I'm doing my assigment to extract rule using neurorule algorithm (basically neural network) which is explained in a journal entitled "NeuroRule: A Connectionist Approach to Data Mining".
It has three major steps
1. Network training
2. Network Pruning
3. Rule extracting
I'm still in the first step which is a kind of backpropagation. It is explained that, there are two activation function which is used. Log sigmoid and hyperbolic tangent sigmoid. I get mad with them. They produced a value that isn't really clear to be 0 or 1 so I search about activation function more and find that applying those needs derivative activation function.
But inside the journal, there's no explaination about that. Should I add it or not? and is there any explicit impact if I add it?
Please for the help.
Thankyou.
It has three major steps
1. Network training
2. Network Pruning
3. Rule extracting
I'm still in the first step which is a kind of backpropagation. It is explained that, there are two activation function which is used. Log sigmoid and hyperbolic tangent sigmoid. I get mad with them. They produced a value that isn't really clear to be 0 or 1 so I search about activation function more and find that applying those needs derivative activation function.
But inside the journal, there's no explaination about that. Should I add it or not? and is there any explicit impact if I add it?
Please for the help.
Thankyou.