Re: bfgsmin iterations

[Levente Torok]

On Wednesday 13 May 2009, you wrote:
> 
>    // this is cache on feval to prevent the extra evaluation of the
>    // objective function when subsequent calls with exactly the same 
> arguments would take place
> 
> I think we have to be cautious with modifying feval to cache the results, 
> because this
> makes two assumptions:
> 
> - the function is 'functional' i.e. should return the same result given the 
> same argument
>   This is not true e.g. for stochastical functions (monte carlo, etc).
> 
> - the function has no side effects, i.e. doesn't change some global state.
> 

You are correct from functional point of view.
However, due to the fact that fmins is intendeded for minimizing a function 
over 
the parameter space, which inherently assumes that the objective function is 
stateless.


For the case of stochastic (MC) stuff, I think the same assumption applies, too.

Levente

-- 
Blogger of http://fapuma.blogspot.com