Excluding data points while using least square method

Hi all. How can I exclude some points from my response data ( entries of value 0 ) while estimating the coefficient for my ( multiple ) linear regression model using least squares, so that obv the corresponding points in the x,y,z matrices ( predictors ) will be excluded too ? Thank you.

Answer 1

[John D'Errico]

How can you exclude them?

You do that by excluding them. Just drop them from your dataset. Create a new copy that lacks those points, then use the modified set for the estimation. You can just use find to find the points with a zero. Or use find to find the points that are not zero.

WTP?

[Damian Wierzbicki]

Yeah, but then how do I exclude the corresponding points in my predictors matrices ? Somehow I'd have to store the positions of the zero entries and then exclude these points of those positions from my x , y , z matrices ? Not sure how to do that.

[Damian Wierzbicki]

Since, when you're using '\' the dimensions of the matrices must agree with each other.

[John D'Errico]

I'd suggest you need to read the getting started tutorials.

Learn to use find, although you might note that I never even needed to use find. A logical vector was sufficient, but I could have used find. Learn to use indexing.

% make up some data
xy = randn(20,2);
z = randn(20,1);

% As an example, select and use only the data with z greater than zero
ind = z > 0;
zhat = z(ind);
xyhat = xy(ind,:);

% use backslash
pred = xy\z;

Or, if you don't want to create those extra variables, don't.

ind = z > 0;
pred = xy(ind,:)\z(ind);

Or use find:

ind = find(z > 0);
pred = xy(ind,:)\z(ind);

Or never even create an index vector at all:

pred = xy(z > 0,:)\z(z > 0);

All basic MATLAB.