Re: voigt error... followup

On Nov 16, 2007, at 3:09 PM, rgibson_at_email.domain.hidden
> On
> the other hand, one might expect the model to handle unphysical 
> parameters by
> simply throwing back a huge chi^2.

In general, when using subplex, you'll find it going to the far corners 
of parameter space.  This is one of the reasons it can be slow, but 
also one of the reasons it can often find that global minimum that 
other minimization routines tend not to find.  This leads to the caveat 
for subplex: always think about your parameter boundaries carefully.

The other thing that one finds for subplex is that "sensitive" codes 
that don't have internal failsafes (not every model function will have 
a "valid" parameter range well-defined by an N-dimensional box) will 
have the biggest problems with subplex.  I have crashed (as in segv'd 
right out of ISIS completely) many a badly written local model with 
subplex.

As John suggested, the try/catch mechanism available in S-lang 2 could 
be helpful in some of these situations.

This is something that's also come up repeatedly in the Sherpa 
redesign, especially when discussing the implementation of a "guess" 
function.  I.e., should the program try to automatically narrow down 
parameter space for you, and save you from yourself?  Our usual retort 
has been the "guess" function already exists - it's called having a 
PhD.

-Mike

----
You received this message because you are
subscribed to the isis-users list.
To unsubscribe, send a message to
isis-users-request_at_email.domain.hiddenwith the first line of the message as:
unsubscribe