forum.alglib.net

I want to minimize a function with the lbfgs algorithm.
I saw there is a optimize function that perform the complete minimization, and we can also give the max number of iteration. But I would like to perform iteration step by step. Something like the following code :

init_minimization
iterate_minimization( 100 ) // 100 iterations
show the result with open_gl
iterate_minimization( 100 )
....

Is it possible ?

Yes, it is possible and can be achieved by two completely different ways.

You can specify two stopping criteria: gradient-based and iteration-based one. When optimizer is terminated by iteration-based criteria, you visualize result and call it one more time. However, optimizer will be restarted from scratch - it will lose all information accumulated so far except for the point where it was stopped.

Maybe it is better to use another option - to turn on reports with minlbfgssetxrep() function and to pass rep callback to minlbfgsoptimize(). In this case you will be able to visualize iterations without restarting algorithm.