2D fused Lasso with Matlab CVX

Question

I wrote a 2D fused lasso code here.

[m n] = size(circle);
cvx_begin
    variable theta(m, n);
    minimize( norm(circle-theta, 'fro'));
    subject to
        sum(sum(abs(theta(:,1:n-1)-theta(:,2:n)))) == 0;
        sum(sum(abs(theta(1:m-1,:)-theta(2:m,:)))) == 0;
cvx_end

Weirdly, the program report,

In cvxprob (line 28) In cvx_begin (line 41) Error using cvxprob/newcnstr (line 192) Disciplined convex programming error:
Invalid constraint: {convex} == {constant}

Error in == (line 12) b = newcnstr( evalin( 'caller', 'cvx_problem', '[]' ), x, y, '==' );

After I remove abs() in the constraint, the program could run, but that's not what constraints I expect to be.

huh what. That is not MATLAB code, is it? – Ander Biguri Sep 09 '16 at 10:26 — Ander Biguri, Sep 09 '16 at 10:26
It is using the cvx package under MATLAB – xxx222 Sep 09 '16 at 10:52 — xxx222, Sep 09 '16 at 10:52

score 0 · Answer 1 · answered Apr 25 '19 at 16:20

0

I think you can try stacking the matrices into vectors, then use L1 norm. In CVX, it's just norm(variable, 1). It will do the same as you wrote here: sum of absolute elementary-wise differences.

answered Apr 25 '19 at 16:20

Linus Zhang

1

2D fused Lasso with Matlab CVX

1 Answers1