createNN2.m

function net = createNN2(neurons_per_layer,par)
% create a CV NN
% nu is the dim of uk
% num_layers n e.g 5 including input and output layer
% neurons_per_layer  
% e.g. [4 2 3 4 2] first is the size of input  end is the  size of output
nu = par.nu;
nx = par.nx;
% lbu = par.lbu;
% ubu = par.ubu;
num_layers = length(neurons_per_layer);

% Load CasADi
import casadi.*

% Create symbolic variables for learable paremters
corrP{num_layers-1} =[];startInd = 1; w0=[];
for i =1:num_layers-1
    per_P_num = (neurons_per_layer(i)+1)*neurons_per_layer(i+1);
    corrP{i} = startInd:startInd+per_P_num-1 ;
    startInd = startInd+per_P_num;
    % rng(i)
    w0 = [w0;initializeGlorot(per_P_num,neurons_per_layer(i),neurons_per_layer(i+1))];
end
toll_P_num = startInd-1;
theta = MX.sym('theta',toll_P_num);

% Create symbolic variables for inputs and outputs

uk = MX.sym('uk',nu);  % Input
Xi = MX.sym('Xi',nx); % Output


% Create hidden layers
outLayer = [Xi];
for i = 1:num_layers - 1
    W = reshape(theta(corrP{i}(1:end-neurons_per_layer(i+1))),neurons_per_layer(i+1),neurons_per_layer(i));
    b = theta(corrP{i}(end-neurons_per_layer(i+1)+1:end));
    if i< num_layers - 1
        outLayer = tanh(W*outLayer+b); % Replace with desired activation function
    else
        % Create output layer (using linear activation)
        outLayer = W*outLayer+b; 
        % outLayer = min(max(W*outLayer+b,lbu),ubu); 
    end
end


% Define the neural network as a CasADi function
nn = Function('nn', {Xi,theta}, {outLayer});

net.nn =nn;
net.corrP = corrP;
net.w0 = w0;
end

function weights = initializeGlorot(sz,numOut,numIn)

Z = 2*rand(sz,1) - 1;
bound = sqrt(6 / (numIn + numOut));

weights = bound * Z;
% weights = dlarray(weights);

end