On Wed, Feb 27, 2008 at 7:08 PM, Robert Gloeckner
<Robert.Gloeckner at gmx.net> wrote:
> WRITE-CHAR on #<CLOSED OUTPUT BUFFERED FILE-STREAM CHARACTER
> #P"/dev/fd/1"> is illegal
Well, the immediate problem here is that an error message
was generated from within lbfgs (in the code translated automatically
from Fortran into Lisp) and Clisp, for reasons known only to itself,
refuses to print the message. (This is Maxima + Clisp here, isn't it?)
Try the same operation with GCL or SBCL -- I think those will
display the error message. I don't know what's wrong with Clisp.
I've seen that error before but didn't resolve it.
The larger problem here is that you have a pretty difficult regression
problem. I think you'll need to identify an approximate solution
somehow and then start lbfgs from there; as a quasi-Newton method,
lbfgs is a local search method. I tried guessing, but
everything I tried yielded an error at the starting point approximately
equal to 1.215 x 10^7; which suggests I was very far from any
solution.
Sorry I can't be more helpful,
Robert Dodier