R help - Apr 2012 - nnet formular for reproduce the expect output

Dear All,

I am recently working on neural network using nnet package. The network has
4 hidden layers and 1 output layer, the target output 1 or 0.

 The model I use is as follows:
nn<-nnet(target~f1+f2+f3+f4+f5+f6+f7+f8+f9+f10,data=train,size=4,linout=FALSE,decay=0.025,maxit=800)

 It works well and give me ROC 0.85. However, when I want to reproduce the
result in java, I cannot get the same prediction as nnet output. 

I checked modern applied statistics with S and I think the output is as
follows: y=sigmoid(a_0+\sum (w_{hk} sigmoid_h(a_h+\sum (w_{ih}*x_i)))) and
this is exactly how I reproduce the result, the activation function is
logistic function for hidden layer and output layer.

 I use the exact weights I get from the model. 

Will anyone be kind of pointing out where thing goes wrong?

Thanks
Eric

--
View this message in context:
http://r.789695.n4.nabble.com/nnet-formular-for-reproduce-the-expect-output-tp4590348p4590348.html
Sent from the R help mailing list archive at Nabble.com.
	[[alternative HTML version deleted]]