Manual Neural Network Wizard

Hi all,

I've tried looking around to figure this one out, but time is against me so here I am.

Basically, I've developed a really good network with the help of the NN wizard. Great. However, due to a complication, I need to essentially do the same network, but with different transfer functions (Hardlimit).

I've tried developing the network from scratch with code, but I just can't get it to work, and I've tried the nntool command too, but I can't seem to clarify the test/val/train split, or plot a good roc or confusion matrix.

Essentially, the wizard is great, but I need simpler transfer functions to code elsewhere.

Thanks all,

C

[Greg Heath]

Insufficient detail.

However, I suspect the reason that training failed is because hardlim is not differentiable.

I don't think anyone is going to try to help you unless you describe your problem in much more detail.

For example, why do you believe you have to have hardlim as an output transfer function? It makes no sense to me.

You don't explain

 1. Type of problem 
2. Dimensions of input and output vectors
3. Amount of data
3. Type of net
4. Number of hidden nodes
5. Type of transfer functions
6. Non-default parameter choices
7. Normalized quantitative results e.g., 

   NMSE = mse(error)/mean(var(target',1))

etc