| forum.alglib.net http://forum.alglib.net/ |
|
| minlbfgs conversion problem. http://forum.alglib.net/viewtopic.php?f=2&t=614 |
Page 1 of 1 |
| Author: | kryp33 [ Tue Sep 18, 2012 2:48 am ] |
| Post subject: | minlbfgs conversion problem. |
Hello All, I just started to use AlgLib 3.6.0 in C++ incarnation, so I may be missing something simple. Can you, please, help me with this problem: I am trying to use function minlbfgs to find a minimum given the user defined function callback which returns a func and grad's. The way how I sue it: real_1d_array x; x.setcontent(initTheta.size(),initTheta.memptr()); //All zeros minlbfgscreate(1, x, state); minlbfgssetcond(state, 0, 0, 0, m_nIterrations); alglib::minlbfgsoptimize(state, fnCostLogReg,0,this); minlbfgsresults(state, x, rep); I noticed(in debugger) that on the second iteration the real_1d_array &x parameter in the call back function has got changed values for the next iteration. But these values are too large. I compare them with the same code I have in Octave and they are exactly 10 times bigger. This break my convergence. I am not sure how this algorithm works internally, but if I use an analogy with standard gradient descend method, then I have a feeling that there should be some parameter(lambda) that regulates the step of descend. Is there something I forgot to set up in minlbfgsoptimize? Thanks a lot, Sergey Andreyev |
|
| Author: | kryp33 [ Wed Sep 19, 2012 7:58 pm ] |
| Post subject: | Re: minlbfgs conversion problem. |
Hi All, I added the call minlbfgssetstpmax and finally got a correct results. Problem solved. |
|
| Page 1 of 1 | All times are UTC |
| Powered by phpBB © 2000, 2002, 2005, 2007 phpBB Group http://www.phpbb.com/ |
|