Closed EyobZewdie closed 6 years ago
Hmm. If you terminate because you hit the maximum number of iterations, it means you haven't really provably converged.
Does f_approx come out as NaN? Could you please send me some more details about the problem?
Is the dual feasible if the method terminates after max iterations?
The issue is fixed by https://github.com/lkapelevich/SDDiP.jl/commit/162237e7ae5aa4fe932245a68cba6fabb55f7e24.
The dual is feasible, so we could be smarter and deliberately terminate early.
The dual is not necessarily feasible...
When you run the level method and it hits the maximum number of iterations it fails with f_approx not defined.