# neural network with bayesian regularization: find weights and biases and recalculate the network

3 views (last 30 days)
Michael Arnold on 11 Dec 2020
Edited: Michael Arnold on 15 Dec 2020
Hey,
i´m trying to use a neural network to guess functional values for unknown points. This is my current solution.
%target f(x)=(x^2 + 22*x - 100)/(4*x)
%for x = [2,9]
inputall = 2:0.01:9;
outputall = (inputall.^2+22*inputall-100)./(4*inputall);
%training data
inputtrain = 2:1:9;
outputtrain = (inputtrain.^2+22*inputtrain-100)./(4*inputtrain);
%neural network
neurons = 5;
net = feedforwardnet(neurons,'trainbr');
net = train(net,inputtrain,outputtrain);
%prediction
predict(1,:) = net(inputall);
%comparison
comp = [outputall' predict']
%visualization
figure('Name','comparison'); hold on;
plot(inputall,outputall);
plot(inputall,predict)
Now I want to know what weights and biases the network finaly used. How can i get them and is it possible to use them to recalculate by myself the solution of the network?
Best regards
Michael

Sai Veeramachaneni on 15 Dec 2020
Hi,
You can use net.IW, net.LW, net.b properties of neural network object to get weights and biases used in the network.
You can use this as a reference to calculate solution using the constructed network.
References:
Michael Arnold on 15 Dec 2020
Edited: Michael Arnold on 15 Dec 2020
Thanks, that helps a lot. But i have trouble to find lines like
net.input.processFcns = { }; % Remove normalization
by my own. Have you a good tip where i can finde them? And can i see somewhere the code behind "net"?