MATLAB: Problem in solution from Linprog

Question

Hi,

I am using linprog to solve for parameters of the following problem –

minimise -a

subject to constraint : a + b*x < y

The parameter estimates that I get from linprog (a & b), should ideally give me errors which are all positive ie

e = y – a -b*x

e > 0, for all y and x

but few of my errors have value -0.0000

how can I get around this problem?

—————-

A = ones(80,1);

f = [-1;0];

options = optimoptions('linprog','Algorithm','simplex')

p_fit=linprog(f,[A xdata],[ydata],[],[],[-Inf -Inf],[],p_initial,options)

———————————–

I have attached data for x and y in an excel file.

Thanks for your help!

Regards, Prachi

Answer 1

Time to learn about the 'ConstraintTolerance' parameter, and about floating point arithmetic in general.

Are those values really negative? Or are they just effectively zero, plus or minus some trash? Note all of those zeros. In fact, it was reported as -0.0000. That should be the hint that it is just negative "zero".

If you cannot tolerate that amount of deviation, then you can add a tiny offset to the constraints.