I am beginner in machine learning…I am using Hidden markovs model to analyse the brain states in OSL dynamics tool box . Hidden Markov Model (HMM) — OSL Dynamics
q1) I have converted fmri data into time series using aal atlas and done the normalization and also done the frequency filteration for the time series data. But after the training i am getting loss function (screen shot attached)… how to know the training is bad or not ?is it only by getting free energy or by the function get_likelihood ? on epochs its reducing from 122 to 108 like that … but usually the loss should be shown as in percentage ? am i doing something wrong on taking the time series? or any other steps to do for better training? attached the image after the training…
q2) I have trained the model with the fmri time series using AAL parcellation. after training i tried to validate the model with other test data (fmri time series) . I was having an issue while trying to get the likelihood using the get_likelihood function.When i prepare the data for training it is an np_array with shape (time points, n_channels). However the get likelihood function expects a np_array with following shape (batch _size, seq_length, n_channels). Is there some step that i am missing in between to get this type of nparray. While i try to reshape my prepared data manually it throws error inside the get_likelihood function. If any one knows to use it would be a great help.