11:46 a.m.
Hi
A warning.
the Model Var Autoregression in the GUI can crash out Gretl -no error
message re estimation no. of variables too few or anything -it just
disappears from the screen. So all preceding work done without the outputs
saved in the session (not necessarily only from previous Var
autoregressions ) will be lost. Mine is the latest Gretl version.
Brian J Revell
Professor Emeritus
12:02 p.m.
Am 20.12.20 um 12:46 schrieb Brian Revell:
Hi
A warning.
the Model Var Autoregression in the GUI can crash out Gretl -no error
message re estimation no. of variables too few or anything -it just
disappears from the screen. So all preceding work done without the
outputs saved in the session (not necessarily only from previous Var
autoregressions ) will be lost. Mine is the latest Gretl version.
Hi,
I cannot replicate this on Ubuntu 20.04 with latest git. Brian could you
provide a detailed step-by-step guide what you do? Also, which OS are
you using?
I did the following:
Open the data set "AWM18.gdt" --> Open the VAR menu --> Select 3
arbitrary series, keep everything else at default values --> Estimate
the model
Artur
1:26 p.m.
Am 20.12.2020 um 13:02 schrieb Artur Tarassow:
Am 20.12.20 um 12:46 schrieb Brian Revell:
> Hi
> A warning.
> the Model Var Autoregression in the GUI can crash out Gretl -no
> error message re estimation no. of variables too few or anything -it
> just disappears from the screen. So all preceding work done without
> the outputs saved in the session (not necessarily only from previous
> Var autoregressions ) will be lost. Mine is the latest Gretl version.
Hi,
I cannot replicate this on Ubuntu 20.04 with latest git. Brian could
you provide a detailed step-by-step guide what you do? Also, which OS
are you using?
As per the changelog (http://gretl.sourceforge.net/ChangeLog.html)
there
is one known crash bug currently since the latest release, quoting:
Fix bug: potential crash on "Save as icon and close" for a model displayed in
tabbed model viewer
Is this by any chance what you did / what was happening?
thanks
sven
3:34 p.m.
Hi Sven
the specification was just pushing data and parameters to their limits with
specific lags of 1 3 6 and 9 years specified in the 2 dependent variables
to try to pick up some cyclical elements in the data and with only 22
observations. I will try it dropping the lag 9 later (I have other ARMAX
results I don't want to risk losing at present until I have finished the
session) to see if it makes the difference between crash or go!
Windows 10 OS. However, i would have expected some failure messages, not a
complete crash.
Brian
On Sun, 20 Dec 2020 at 13:26, Sven Schreiber <svetosch(a)gmx.net> wrote:
Am 20.12.2020 um 13:02 schrieb Artur Tarassow:
Am 20.12.20 um 12:46 schrieb Brian Revell:
Hi
A warning.
the Model Var Autoregression in the GUI can crash out Gretl -no error
message re estimation no. of variables too few or anything -it just
disappears from the screen. So all preceding work done without the outputs
saved in the session (not necessarily only from previous Var
autoregressions ) will be lost. Mine is the latest Gretl version.
Hi,
I cannot replicate this on Ubuntu 20.04 with latest git. Brian could you
provide a detailed step-by-step guide what you do? Also, which OS are you
using?
As per the changelog (http://gretl.sourceforge.net/ChangeLog.html) there
is one known crash bug currently since the latest release, quoting:
Fix bug: potential crash on "Save as icon and close" for a model displayed in
tabbed model viewer
Is this by any chance what you did / what was happening?
thanks
sven
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
3:44 p.m.
Am 20.12.2020 um 16:34 schrieb Brian Revell:
Hi Sven
the specification was just pushing data and parameters to their limits
with specific lags of 1 3 6 and 9 years specified in the 2 dependent
variables to try to pick up some cyclical elements in the data and
with only 22 observations. I will try it dropping the lag 9 later (I
have other ARMAX results I don't want to risk losing at present until
I have finished the session) to see if it makes the difference between
crash or go!
Windows 10 OS. However, i would have expected some failure messages,
not a complete crash.
Brian
Thanks, that's already enough - I can confirm the crash here on Windows
with the latest snapshot (with 22 obs and using the GUI to enter the lag
choice "1,3,6,9", without the quotes).
Of course, with a max lag of 9 you only have 13 effective obs left and
that is just barely enough to estimate 4*2 + 2 = 10 parameters in each
equation. Nonetheless, in theory it should still work (although I
wouldn't recommend it econometrically), and needless to say a crash is
never acceptable.
I suspect that the fix will be relatively easy (for Allin, that is ;-)
cheers
sven
4:45 p.m.
Nope I agree -it was just a look see exercise as a VAR as there might have
been some feedback in the system between the 2 variables. Back to ARMAX
univariate specifications..
AT least I have pointed out an issue that is replicable.
cheers
Brian
On Sun, 20 Dec 2020 at 15:44, Sven Schreiber <svetosch(a)gmx.net> wrote:
Am 20.12.2020 um 16:34 schrieb Brian Revell:
Hi Sven
the specification was just pushing data and parameters to their limits
with specific lags of 1 3 6 and 9 years specified in the 2 dependent
variables to try to pick up some cyclical elements in the data and with
only 22 observations. I will try it dropping the lag 9 later (I have other
ARMAX results I don't want to risk losing at present until I have finished
the session) to see if it makes the difference between crash or go!
Windows 10 OS. However, i would have expected some failure messages, not a
complete crash.
Brian
Thanks, that's already enough - I can confirm the crash here on Windows
with the latest snapshot (with 22 obs and using the GUI to enter the lag
choice "1,3,6,9", without the quotes).
Of course, with a max lag of 9 you only have 13 effective obs left and
that is just barely enough to estimate 4*2 + 2 = 10 parameters in each
equation. Nonetheless, in theory it should still work (although I wouldn't
recommend it econometrically), and needless to say a crash is never
acceptable.
I suspect that the fix will be relatively easy (for Allin, that is ;-)
cheers
sven
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
4:50 p.m.
Am 20.12.2020 um 17:45 schrieb Brian Revell:
On Sun, 20 Dec 2020 at 15:44, Sven Schreiber <svetosch(a)gmx.net
<mailto:svetosch@gmx.net>> wrote:
Am 20.12.2020 um 16:34 schrieb Brian Revell:
> Hi Sven
> the specification was just pushing data and parameters to their
> limits with specific lags of 1 3 6 and 9 years specified in the
> 2 dependent variables to try to pick up some cyclical elements in
> the data and with only 22 observations. I will try it dropping
> the lag 9 later (I have other ARMAX results I don't want to risk
> losing at present until I have finished the session) to see if it
> makes the difference between crash or go!
> Windows 10 OS. However, i would have expected some failure
> messages, not a complete crash.
> Brian
>
Thanks, that's already enough - I can confirm the crash here on
Windows with the latest snapshot (with 22 obs and using the GUI to
enter the lag choice "1,3,6,9", without the quotes).
Although I now realize that using commas in there was probably the wrong
syntax, using spaces instead also replicates the crash...
-s
5:06 p.m.
sven
the problem in the real world is often we do not have as many observations
as we would like to work with -so the data I am looking at is annual
observations available from 1986 for 1 variable (rod catches of salmon)
with annual data for another (hours spent fishing), recorded from 1994 but
the basis of its recording changed in 1998 from a paper return to online,
which significantly affected the number of records submitted for 3 years as
for catches recorde in those years. etc so one can only work with a common
timeframe . Frustrating but reality.
cheers
Brian
On Sun, 20 Dec 2020 at 16:50, Sven Schreiber <svetosch(a)gmx.net> wrote:
Am 20.12.2020 um 17:45 schrieb Brian Revell:
On Sun, 20 Dec 2020 at 15:44, Sven Schreiber <svetosch(a)gmx.net> wrote:
> Am 20.12.2020 um 16:34 schrieb Brian Revell:
>
> Hi Sven
> the specification was just pushing data and parameters to their limits
> with specific lags of 1 3 6 and 9 years specified in the 2 dependent
> variables to try to pick up some cyclical elements in the data and with
> only 22 observations. I will try it dropping the lag 9 later (I have other
> ARMAX results I don't want to risk losing at present until I have finished
> the session) to see if it makes the difference between crash or go!
> Windows 10 OS. However, i would have expected some failure messages, not
> a complete crash.
> Brian
>
> Thanks, that's already enough - I can confirm the crash here on Windows
> with the latest snapshot (with 22 obs and using the GUI to enter the lag
> choice "1,3,6,9", without the quotes).
>
Although I now realize that using commas in there was probably the wrong
syntax, using spaces instead also replicates the crash...
-s
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
12:53 a.m.
On Sun, 20 Dec 2020, Sven Schreiber wrote:
Am 20.12.2020 um 16:34 schrieb Brian Revell:
> Hi Sven
> the specification was just pushing data and parameters to their limits
> with specific lags of 1 3 6 and 9 years specified in the 2 dependent
> variables to try to pick up some cyclical elements in the data and
> with only 22 observations. I will try it dropping the lag 9 later (I
> have other ARMAX results I don't want to risk losing at present until
> I have finished the session) to see if it makes the difference between
> crash or go!
> Windows 10 OS. However, i would have expected some failure messages,
> not a complete crash.
> Brian
>
Thanks, that's already enough - I can confirm the crash here on Windows
with the latest snapshot (with 22 obs and using the GUI to enter the lag
choice "1,3,6,9", without the quotes).
Of course, with a max lag of 9 you only have 13 effective obs left and
that is just barely enough to estimate 4*2 + 2 = 10 parameters in each
equation. Nonetheless, in theory it should still work (although I
wouldn't recommend it econometrically), and needless to say a crash is
never acceptable.
I suspect that the fix will be relatively easy (for Allin, that is ;-)
It turns out this bug was not to do with minimal degrees of freedom
(in fact, gretl will estimate a VAR with df = 0 if you really want).
It was specific to the handling of "specific lags" (with gaps) in
the GUI. I think the bug must have there for a long time but nobody
has reported it until now.
Anyway, it's now fixed in git and snapshots.
Allin
3:43 p.m.
No-I did not try to save as icon.
Brian
On Sun, 20 Dec 2020 at 13:26, Sven Schreiber <svetosch(a)gmx.net> wrote:
Am 20.12.2020 um 13:02 schrieb Artur Tarassow:
Am 20.12.20 um 12:46 schrieb Brian Revell:
Hi
A warning.
the Model Var Autoregression in the GUI can crash out Gretl -no error
message re estimation no. of variables too few or anything -it just
disappears from the screen. So all preceding work done without the outputs
saved in the session (not necessarily only from previous Var
autoregressions ) will be lost. Mine is the latest Gretl version.
Hi,
I cannot replicate this on Ubuntu 20.04 with latest git. Brian could you
provide a detailed step-by-step guide what you do? Also, which OS are you
using?
As per the changelog (http://gretl.sourceforge.net/ChangeLog.html) there
is one known crash bug currently since the latest release, quoting:
Fix bug: potential crash on "Save as icon and close" for a model displayed in
tabbed model viewer
Is this by any chance what you did / what was happening?
thanks
sven
_______________________________________________
Gretl-users mailing list -- gretl-users(a)gretlml.univpm.it
To unsubscribe send an email to gretl-users-leave(a)gretlml.univpm.it
Website:
https://gretlml.univpm.it/postorius/lists/gretl-users.gretlml.univpm.it/
1:35 p.m.
On Sun, 20 Dec 2020, Brian Revell wrote:
Hi
A warning.
the Model Var Autoregression in the GUI can crash out Gretl -no error
message re estimation no. of variables too few or anything -it just
disappears from the screen. So all preceding work done without the outputs
saved in the session (not necessarily only from previous Var
autoregressions ) will be lost. Mine is the latest Gretl version.
Thanks for the report, Brian.
Estimation of a VAR is a common task with gretl and this problem has
not been noted before, so it must appear only under very particular
conditions. Please give us details:
What dataset? What choice of variables? Lag length? Any options
selected?
Allin Cottrell
1467
days inactive
1468
days old
10 comments
4 participants
participants (4)
-
Allin Cottrell -
Artur Tarassow -
Brian Revell -
Sven Schreiber