I use Matlab's neural network toolbox for predicting and estimating my problem by experimental data. Now, I want to drive out the equation which correlates the inputs to the outputs.(for example- inpouts:a,b,c and output:m - goal: a0.1+b0.5+3 c1.1=m)
why do not try curve fitting toolbox in Matlab?
just type cftool in command window and insert your data
you should clarify the function type(i.e. linear, non-linear, sin,...) and the toolbox returns you the parameters you want
Good Luck
Thanks dear hossein, I know that it's possible by curve fitting but I want to drive out the equation by neural network which correlates the inputs to the outputs so that I can use it as an objective function in order to be optimized by Genetic Algorithm and combine NN and GA.
I suppose you mean layered feed forward networks.
The thresholds of each layer are vectors,
and the mapping between layers are matrices.
Lots of floating point parameters, and you
can bundle all of them in a single list,
call it the genome, and apply the evolution.
Regards,
Joachim
Ali:
Lets suppose that you want to train a NN for regression, some inputs and one target like this:
hiddenLayerSize = [6 2];
net = fitnet(hiddenLayerSize);
[net,tr] = train(net,inputs,targets);
The following code in MATLAB should print the NN configuration, weights, activation function and bias. Then you can build your equation... you may try to adapt it to your personal problem. If you found any error, please, let me know.
clc;
fprintf('Network Configuration\n');
fprintf('Internal weights for each layer\n\n');
fprintf(':::: Input Layer/Hidden1 ::::\n');
for R=1:net.inputs{1}.size
fprintf('X%d\n',R)
for S=1:hiddenLayerSize(1)
fprintf('W(%d,%d)=%f\n',R,S,net.IW{1}(S,R));
end
fprintf('-------\n');
end
fprintf('Node Bias\n');
for R=1:net.layers{1}.dimensions
fprintf('b(%d)=%f\n',R,net.b{1}(R));
end
fprintf('-------\n');
fprintf('Activation function: %s\n',net.layers{1}.transferFcn);
fprintf('*********************************\n');
for R=2:net.numLayers-1
fprintf(':::: Input Hidden %d ::::\n',R);
%fprintf('%d,%d\n',R,R-1);
[NODES,IN_NODES]=size(net.LW{R,R-1});
for K=1:IN_NODES
fprintf('Node %d \n',K)
for S=1:NODES
fprintf('W(%d,%d)=%f\n',K,S,net.LW{R,R-1}(S,K));
end
fprintf('-------\n');
end
fprintf('Node Bias\n');
for m=1:net.layers{R}.dimensions
fprintf('b(%d)=%f\n',m,net.b{R}(m));
end
fprintf('-------\n');
fprintf('Activation function: %s\n',net.layers{R}.transferFcn);
fprintf('---------------------------------\n');
end
fprintf('*********************************\n');
fprintf(':::: Output Layer ::::\n');
R=R+1;
[NODES,IN_NODES]=size(net.LW{R,R-1});
for K=1:IN_NODES
fprintf('Node %d \n',K)
for S=1:NODES
fprintf('W(%d,%d)=%f\n',K,S,net.LW{R,R-1}(S,K));
end
fprintf('-------\n');
end
fprintf('Output bias\n');
fprintf('b=%f\n',net.b{net.numLayers}(1));
fprintf('-------\n');
fprintf('Activation function: %s\n',net.layers{R}.transferFcn);
fprintf('*********************************\n');
fprintf('Done...!\n');
I've got the same question. Could you finally derive an equation from your neural network? I'd appreciate if you could keep me posted. Ali N. Bakhtiyari
I think its impossible to get a certain equation between all input and target data, because the NN involve with some weights so that the best answer not obtained by equation but the best answer achieved by the distance concept(MSE) between input and target data.
- Badges
- Science topic