Performance degradation after ~3000 iterations during parameter estimation

[electricmomo]

Hi,

Thanks for the library.

I adapted example-001.cc to my case where I model a noisy timeserie by a MLP. Estimation performance is good up to about 4000-5000 samples, then it inexplicably starts degraging quite horribly.

I changed the starting point to rule out some data effect issue,and consistently, after 4000-5000 samples, it got bad.

Any idea?

Thanks

[jeremyfix]

Well basically , I have no idea what might be going on ....

when you say you changed your starting point, you mean the initlalization of the parameter vector or the samples you draw from your timeserie ?

maybe getting access to your code might help debugging your code .

Jeremy.