BACK PROPAGATION WITH 2 HIDDEN LAYERS

I need to train a neural network with 2 hidden layers.please post the matlab code for 2 hidden layers. In neural network,nntool i can only change the no. of hidden nodes how can i change the no. of hidden layers.

[Greg Heath]

A single hidden layer is sufficient. Why do you think you need more???

[NAYANA KAMAL]

YES, i need the code for 2 hidden layers.please help me.

[NAYANA KAMAL]

My guide told me to use 2 hidden layers to reduce the mean square error

Answer 1

[Greg Heath]

My recommendation is to FIRST use one hidden layer and try to minimize the number of hidden nodes while achieving an adjusted R-square >= 0.99. For more advice, I need more information re number and dimensions of input/target examples as well as an explanation of what you are trying to model.

I have posted tens of examples. Search on

 greg fitnet      % for regression/curve-fitting

 greg patternnet  % for classification/pattern-recognition

Once a single hidden layer solution is found, you can find a double hidden layer solution by changing fitnet(H) to fitnet([H1,H2]). I explain how to choose H. You are on your own with [H1,H2] except I have heard that, typically, H1+H2 < H. However, I don't believe a proof exists.

Hope this helps.

Thank you for formally accepting my answer

Greg

[NAYANA KAMAL]

thank you for your reply.I am doing a rainfall estimation model using the cloud temperature and rainfall data,so i have 80 input values(temperature) and 80 target values(rainfall ).when i use 1 hidden layer the MSE is very high, so i need to test with 2 hidden layers.

[Greg Heath]

Again, 1 hidden layer is sufficient for minimizing MSE. Using more hidden layers does not guarantee a better result. You just need to increase the number of hidden nodes, H, AND try multiple (e.g., 10) cases of initial weights for each value of H.

See my examples. Search using

 greg fitnet Ntrials