testcases and pet projects
by Klein, Christoph
Hi,
I've attached a badly specified mle model which provokes the error. I don't know whether one can find a well specified model which has these problems, too. The script is named test_bfgs.inp.
I've attached two more patches. The first one adds a function debugger to gretl. It adds 4 new commands to gretl: debug and undebug for initiating and stopping debugging of a user defined function and continue and next (aliases c and n) which allow to step through a function. After having these problems with my code last Thursday/Friday I thought that a real debugger would be really helpful. If you have a large programm set debug 2 is not sufficient anymore. The debugger is not thoroughly tested. There are two things that I don't like about it. First of all a callback into the client application is needed in order to talk with the users out of the depths of libgretl. Perhaps I can extend the PRN structure somehow in order to have a more coherent interface. Another problem here is that I didn't find a way to transport the callback address to exec_user_func. I extended libset, but I think libset was meant only to care for user editable properties. I would be happy to receive comments about this from you. If I have time in the next days I will try to implement breakpoints and "step into function"-commands. Up to now the debugging feature is only available to gretlcli users. Below you find a sample debug session:
script:
set debug 2
function fred(scalar a)
scalar d=a+1
scalar f=1
scalar g=f
return scalar d
end function
debug fred
scalar q=fred(2)
scalar q=fred(2)
session:
? run test.inp
/home/chris/dev/gretl-cvs8/test.inp
? set debug 2
? function fred(scalar a)
#scalar q=fred(2)
#print q
? debug fred
? scalar q=fred(2)
*** executing function fred
? scalar d=a+1
Generated scalar d = 3
Debugging fred, line 0
? d
3
? d=4
Replaced scalar d = 4
? n
? scalar f=1
Generated scalar f = 1
Debugging fred, line 1
? n
? scalar g=f
Generated scalar g = 1
Debugging fred, line 2
? c
Generated scalar q = 4
? scalar q=fred(2)
*** executing function fred
? scalar d=a+1
Generated scalar d = 3
Debugging fred, line 0
? c
? scalar f=1
Generated scalar f = 1
? scalar g=f
Generated scalar g = 1
Replaced scalar q = 3
Done
The second patches is more like a proof of concept. I really liked the integration of R into gretl via the foreign code block. But up to now it has 2 problems: It is slow and for all foreign blocks you need gretl glue code which transports parameters and results into and out of R. The second patch speeds up the execution of R functions by calling directly into libr. I have extended the makefiles in order to link against libR. The position of libr is hand coded in order to match the location on my ubuntu system. Another feature of this patch is the seamless integration of R functions into gretl. After defining a function in a gretl foreign-block, it can be called from within gretl using standard genr statements: No more glue needed. But this functionality is still a bit experimental. I haven't tested this much too. Another problem is that using libr may have some side-effects. For example R installs its own segfault-handler. This feature, too, is only avaiable to gretlcli users. Below you find again an example:
script:
nulldata 10
foreign language=RLib
abc<-function(q) {
z=q+1
invisible(z)
}
end foreign
scalar b=abc(2)
session:
? nulldata 10
periodicity: 1, maxobs: 10,
observations range: 1-10
? foreign language=RLib
? abc<-function(q) {
? z=q+1
? invisible(z)
? }
? end foreign
> abc<-function(q) {
+ z=q+1
+ invisible(z)
+ }
? scalar b=abc(2)
abc(2)
Generated scalar b = 3
Done
Christoph
________________________________________
Von: gretl-devel-bounces(a)lists.wfu.edu [gretl-devel-bounces(a)lists.wfu.edu] im Auftrag von Allin Cottrell [cottrell(a)wfu.edu]
Gesendet: Samstag, 4. Juli 2009 21:21
An: Gretl development
Betreff: Re: AW: AW: [Gretl-devel] bfgs: failed to match initial value
On Sat, 4 Jul 2009, Sven Schreiber wrote:
> Am 03.07.2009 22:35, Klein, Christoph schrieb:
> > I'm sorry for all the trouble. Looking carefully through my code, it seems more like the problem is within my scripts than in gretl. I will try to find out why this problem did arise in the next days.
> >
> > Sorry again and thanks for your helping pointers!
> >
> > Christoph
>
> Nevertheless, if it was the case that you get an error message "which
> should never happen" then it's worthwhile to fix it, no?
Yes, I think it would still be useful to have a testcase which
provokes that error message.
Allin.
_______________________________________________
Gretl-devel mailing list
Gretl-devel(a)lists.wfu.edu
http://lists.wfu.edu/mailman/listinfo/gretl-devel
12 years, 10 months
crash on defining matrix?
by Allin Cottrell
I'm just about ready to put out gretl 1.8.2, but there's one
outstanding item of business. Ignacio reported to me that he was
getting a crash from the following script:
<script>
nulldata 115
matrix trendm=t-1
</script>
I've tried at some length but I'm not able to replicate the
problem. I'm not saying the problem doesn't exist. Ignacio gave
me a different hard-to-replicate problem not so long ago and
eventually I found a definite, if elusive, bug.
I'm assuming that the problem ocurred while using the GUI program
in Spanish. In my attempts at replication I've used both gretlcli
and gretl_x11, in both English and Spanish.
Can anyone else replicate the crash? If so, could you please try
it under gdb and/or valgrind (with gretl compiled with debugging
symbols) and send me the trace?
Thanks very much,
Allin.
12 years, 10 months
bfgs: failed to match initial value
by Klein, Christoph
Hi,
I had a problem with a mle estimation today. My script canceled execution with the following error message:
Using numerical derivatives
Iteration 1: Log-likelihood = 7274,73993927
Parameters: 3,4478 -0,17644 6,7838
Gradients: 2,6526e+06 2,6526e+06 -2,6526e+06
failed to match initial value of objective function, 7274,74
--- FINAL VALUES:
Log-likelihood = 7274,73993926 (steplength = 1,04858e-14)
Parameters: 3,4478 -0,17644 6,7838
Gradients: 2,6526e+06 2,6526e+06 -2,6526e+06
Tolerance = 1,81899e-12
The convergence criterion was not met
seems like the starting set was already quite optimal. Though gretl steps to the new point, because the criteria for advancing in line 544 in gretl_bfgs.c is (f >= fmax + d). d is set as d = sumgrad * steplen * acctol and in my case seems to be negative. The criteria for printing the error message (line 645) is: (fmax < f0). The attached patch (one-liner) changes this to ((fmax + d) <= f0). Is this change ok?
Christoph
12 years, 10 months
NA patch for fracdiff functions
by Klein, Christoph
Hi,
the current behaviour of fracdiff when invoked with a series containing NAs is to issue an E_DATA error. I propose that in this case fracdiff returns a series of NAs. I think returning NAs would be more in line with the rest of gretl, but I'm not sure.
The attached patch changes the fracdiff function in the proposed way.
Christoph
12 years, 10 months
