27 Jun
2002
27 Jun
'02
9:34 a.m.
Thanks Marc. For what it is worth, we have a new front end under
development. We hope to make it as reliable as the back end that does the
hard work, but we'll let you know!
Thanks,
Gary
: Gary King, King(a)Harvard.Edu http://GKing.Harvard.Edu :
: Center for Basic Research Direct (617) 495-2027 :
: in the Social Sciences Assistant (617) 495-9271 :
: 34 Kirkland Street, Rm. 2 HU-MIT DC (617) 495-4734 :
: Harvard U, Cambridge, MA 02138 eFax (928) 832-7022 :
On Thu, 27 Jun 2002, Marc R. Rosenblum wrote:
Thanks so much to both of you for your prompt
responses. (Customer service
is yet another reason Amelia is taking over the multiple imputation market ;-)
I'll give you (and the list-serv) the play by play in case it is helpful to
anyone
Initially I had left all the default settings along on the first options
page except for telling amelia which vars. were fully observed.
First mistake I solved was that although I was using TSCS data and telling
Amelia my AMt's and AMc's, I did not change the default settings on AMusets
and AMusecs. (I was trying to ask Amelia to do as little as possible, but I
think giving it TSCS data but not changing these settings was one source of
a crash.)
Then I could get Amelia to run on an extremely abridged version of my data
(down to 8 variables from an original 30 or so) by re-setting these to 1
(should have been more obvious to me than it was!)
When I added variables back in I was now getting one of the following messages:
C:\GAUSS\SRC\AMELIA.SRC(395) : error G0036 : Matrices are not conformable
Currently active call: AMELIA [395]
C:\GAUSS\SRC\SORTMC.SRC(89) : error G0057 : Procedure stack overflow -
expression too complex
Currently active call: SORTMC [89]
C:\GAUSS\SRC\SWEEP.SRC(102) : error G0048 : Matrix singular
Currently active call: SWEEP [102]
C:\GAUSS\SRC\SORTMC.SRC(69) : error G0030 : Insufficient workspace memory
Currently active call: SORTMC [69]
(Even though I have over 400 Mb of available memory, and had set max
workspace in the GSRUN.cfg file to 300.)
Then I followed the advice on the FAQs for speeding Amelia up:
1) I set _AMempri = 2
2) I set _AMmthd=2
3) I set _AMdrTol = .0001
SUCCESS!
I think my real problem was probably correlation among some of my
variables, so today I will re-run leaving everything alone except the
_AMempr option and see if it still works
Thanks again for your help and for making the software available.
Marc Rosenblum
Assistant Professor of Political Science
University of New Orleans
At 02:11 AM 6/27/2002, James Honaker wrote:
-
amelia mailing list served by Harvard-MIT Data Center
List Address: amelia(a)latte.harvard.edu
Subscribe/Unsubscribe: http://lists.hmdc.harvard.edu/?info=amelia
Dear Mr. Rosenblum,
Unfortunately, _AMvarnm can not be set in the windows version of Amelia,
or the solution that Gary gives, and that from your note you probably
have already looked to try, would be the short cut around this problem.
I have not been able to replicate your problem, but have been working on
it. Are you setting other globals at the same time? Could you give me a
list of what gloabls in Amelia you are setting?
Regardless of which combinations of variables is causing this problem, the
solution is fairly simple.
If you are curious, this is what is causing your problem:
The problem you have is one that has come up before. Because of the
interface we use for the windows version, it is not possible to produce
error messages during the actual running of Amelia as they occur.
Threfore all the checks that occur during the running of Amelia have to be
moved into the front end and run before any processing in Amelia can
occur. Sometimes sets of globals (both the ones the user can set, the
internal ones used for reordering and scaling the dataset for
computational efficiency) are valid in combination when they are not
individually. Once the computation has begun, all the globals are set, so
it is easier to check they are jointly valid, whereas at the point of the
front end, the code sometimes prematurely challenges the validity of
globals because not all user globals have been set yet, or internal
globals have not yet had a chance to be changed. The routines that check
the validity of globals were written to help the user from making common
mistakes, but sometimes they trip up (as here) and prevent the user from
doing valid things simply because the routines that check don't see the
full picture yet.
I hope to give you a new version of Amelia for windows with the check
routines corrected on Thrusday,
sincerely,
James Honaker
On Wed, 26 Jun 2002, Gary King wrote:
-
amelia mailing list served by Harvard-MIT Data Center
List Address: amelia(a)latte.harvard.edu
Subscribe/Unsubscribe: http://lists.hmdc.harvard.edu/?info=amelia
why don't you set _AMvarnm to missing values (a dot). then Amelia will
name the variables for you and this error shouldn't occur.
THe other problem I think is with the operating system.
Gary King
: Gary King, King(a)Harvard.Edu http://GKing.Harvard.Edu :
: Center for Basic Research Direct (617) 495-2027 :
: in the Social Sciences Assistant (617) 495-9271 :
: 34 Kirkland Street, Rm. 2 HU-MIT DC (617) 495-4734 :
: Harvard U, Cambridge, MA 02138 eFax (928) 832-7022 :
On Tue, 25 Jun 2002 mrosenbl(a)uno.edu wrote:
> I am getting the following response when I try to run Amelia for windows:
>
>
> _AMvarnm has too many variable names to match each variable
>
> in the dataset. Delete variable names, or set _AMvarnm=.
>
> for default variable names (default = var1 var2 ... ).
>
> I have cut the dataset down to 20 variables, 8 of which are fully
observed,
and
> none of which is missing more than 85 or so
out of N=586. It is TCSC
data w/ 15
> countries, 39 years.
>
> Any suggestions?
>
> I previously ran into a separate problem on a different computer of
Amelia
> crashing after saying that it was unable to
save window size and font
size.
> there was a message about this on the
listserv, but I never did solve
that
> problem (do not know yet whether it will pop
up now that I am back on
my own
> > computer; was using a colleague's before.)
> >
> > Thanks so much for any insight.
> >
> > Marc Rosenblum
> > Assistant Professor of Political Science
> > University of New
> > Orleans
> >
>
>
> -
> amelia mailing list served by Harvard-MIT Data Center
> List Address: amelia(a)latte.harvard.edu
> Subscribe/Unsubscribe: http://lists.hmdc.harvard.edu/?info=amelia
>