I use the HCRF Library for the sequence prediction task. I use LBFGS to study model parameters, although there are options for using CG and BFGS.
I notice that it converges for a very long time. I set the maximum number of iterations (currently) to 5000. If my interpretation of the HCRF library is correct, the changes in the estimates from iteration to iteration become smaller. However, the program does not seem to end in less than 5,000 iterations, and it really takes a lot of time.
If I set the maximum iteration to 500, then at the 500th iteration, the result will look like this:
Iteration 500
fx = 465.359
xnorm 38.0831, gnorm = 46.3165, step = 1line search = 1
He completes this message:
L-BFGS optimization terminated with status code = -997
fx = 465.359
which means that the maximum iteration is reached.
?
(, 300 )? , "" ?
, , 20 000 - 30 000 14 .
, , , .