what is the main reason of no convergence after of alot of iterations?

Question

Hi there

sometimes when I run command file which uses cmlmt, Despite, more iterations, could not get convergence.

I want to know, how can I deal with this? do make some changes in Algorithm setting, can be useful or I have to do somethings else?

2 Answers

Your Answer

aptech · Answer 1

There are a few common possibilities. If the surface of the log-likelihood function is relatively flat, the algorithm can have trouble finding a real local minimum. Also if the data is scaled poorly, there can be catastrophic cancellation in the computation of the derivatives. In that case, the algorithm will not be able to find a correct search direction.

What are your CMLMT control settings?

link

aptech

1,773

mazegar · Answer 2

// Algorithm settings
struct cmlmtControl c0;
c0=cmlmtControlCreate;
c0.Algorithm=4;
c0.LineSearch=4;
c0.CovParType=2;
c0.printIters=1;
c0.DirTol=0.0001;
//c0.TrustRadius=0.001;
//c0.Switch=1|10000000|0;
//c0.Switch={3 1, 0.1 0.1, 1 1, .001 .001};

link

mazegar

8

aptech · Answer 3

There are a few common possibilities. If the surface of the log-likelihood function is relatively flat, the algorithm can have trouble finding a real local minimum. Also if the data is scaled poorly, there can be catastrophic cancellation in the computation of the derivatives. In that case, the algorithm will not be able to find a correct search direction.

What are your CMLMT control settings?

link

aptech

1,773

mazegar · Answer 4

// Algorithm settings
struct cmlmtControl c0;
c0=cmlmtControlCreate;
c0.Algorithm=4;
c0.LineSearch=4;
c0.CovParType=2;
c0.printIters=1;
c0.DirTol=0.0001;
//c0.TrustRadius=0.001;
//c0.Switch=1|10000000|0;
//c0.Switch={3 1, 0.1 0.1, 1 1, .001 .001};

link

mazegar

8

what is the main reason of no convergence after of alot of iterations?

2 Answers

Your Answer

2 Answers

You must login to post answers.

Have a Specific Question?

Need Support?