my neural network is giving same output for all inputs...do you have any idea why?
5 views (last 30 days)
Show older comments
net=network(8,3,[1;1;1],[1 1 1 1 1 1 1 1;0 0 0 0 0 0 0 0;0 0 0 0 0 0 0 0],[0 0 0;1 0 0;0 1 0],[0 0 1]); net.layers{1}.transferFcn='logsig'; net.layers{2}.transferFcn='logsig'; net.layers{3}.transferFcn='logsig'; net.layers{2}.dimensions=10; net.trainFcn='traingd'; net.trainparam.min_grad=0.00001; net.trainparam.epochs=10000; net.trainparam.lr=0.3; net.trainparam.goal=0.0001; net=init(net); net.layers{1}.initFcn='initwb'; net.layers{2}.initFcn='initwb'; net.biases{1,1}.initFcn='rands'; net.biases{2,1}.initFcn='rands'; i=load('input.txt'); t=load('target.txt'); i=i'; t=t'; in=zeros(8,53); %normalized input tn=zeros(1,53); %normalized target
for r=1:8 %normalization of input min=i(r,1); max=i(r,1); for c=2:53 if i(r,c)<min min=i(r,c); end if i(r,c)>max max=i(r,c); end end for c=1:53 in(r,c)=0.1+(0.8*(i(r,c)-min)/(max-min)); end end
min=t(1); %normalization of target max=t(1); for c=2:53 if t(1,c)<min min=t(1,c); end if t(1,c)>max max=t(1,c); end end for c=1:53 tn(1,c)=0.1+(0.8*(t(1,c)-min)/(max-min)); end
net.divideFcn='divideblock'; net.divideParam.trainRatio = 0.85; net.divideParam.valRatio = 0.05; net.divideParam.testRatio = 0.1; net.performFcn='mse'; [net,tr]=train(net,in,tn); y=sim(net,in);
2 Comments
Greg Heath
on 10 Sep 2013
1. Why would you post a long code that will not run when cut and pasted into the command line because there is no sample data???
2. NEVER use MATLAB function names for your own variables (e.g., max and min)
2. When beginning to write a program it is smart to try to use all of the defaults of the functions and use MATLAB data that is most similar to yours.
help nndata
3. Once that runs you can begin to modify it to fit your original problem.
4. Cut and paste the program to make sure it runs or to obtain the error messages.
5. Post code that can be cut and pasted into the command line.
6. Include relevant error messages.
Hope this helps.
Greg
Accepted Answer
Greg Heath
on 10 Sep 2013
1. There is no reason to use more than one hidden layer
2. You have created a net with 8 inputs instead of 1 8-dimensional input.
3. After creating a net view it using the command
view(net)
4. Why not just use fitnet?
help fitnet
5. After you rewrite your code you can test it on the 8-input/1-output chemical_dataset if you want to post further questions.
Hope this helps.
Thank you for formally accepting my answer
Greg
7 Comments
Greg Heath
on 25 Mar 2014
That doesn't make much sense to me because it is usually more important to decrease large errors than it is to decrease small errors with high relative errors.
More Answers (0)
See Also
Products
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!