how can i use particleswarm command to update weights and biases of a neural network ?

%this is a translation of the NARX neural network equation into matlab, where all weights and biases are equals to 0, i want to update them using particleswarm optimization to minimize the error.

order=3;

hiddenlayers=2;

H=[];iw=[];lw=[];rlw=[];bi=[];bo=1;

for ligne_countiw=1:hiddenlayers

for col_countiw=1:order

iw(ligne_countiw,col_countiw)=1;

end

end

for ligne_countlw=1:hiddenlayers

for col_countlw=1:order

lw(ligne_countlw,col_countlw)=1;

end

end

for countbi=1:hiddenlayers

bi(countbi)=1;

end

for countrlw=1:hiddenlayers

rlw(countrlw)=1;

end

H(1,1)=tanh(bi(1));

H(1,2)=tanh(bi(2));

IW=[];LW=[];

for ligne=1:1000

for col=1:1000

IW(ligne,col)=0;

end

end

for ligne=1:1000

for col=1:1000

LW(ligne,col)=0;

end

end

for t=2:order

for i=1:hiddenlayers

for r=1:t-1

IW(t,i)=IW(t,i)+iw(i,r)*XTrain(t-r);

LW(t,i)=LW(t,i)+lw(i,r)*TTrain(t-r);

end

H(t,i)=tanh(IW(t,i)+LW(t,i)+bi(i));

end

end

for t=order+1:length(XTrain)

for i=1:hiddenlayers

for r=1:order

IW(t,i)=IW(t,i)+iw(i,r)*XTrain(t-r);

LW(t,i)=LW(t,i)+lw(i,r)*TTrain(t-r);

end

H(t,i)=tanh(IW(t,i)+LW(t,i)+bi(i));

end

end

Y_estimated=[];

for counter=1:1000

Y_estimated(counter)=0;

end

for t=1:length(XTrain)

for i=1:hiddenlayers

Y_estimated(t)=Y_estimated(t)+rlw(i)*H(t,i)+bo;

end

end

mse = sum((Y_estimated-TTrain).^2)/length(Y_estimated);

Answer 1

[Darshak]

Hello sami dahdouh,

I understand that you want to use particle swarm optimization in your MATLAB code.

Particle swarm optimization can be done in MATLAB using the “particleswarm” function. You may refer to the MATLAB code below which illustrates the use of “particleswarm” function:

objective = @(W) sum((narx_predict(W, XTrain, order, hiddenlayers) - TTrain).^2) / length(XTrain);

options = optimoptions('particleswarm','SwarmSize',50,'Display','iter','MaxIterations',100);

[W_opt, fval] = particleswarm(objective, nvars, [], [], options);

This is a sample MATLAB code that illustrates the use of “particleswarm” function. You can customize it according to the objective function.

Kindly refer to the documentation link below for any further queries on “particleswarm” function.

https://www.mathworks.com/help/releases/R2021b/gads/particle-swarm.html

Hope this helps you!