Use regularization with trainlm - negative performance

I train my network (2-layers) from the Neural Network toolbox. I train it with the train function trainlm and performance function mse. When I add some regularization through

 net.performParam.regularization = 0.6

I get negative performance (for example -0.02). Is it possible to use regularization with trainlm. And how is it possible to get negative performance?

[Daniel Perez Rapela]

Same here. Any suggestions?!

Answer 1

[Greg Heath]

0. SEE BELOW AND POST ANY DIFFICULTIES

1. Read the documentation and try the examples

   help trainbr
and
   doc trainbr

2. Search previous posts using

       trainbr
and
       greg trainbr

3. More example data is available using

   help nndatasets

   doc  nndatasets

 Post any problems and alert me.

Hope this helps.

Thank you for formally accepting my answer

Greg

[Daniel Perez Rapela]

Hello Greg, I have looked the everywhere (MatLab and not MatLab) and still cannot find an answer for this question. How is it possible that when I train my neuron with trainln with regularization the performance turns out negative for some cases? In these cases, best_perf and best_vperf are negative but even more strange is that if I take this already trained neuron and run a perform function with either the training or validation samples, the results are positive and obviously differ from the best_perf and best_vperf. Any thoughts? Thanks, Daniel

[Greg Heath]

What do you get with TRAINBR??

Greg