CODEWITHSUNDEEP
matlaboptimizationlinear-regressiongradient-descent
ajax2000
ajax2000

Reputation: 711

MATLAB constrained optimization with theta constraints

I want to find the theta that minimizes l2 norm of y-X*theta, and constraint is that some element of theta should equal to 0. How do I do this in MATLAB?

My y is 10x1, X is 10x9, theta is 9x1

I think using gradient descent may be helpful not sure

Upvotes: 0

Views: 99

Answers (1)

sascha
sascha

Reputation: 33532

There are multiple interpretations possible here, because of the informal description.

If you just want to solve the linear least-squares problem and a-priori fix some variables to zero, use matlab's lsqlin and use upper-bound=lower-bound=0 for those variables.

If somehow your task is to solve the same objective subject to something like: at least w values of theta are 0, this is probably NP-hard (in the general setting) and you would need Mixed-integer Quadratic-Programming or Mixed-integer Second-order-cone Programming. In this case, i'm not sure if there is a solver available. Some open-source solvers would be Bonmin (Couenne, to a lesser degree) or most commercial solvers like (Gurobi, CPLEX, Mosek).

Upvotes: 2

Related Questions