problem defining linear regression condition

Hello, i have built the first code( the top one which uses matlabs regress command to aproximate [m10;m5] (NX1) vector with

[repmat([0.04,10],n,1);repmat([0.04,12],n,1)] NX2 matrices getting B(2X1) vector. it works great.

then i tried to do the same thing by myself using the linear regression theory and using itterations to get the same result as regress.

However I am stuck on the "fixing condition" the gradient descent. D_m(2)=(-2./n)*sum(x.*(y-y_pred));

is there some one who can help me with the gradient descent?

Thanks.

Regress function code:

n=10000;
min_value = 30+273-5;
max_value = 30+273+5;
%We need all out samples to be in the range of max_value min_value
T = min_value + (max_value - min_value) * rand(n,1);
%our emesivity samples 
e_delta=0.31622*randn(n,1);
var_e=var(e_delta); %emesivity variance
mean_e=mean(e_delta); %emesivity avarage
e = 0.9+0.1*e_delta;
var(e)
mean(e)
eT = e.*T;

m10 = 0.04*eT+10*e ;
m5  = 0.04*eT+12*e ;

[B,BINT,R] = regress([m10;m5],[repmat([0.04,10],n,1);repmat([0.04,12],n,1)]);

tested_T=B(1)/B(2)

plot(T);
xlabel('Sample number');
ylabel('Temperature');
ylim([min_value-2, max_value+2]);

My itterative code:

n=100000;

x=[repmat([0.04,10],n,1);repmat([0.04,12],n,1)];
m=[0;0];

L=0.0001;
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
min_value = 30+273-5;
max_value = 30+273+5;
T = min_value + (max_value - min_value) * rand(n,1);
e_delta=0.31622*randn(n,1);
e = 0.9+0.1*e_delta;
eT = e.*T;
m10 = 0.04*eT+10*e ;
m5  = 0.04*eT+12*e ;
y=[m10;m5];
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%%
%plot(x,y,'.b');
%hold;

for i=1:100
   y_pred=x*m;
   D_m(1)=(-2./n)*sum(x.*(y-y_pred));
   D_m(2)=(-2./n)*sum(x.*(y-y_pred));
   
   m=m-L*D_m;
end