How to do Gradient Descent Optimization Algorithm in MATLAB | MATLAB Tutorial

2022 Gradient Descent Algorithm in MATLAB! How to optimize a function using Gradient Descent. 3D example with equations, code, and explanation for beginners.

#matlab #LearnMATLAB #GradientDescent

CHAPTERS
0:00 Overview
0:25 Equations
2:43 Algorithm
8:36 Learning Rate

MATLAB BASICS

ADVANCED MATLAB

HELP A BROTHER OUT
Thank you so much for watching! Please consider subscribing as a thank you if you benefited from this content! :D

HELPING MY BROTHERS AND SISTERS OUT
Comment below and I will happily answer all your questions, queries, and conundrums... whether or not they pertain to programming ;)

[alexarunyan]

FINALLY a true video explaining gradient descent and the algorithm behind it

[durlov]

Thanks for the explanation broken down in simple but detailed steps!

[mahadevanar]

Nice one ...
Please make videos about divergence and others if you have time♥️♥️

[s.a.]

Great video!! Just one question... what is this anonymous function used for? f = 1/4*x^2 + y^2 ? Why did you select and use this function in particular?

[madreerdam]

I have a question: I have to use GD for another type of function, but i got stuck.
I changed object function for both f and fs, respectively:
"f = @(x, y) x.*exp(-x.^2-y.^2);
syms fs xs ys
fs = xs.*exp(-xs.^2-ys.^2);" but i got this error:
"Not enough input arguments.
Error in symengine>@(xs, ys)exp(-xs.^2-ys.^2)-xs.^2.*exp(-xs.^2-ys.^2).*2.0
Error in Gradient_Descent (line 46)
pts(i+1, 1) = pts(i, 1) - gamma*grad_x(pts(i, 1));"
Code is the same, just changing domain, starting point and function.
Could be that exp create problems? Why that?
( Really appreciated your work, thanks for sharing <3 )

[maazmohammad]

Thank you for the great explanation. I had a question. If I have weights in an equation and changing those weights changes the output value constant(F). Can I still use gradient descent on the constant and change the weights to get the lowest constant(F) possible?

[skidzeess]

Man, thank you so much for your video. I've tried it myself to make a univeristy homework. but I've got a problem with my function:

f = @(x, y)

I've changed your code, to fit my function, but It did't work "well". One guy had same problem down the comment section ("Not enough input arguments."), I think I've caught how to solve it, but the resoult is shit :(

Here is what I did:

while err_x > 0.01 || err_y > 0.01

gradix = grad_x(pts(i, 1), pts(i, 2)); %first component of gradient
gradiy = grad_y(pts(i, 1), pts(i, 2)); %second component of gradient
grad_f = sqrt(gradix^2 + gradiy^2); %gradient length


pts(i+1, 1) = pts(i, 1) - betta/grad_f*grad_x(pts(i, 1), pts(i, 2)); %x
pts(i+1, 2) = pts(i, 2) - betta/grad_f*grad_y(pts(i, 1), pts(i, 2)); %y
pts(i+1, 3) = f(pts(i+1, 1), pts(i+1, 2)); %z (from x, y)
% Update difference between consecutive points
err_x = abs(pts(i+1, 1) - pts(i, 1));
err_y = abs(pts(i+1, 2) - pts(i, 2));
% increment i (move to next row)
i = i + 1;
% break loop if going too long
if i > 100
break
end
end


Where am I wrong?

Thank you, hope you will answer!