at first i have to mention that I've read and watched different tutorials about the implementation for murphys toolbox and also have some backround to the topic of HMM. But all application examples handle either the subject to predict the weather or voice recognition. Neither of the two cases is similar to my application.
I have to make a classification for my master's thesis. In that case the model should distinguish between two different driving maneuvers. As input variables i have 20 differnent vehicle specific quantities for every maneuver from several test persons. This means i want to train the model with train data and made a classification for test people. I know that for each state a separate model must be trained, in my issue two. In this case i would set:
- O=2; %number of states (maneuver1, maneuever2)
- Q=20; %number of discrete observation (all 20 vehicle specific quantities)
- I know the meaning of prior0, transmat0 and obsmat0, but did i need these one if i want to train the model with training data? And if yes how i get these Informations?
If read that the murphys toolbox offers different subdivison like dhmm, ghmm and mhmm. Which of these is most appropriate or where is the difference.
The em algorithm is used to train the model, how do I put my data record here? Is it necessary to make the variable prior, transmat and obsmat random, as seen in different tutorials?
As you can see, I have many questions, which have not yet been answered, hopefully you can help me.