El Cid,

I don't understand what you want to try to achieve:
- to replicate Better's EA ?
- to do your own NN EA?
- to do something completely different

For the moment I guess it is the last one because from the third post there is no link to NN anymore.

What we know from Better's EA is that inputs are based on Moving Averages and that it uses a Probabilistic Neural Network. And that's it.
I think that there is no point to try to replicate the EA because it will be anyway impossible as it needs to be trained the same way.

If you want to do your own or a Forex TSD NN EA , that is much more interesting because the only thing that exists currently is just a Perceptron that was taken from the metaquotes site and I think nobody has bring it further as far as I know.

To use a Probabilistic NN would be very interesting.
But then you cannot put rules such as if .. then
You need to define your inputs (moving averages or something else like OHLC...) outputs (for instance probability for a Buy / Sell) the data you request for the learning process, the data for the testing process and to have a PNN software that you can train and from which you can "import" the training results in an EA.
I guess that at the beginning the first issue to tackle is to find or program a PNN software.