Train and simulation of neural network

HI.... I have following questions?

1-I train my NN in nntool and get the weights and bias values. Then using the nntool i simulates some values and get very close results up 95%. But when I use these values and develop the same network with activation functions on my excel sheet i dont get the same simulated result.

2-How can I introduce 2 hidden layers in nntool?

3-I have this freedom in newff...but in some literature it mention that newff is old one use nntool,please help in this matter?

Best Regards

Answer 1

[Greg Heath]

Yes, NEWFF and it's special cases NEWFIT and NEWPR have been obsolete since 2010b. However, they still can be used.

It would be preferable, if you have them, to use the current functions (Special cases FITNET and PATTERNNET that call FEEDFORWARDNET)

Details on use can be obtained using the HELP and DOC commands.

Typically, the inability to match computer results, given the weights, is because the default MINMAX NORMALIZATION of the input and target AND MINMAX DENORMALIZATION of the output are not taken into account.

You can probably find examples by searching the NEWSGROUP and ANSWERS including MINMAX and/or NORMALIZATION and/or DENORMALIZATION as search words. Also consider using

 help minmax
 doc minmax

Hope this helps.

Thank you for formally accepting my answer

Greg