Hi John,
Let me answer to some of your interesting comments below:
- However, I have to admit that I tried a couple of examples (ie the square root that is already in main) with more training data and various training parameters but I couldn't get it to produce any good results :\
> You are right, it is not easy to make the training converge. So far, it is a very tricky step but I think it is part of the game since, as far as I know, any optimization problem is tricky to solve. But, it doesn't mean that we cannot improve the situation ;)
- Is it planned to be a library that one can use in his own code or a standalone application? If it is going to be a standalone application then I guess it needs to somehow be able to read the input data and the neural network's structure and parameters from a file. Right?
> At the moment, I am implementing a scripting language (which should be completed today, so it should be released VERY soon) so I personally see Gneural Network as a standalone package which can be utilized by people who can't program but can understand simple scripts. But more advanced user could comment the parser out in the code and call the various methods implemented in their own way along with their won routines.
- Currently it only accepts as input numbers as inputs, but I believe it would be good to accept other types of inputs like for example strings. ...
> This is another VERY GOOD point! At the moment, Gneural Network can deal with scalars only but soon there will be a way to generalize this situation. One thing at a time ;)
I hope this somehow answers to your questions!
JM