve removing (all countries for certain
years) or (all years for certain countries).
This works nicely, now. Actually I
never fully understood why the panel
structure was destroyed in the case of setting sample restrictions
before. I think STATA never does something like this. But maybe I am
wrong. And of course STATA doesn't have to be the reference case, anyway...
2) If you want a "really" smaller dataset, then
i) Apply all your restrictions.
ii) Save the reduced dataset ("store")
iii) Open the reduced dataset and do setobs ID year --panel
Yes, that's what I
am usually doing if I need a specific subsample for
some more complex operations.
Allin
_______________________________________________
Gretl-users mailing list
Gretl-users(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-users
Thank you again!
Artur
4126
days inactive
4127
days old
3 comments
2 participants
participants (2)
-
Allin Cottrell -
Artur Tarassow