Need help with Least Squares fit problem! Not overly difficult I don't believe.

Hello, I am doing some homework for my computational methods class and need some help. We are covering linear regressions and least fit squares. So the problem I am doing is having me derive by hand the least fit squares for the equation y=a1x+e, meaning determine the slope that results from a straight line with a zero intercept. I did it by hand and got a slope of 0.61436, but was wondering how I would go about checking on matlab if this was correct?

The x and y data are as follows: X: 2 4 6 7 10 11 14 17 20 Y: 4 5 6 5 8 8 6 9 12 I understand basic plotting but am unsure of how to fit the line to it on here. Thanks!

Answer 1

[Star Strider]

Use the MATLAB mldivide,\ (link) function to calculate the least-squares fit:

X = [2 4 6 7 10 11 14 17 20];
Y = [4 5 6 5 8 8 6 9 12];
a1_1 = X(:)\Y(:)                                    % Using ‘mldivide,\’

You may want to check your calculations and derivation, since I get a different result from taking the first (partial) derivative of:

S = sum((Y - a1*X)^2)

with respect to ‘a1’, then setting it to zero and solving for ‘a1’:

a1_2 = sum(X.*Y)/sum(X.^2)                          % Using Computed Least Squares

My derivation result agrees with the mldivide result.

Plotting:

figure(1)
plot(X, Y, 'pg')
hold on
plot(X, a1_1*X, '-r')
hold off
grid
text(7, 11, sprintf('Y = %.3f\\cdotX', a1_1))
text(7, 13, sprintf('Y = %.3f\\cdotX', a1_2))

[Grant Piephoff]

Thank you so much! I recalculated my numbers and got the same value as you.

[Star Strider]

As always, my pleasure!

I decided to do the derivation as well (by hand, not using the Symbolic Math Toolbox) to be certain I could get the same result.

Answer 2

[Kaushik Lakshminarasimhan]

regress(Y,X)