Query: HMM predictions with trivial data

[gunjanthesystem]

Hi,

I am trying to check sanity of my usage of HMM. I was trying to predict the next (t+1) state of an observed variable (Activity Name ∈ {A,B,C,D}). The training data is trivial with 16 sequences of "<A B C D A>" . Hence, I would expect very good predictions. However, for each of the test instance, the prediction is 'A' (A gets the highest probability all the time). Could you kindly suggest if I this is to be expected or I am making some mistake?

note: training and test dataset are same.
using: FactoredFrontierForDBN with ImportanceSampling

Link to code
Link to data

Thanks a million,
gunjanthesystem