From: Danilo on
Hi,
I have created my own transfer function "myTF.m" and I want to use it in a neural network generated by NNToolbox, i.e. with newff.
I want to use my own transfer function instead of those existing ('tansig', 'purelin', ecc.), i.e.:

>> net.layers{i}.transferFcn = 'myTF';

What I have to do in order to use 'myTF' in the same way of existing transfer functions?
Thank you so much.
From: Greg Heath on
On Jun 22, 11:30 am, "Danilo" <nature....(a)virgilio.it> wrote:
> Hi,
> I have created my own transfer function "myTF.m" and I want to use it in a neural network generated by NNToolbox, i.e. with newff.
> I want to use my own transfer function instead of those existing ('tansig', 'purelin', ecc.), i.e.:
>
> >> net.layers{i}.transferFcn = 'myTF';
>
> What I have to do in order to use 'myTF' in the same way of existing transfer functions?
> Thank you so much.

Obviously you have to define a derivative function.
Beyond that I don't know.

If you find out, please post because this may be a way to
use newff with radbas in the hidden layer and/or softmax in
the output layer.

By the way, what is the function you want to use?

Hope this helps.

Greg
From: Steven Lord on

"Greg Heath" <heath(a)alumni.brown.edu> wrote in message
news:16f32dab-c005-4f8e-b4b2-9289ace3eb09(a)18g2000vbh.googlegroups.com...
> On Jun 22, 11:30 am, "Danilo" <nature....(a)virgilio.it> wrote:
> > Hi,
> > I have created my own transfer function "myTF.m" and I want to use it in
> > a neural network generated by NNToolbox, i.e. with newff.
> > I want to use my own transfer function instead of those existing
> > ('tansig', 'purelin', ecc.), i.e.:
> >
> > >> net.layers{i}.transferFcn = 'myTF';
> >
> > What I have to do in order to use 'myTF' in the same way of existing
> > transfer functions?
> > Thank you so much.
>
> Obviously you have to define a derivative function.
> Beyond that I don't know.
>
> If you find out, please post because this may be a way to
> use newff with radbas in the hidden layer and/or softmax in
> the output layer.

http://www.mathworks.com/access/helpdesk/help/toolbox/nnet/advance4.html#17617

--
Steve Lord
slord(a)mathworks.com
comp.soft-sys.matlab (CSSM) FAQ: http://matlabwiki.mathworks.com/MATLAB_FAQ
To contact Technical Support use the Contact Us link on
http://www.mathworks.com


From: Greg Heath on
On Jun 22, 3:54 pm, "Steven Lord" <sl...(a)mathworks.com> wrote:
> "Greg Heath" <he...(a)alumni.brown.edu> wrote in message
>
> news:16f32dab-c005-4f8e-b4b2-9289ace3eb09(a)18g2000vbh.googlegroups.com...
>
>
>
>
>
> > On Jun 22, 11:30 am, "Danilo" <nature....(a)virgilio.it> wrote:
> > > Hi,
> > > I have created my own transfer function "myTF.m" and I want to use it in
> > > a neural network generated by NNToolbox, i.e. with newff.
> > > I want to use my own transfer function instead of those existing
> > > ('tansig', 'purelin', ecc.), i.e.:
>
> > > >> net.layers{i}.transferFcn = 'myTF';
>
> > > What I have to do in order to use 'myTF' in the same way of existing
> > > transfer functions?
> > > Thank you so much.
>
> > Obviously you have to define a derivative function.
> > Beyond that I don't know.
>
> > If you find out, please post because this may be a way to
> > use newff with radbas in the hidden layer and/or softmax in
> > the output layer.
>
> http://www.mathworks.com/access/helpdesk/help/toolbox/nnet/advance4.h...

That's good to know. However,

1. my MATLAB 6.5 doesn't have either nncustom nor template_transfer.
2. Neither nncustom nor template_transfer consider softmax

.... which is really weird because softmax is the primary output
function to use for classification. See the classic NN bible by
Bishop.

Hope this helps.

Greg