[Developers] MCMC not working in ADMB10

[Johnoel Ancheta]

Thanks Dave,

I'll add valgrind to the automated testing (buildbot).

JOhnoel

On 11/19/10 7:07 AM, dave fournier wrote:
> The old better_rand function saves its state in static variables.
> this is the old C approach. the problem is that you can't have two 
> different
> random number generators using this code. the general approach is to take
> all the static variables and put them in a class. then different 
> instances of the
> class are different random generators with distinct states.
>
> For people adding code to ADMB there is the follwong rule of thumb. 
> Whenever
> you use new in a constructor you need to write a destructor with delete.
>
> If you had run the example with valgrind it would have picked this up 
> immediately.
>
>
> ==10191== 1,137,337,344 bytes in 227,832 blocks are definitely lost in 
> loss record 16 of 16
> ==10191==    at 0x4C27939: operator new[](unsigned long) 
> (vg_replace_malloc.c:305)
> ==10191==    by 0x5464B6: 
> random_number_generator::random_number_generator(int) (rngen.cpp:70)
> ==10191==    by 0x463D1A: better_rand(long&) (bet_rand.cpp:147)
> ==10191==    by 0x4772C7: bounded_multivariate_normal(int, dvector 
> const&, dvector const&, dmatrix&, long&, double const&) (monte.cpp:81)
> ==10191==    by 0x45D655: function_minimizer::mcmc_routine(int, int, 
> double, int) (xxmcmc.cpp:703)
>
>
> destructor should be something like
>
> random_number_generator::~random_number_generator()
> {
>   delete [] mt;
>   mt=0;  // not  necessary but habit.
> }
>
>
> _______________________________________________
> Developers mailing list
> Developers at admb-project.org
> http://lists.admb-project.org/mailman/listinfo/developers