Slow performance using polyfit on large arrays - how to speed up?

I have two large arrays PatlakX & PatlakY where I perform polyfit for each row in a for loop (Matlab2015b). The problem is the slow performance of polyfit in a for loop. Any good tips how to speed up?

Currently it takes around 20 min to complete.

%code below

PatlakX = 11337728x6 double
PatlakY = 11337728x6 double

for x=1:length(PatlakX);

P = fast_polyfit(PatlakX(x,:),PatlakY(x,:),1)%
k(x,:) = P(1);
m(x,:) = P(2);
r2(x,:) = rsquare(PatlakY(x,:),polyval(P,PatlakX(x,:)));

end

Best Regards

Emil

[Walter Roberson]

You do not appear to be using polyfit: you appear to be using fast_polyfit, which is not a Mathwork supplied routine.

[Emil]

It is a slightly modified polyfit with some unnecessary code removed (checkpoints) in order to speed up. Forgot to mention that modification.

Answer 1

[Walter Roberson]

http://www.mathworks.com/matlabcentral/answers/1836-multiple-use-of-polyfit-could-i-get-it-faster

Answer 2

[Dennie]

I don't believe the problem is that the for loop itself is slow. However, you have a tremendous amount of loops.

If the operation takes 20 min, that means that each loop takes around 0.1 ms (10 kHz).

It seems to me like you are making a linear fit of 6 points, you can also do this without polyfit and just make a simplified matrix operation out of this. Although I am not sure if this will be faster.

a=(PatlakX(:,6)-Patlakx(:,1))./(Patlaky(:,6)-Patlaky(:,1));
b= Patlaky(:,1)-a.*Patlakx(:,1);

This will give you the values for y=ax+b.

Ofcourse you could extend the linearization of the matrix to more complex models that average the slope of the 6 points, this was just an example.