Use a custom activation function in neural network toolbox? [Solved]

Hi everyone,

I've seen that it is possible to create our own transfer function here : https://www.mathworks.com/help/nnet/ug/custom-neural-network-helper-functions.html

So I did what is said :

1) edit hardlim,

2) copy paste in another 'function' file, change the name "mytransfer", save

3) put it in my working directory

% I'm just willing to have a hardlim function with output values between [-1, 1]

Of course, I'm able to compute my function in another file like that :

x = -3:0.1:3;
y = mytransfer(x);

But I can't use it in my neural network :

net=newff([-1 1; -1 1], [2 1], {'mytransfer','mytransfer'}, 'trainlm');% code

it gives me an error, looks like I should create that name ? how to do so ?

Error using feval
Undefined function or variable 'mytransfer.name'.

Error in nnModuleInfo (line 7)
info.name = feval([f '.name']);

Error in nnet7.transfer_fcn (line 6)
info = nnModuleInfo(fcn);

Thanks for helping :)

[EDIT]

Ok I found a more precise answer to my question here, the documentation is a bit fuzzy :

https://www.mathworks.com/matlabcentral/answers/56137-how-to-use-a-custom-transfer-function-in-neural-net-training

So I did what Bob said :

I invented a transfer called "bobfer". To implement it, go into .../toolbox/nnet/nnet/nntransfer.

Copy tansig.m and the folder +tansig to a directory on your path.

Give each a new name, such as bobfer.m and +bobfer.

Modify bobfer.m at the last line so it calls the bobfer.apply method:

a = bobfer.apply(n);

Now, in +bobfer, modify apply.m. This is pretty easy; it's only a few lines. This is where your transfer function goes.

Let me know if this helps...

Bob

[Christoph Aistleitner]

Works also in R2021a. For simulink-implementation you need to create a new transferfunction block.

[Christoph Aistleitner]

In order to create a new transferfunction block, edit the simulink library. Copy paste a existing activation function and break the links. Then replace the logic and rename the block.