On Wed, Oct 28, 2009 at 7:30 AM, Mauro Prencipe <mauro.prencipe at unito.it> wrote:
> 1) Is there the possibility to get estimations of the errors associated to
> each refined parameter (I've experience of other environments like xlispstat
> which provide such estimations).
Instead of calling lbfgs directly, maybe take a look at the lsquares
package. You can construct an expression which is the mean-square
error, and then differentiate that or whatever (I forget what you need
to do exactly) to get error estimates.
lsquares assumes the errors are Gaussian and therefore maximum
likelihood estimates are just the least squares estimates.
I've thought about generalizing it so that the user can specify
a distribution for the errors but I haven't gotten around to it.
> 2) If each of the variables used as argument to the lbfgs call is a bfloat
> object, so are the functions employed in the fit?
lbfgs uses only ordinary floating point arithmetic.
It can't use Maxima's bigfloat arithmetic, sorry.
best,
Robert Dodier