Hi Xingyu, On 15/02/13 14:01, 那兴宇 wrote:
I didn't try that long window, but I agree with you on this. What I am interested is, when you use a real long window of the history, is it comparable with autoregressive HMM?
The reason I made my modifications to allow -1e10 values in the windows was indeed for the autoregressive HMM. The autoregressive HMM and the standard framework are just different models, though, no matter what windows are used.
Personally I have altered my HTS code to use -1e10 as a special value distinct from 0.0 in the window definition files to allow using windows of truly arbitrary sizes.Do you mean -1e10 or 1e-10?
I use -1e10 because it is unlikely to be used in a real window, whereas 1e-10 might conceivably be. It is a bit of a hack, though!
Matt