HMM input weighting and continuous learning HMM / HCRF

[cjst...@gmail.com]

Hi Cesar!

Long time. For HMM inputs is it possible to weight the relevance of a specific input variable over another?

Also, for the HMM and HCRF teacher. Can it be run continuously, introducing new samples as it learns or do I create a new implementation every time as I do currently?

Craig

[César]

Hi Craig! 

Indeed, it has been some time! Unfortunately the HMM doesn't support sampling individual weight variables at this time. But changing this would depends on the distribution that you are using to create the HMMs. Depending on the distribution this could be possible.

As for the teachers, unfortunately it is not currently possible to run the HMM teacher continuously; but for the HCRF then it might be possible if you use HiddenGradientDescentLearning or HiddenResilientGradientLearning (I recommend the resilient version). You might need to set their Stochastic property to true in order to train them properly.

Hope it helps!

Best regards,

Cesar