[Date Index] [Thread Index] [Author Index]

Re: Catching messages

• To: mathgroup at smc.vnet.net
• Subject: [mg91620] Re: Catching messages
• From: thorgal <pbruynd at vub.ac.be>
• Date: Fri, 29 Aug 2008 04:24:59 -0400 (EDT)
• References: <g90bg1$n91$1@smc.vnet.net> <g95k03$4tr$1@smc.vnet.net>

On Aug 28, 9:26 am, David Bailey <dave at Remove_Thisdbailey.co.uk>
wrote:
> thorgalwrote:
> > Is there an easy way to immediately abort a function that was called
> > when it generates a given message and return to the part of the
> > program that called this function ?  I want to do this when I call
> > some function from Mathematica packages, i.e. i don't want to change
> > to function code in the packages and include e.g. catch-throw
> > sequences.
>
> I don't think your request is very meaningful as stated. When you call a
> function - even some built-in ones - that function may make several
> sub-calls before the diagnostic is generated. Falling back one level
> (which is more or less what happens anyway after the diagnostic has been
> generated) is not really useful, and tends to result in a cascade of
> errors.
>
> Why not tell us a bit more about what you are trying to achieve. For
> example, if you are trying to do something like a minimisation (where
> the package contains the minimation code), and the function can
> misbehave for some values, wrap the function in another which catches
> the errors and returns a suitable numeric value in these cases.
>
> David Baileyhttp://www.dbaileyconsultancy.co.uk

David,

It is indeed something like you suggested.  I am using the Wolfram
Neural network package.  Sometimes it happens that the algorithm has
problems, converges slowly or other warnings.  Usually it is
sufficient to stop the training, re-initialize the neural network and
train again.  Now I do this manually and I would like to find a way
that the neural network training function I call from my program would
return to my program when a given warning message is issued, i.e so
not just falling back one level, but falling back to the level where
my program called the training function , so that the program can re-
initialize the network and try again.

I haven't found any Mathematica command that can force a program to
return to a given level when an error/warning occurs at a deeper
level.