Hi Warren,
Can you check this code for your need once:
% Define the input and output patterns
P = [1 -1 1];
T = 1;
% Define the network architecture and parameters
nInputs = length(P);
nHidden = 1;
nOutputs = 1;
learningRate = 0.3;
bias = 1;
lambda = 0.6;
% Initialize the weights randomly
w1 = randn(nInputs,nHidden);
w2 = randn(nHidden,nOutputs);
% Define the activation functions
sigmoid = @(x) 1 ./ (1 + exp(-x));
dsigmoid = @(x) sigmoid(x) .* (1 - sigmoid(x));
unipolar = @(x) 0.5 * (1 + tanh(lambda * x/2));
dunipolar = @(x) lambda/2 * (1 - unipolar(x).^2);
% Initialize the delta arrays
delta2 = zeros(nOutputs,1);
delta1 = zeros(nHidden,1);
% Loop through the iterations of the backpropagation algorithm
for i = 1:100
% Forward pass
net1 = P * w1 + bias;
y1 = unipolar(net1);
net2 = y1 * w2 + bias;
y2 = unipolar(net2);
% Backward pass
delta2 = dunipolar(net2) * (T - y2);
delta1 = dunipolar(net1) .* (delta2 * w2');
% Update the weights
w2 = w2 + learningRate * y1' * delta2;
w1 = w1 + learningRate * P' * delta1;
end
% Test the network with the input pattern (1 -1 1)
net1 = P * w1 + bias;
y1 = unipolar(net1);
net2 = y1 * w2 + bias;
y2 = unipolar(net2);
% Print the output
fprintf('The predicted output is: %f\n', y2);
Hope it helps!
