Re: st: mata optimize problem

[Stas Kolenikov <skolenik@gmail.com>]

It can go either way. You might want to output some intermediate
results to see what your evaluator really returns. If it only returns
missing values or a constant value, this would explain the optimizer
error message.

On Sat, Jul 3, 2010 at 9:11 PM, Michael Ralph M. Abrigo
<mmabrigo@gmail.com> wrote:
> Many thanks Professors Austin and Stas. Points well taken. I already
> have found the bug, thanks to your help, and did the necessary
> tweaking. I was planning to set the externals as parameters later
> after I set it running, but as Prof Stas as pointed out I should set
> them as parameters outright. I was able to run it (yey!), but now I
> was prompted
>
> could not calculate numerical derivatives -- discontinuous region with missing
> values encountered
>
> I don't now whether this is what Lambert, Brown and Florax's (2010)
> are saying that numerical solutions using various search algorithms
> were difficult to obtain, or because of my codes. In any case thank
> you very much. This is my very first code in Mata.
>
> Cheers,
> Michael
>
> On Sun, Jul 4, 2010 at 2:08 AM, Stas Kolenikov <skolenik@gmail.com> wrote:
>> set matalnum on
>>
>> to see which line of your procedure generates the error. To make your
>> program well defended against bad inputs, you can code something like
>>
>> if ( cols(A_inv) != rows( X ) ) {
>>  printf( "{err}A_inv is not conformant to X"\n)
>>  lndensity[,] = J( rows(lndensity), 1, . )
>>  return
>> }
>>
>> Note also that your observations lie on the straight line x==y. A more
>> sensible set of values might have been
>>
>> gen xcoor = uniform()
>> gen ycoor = uniform()
>>
>> producing points distribution U[0,1]^2.
>>
>> From programming style, I would try to get rid of the -external-
>> matrices in your routine. Any -external- statement read like "I don't
>> bother explaining to you what's going on here, and what these matrices
>> are", which makes it more difficult to debug. You can transfer them as
>> parameters of -optimize()-. And of course transferring a known matrix
>> I is kinda silly, you can generate it within the routine.
>>
>> On Sat, Jul 3, 2010 at 8:44 AM, Austin Nichols <austinnichols@gmail.com> wrote:
>>> Michael Ralph M. Abrigo <mmabrigo@gmail.com> :
>>>
>>> I have not read through this, but you should put the dimensions of
>>> every matrix in as a comment, and also print out the matrices once,
>>> and then maybe you will see where the error is.
>>>
>>> On Sat, Jul 3, 2010 at 2:08 AM, Michael Ralph M. Abrigo
>>> <mmabrigo@gmail.com> wrote:
>>>> Good day statalisters!
>>>> I am practically new in programming, especially in Mata. I'm trying to
>>>> implement Lambert, Brown and Florax's (2010) poisson spatial lag model
>>>> ("A two-step estimator for a spatial lag model for counts: Theory,
>>>> small sample performance and an application", Urban and Regional
>>>> Economics) in Stata using full-information ML (instead of the two-step
>>>> LIML). However, I got stuck because of a conformability error which I
>>>> cannot trace (I was able to generate the matrices interactively using
>>>> ad hoc values). I was hoping I can do the FIML first before I venture
>>>> to their two-step estimator.
>>>> Any hint will be greatly appreciated.
>>>> Below is my code and the error messages.
>>>> Cheers,
>>>> Michael
>>>> drop _all
>>>> ** (1) Set version; Open file
>>>> version 11.0
>>>> webuse "dollhill3.dta", clear
>>>> ** (2) Generate constant; Local variables
>>>> gen cons = 1
>>>> local y deaths
>>>> local xlist smokes pyears cons
>>>> ** (3) Generate weight matrix
>>>> sort pyears
>>>> gen xcoor = _n
>>>> gen ycoor = _n
>>>> spwmatrix gecon xcoor ycoor, wname(W) wtype(bin) ///
>>>>    dband(0 1000) rowstand mata replace
>>>> ** (4) Optimize using Mata
>>>> mata
>>>>   void spmlpoisson (
>>>>     real scalar todo,
>>>> real rowvector rho_b,
>>>> real colvector lndensity,
>>>> real matrix g,
>>>> real matrix H)
>>>> {
>>>>       external y, X, W, I
>>>>  transmorphic A_inv
>>>>  real colvector Xb_sar, mu_sar
>>>>  real rowvector b
>>>>  real scalar rho
>>>>       rho_b = rho, b
>>>> A_inv = cholinv(I - rho * W)
>>>> Xb_sar = A_inv * X * b'
>>>>         mu_sar = exp(Xb_sar)
>>>> lndensity = (y :* Xb_sar) - mu_sar - lnfactorial(y)
>>>> }
>>>>   st_view(y = ., ., "`y'")
>>>>   st_view(X = ., ., tokens("`xlist'"))
>>>>   fh = fopen("W", "r")
>>>>     W = fgetmatrix(fh)
>>>>    fclose(fh)
>>>>   I = I(cols(W))
>>>>   S = optimize_init()
>>>>   optimize_init_evaluator(S, &spmlpoisson())
>>>>   optimize_init_which(S, "max")
>>>>   optimize_init_evaluatortype(S, "d0")
>>>>   optimize_init_params(S, J(1, cols(X) + 1, 0))
>>>>   rho_b = optimize(S)
>>>>                        *:  3200  conformability error
>>>>            spmlpoisson():     -  function returned error
>>>>       opt__calluser0_d():     -  function returned error
>>>>       opt__d0_calluser():     -  function returned error
>>>>     deriv__call1user_d():     -  function returned error
>>>>  _deriv__compute_value():     -  function returned error
>>>>                 _deriv():     -  function returned error
>>>>        opt__eval_nr_d0():     -  function returned error
>>>>              opt__eval():     -  function returned error
>>>> opt__looputil_iter0_common():     -  function returned error
>>>> opt__looputil_iter0_nr():     -  function returned error
>>>>           opt__loop_nr():     -  function returned error
>>>>               optimize():     -  function returned error
>>>>                  <istmt>:     -  function returned error
>>>> r(3200);
>>>>
>>>>
>>>> end
>>>
>>> *
>>> *   For searches and help try:
>>> *   http://www.stata.com/help.cgi?search
>>> *   http://www.stata.com/support/statalist/faq
>>> *   http://www.ats.ucla.edu/stat/stata/
>>>
>>
>>
>>
>> --
>> Stas Kolenikov, also found at http://stas.kolenikov.name
>> Small print: I use this email account for mailing lists only.
>>
>> *
>> *   For searches and help try:
>> *   http://www.stata.com/help.cgi?search
>> *   http://www.stata.com/support/statalist/faq
>> *   http://www.ats.ucla.edu/stat/stata/
>>
>
>
>
> --
> "I am most anxious for liberties for our country... but I place as a
> prior condition the education of the people so that our country may
> have an individuality of its own and make itself worthy of
> liberties... " Jose Rizal,1896
>
> *
> *   For searches and help try:
> *   http://www.stata.com/help.cgi?search
> *   http://www.stata.com/support/statalist/faq
> *   http://www.ats.ucla.edu/stat/stata/
>



-- 
Stas Kolenikov, also found at http://stas.kolenikov.name
Small print: I use this email account for mailing lists only.

*
*   For searches and help try:
*   http://www.stata.com/help.cgi?search
*   http://www.stata.com/support/statalist/faq
*   http://www.ats.ucla.edu/stat/stata/