Re: Neural Networks-trainlm wont work


Felix wrote:
> Hello,
> I would like to use the levenberg-marquart backprop to train my ffd
> net. However invoking train with 'trainlm' causes:
>
> ??? Maximum variable size allowed by the program is exceeded.
>
> Training set ist not huge-3000*170 values.
> Can anyone give me a hint?

help trainlm

There you will see the default spec

net.trainParam.mem_reduc 1 Factor to use for memory/speed trade
off.

By increasing this above 1 you will decrease memory requirements
at the cost of slower training speed.

If the resulting speed is too slow try one of the Conjugate Gradient
algorithms.

Hope this helps.

Greg
Hope this helps

.