1

Can someone explain me the difference between the parameters transmat_ andtransmat_prior, when I setup the model ( e.g. GaussianHMM() in hmmlearn.hmm )?

class hmmlearn.hmm.GaussianHMM( n_components    =  1,
                                covariance_type = 'diag',
                                min_covar       =  0.001,
                                startprob_prior =  1.0,
                                transmat_prior  =  1.0,
                                means_prior     =  0,
                                means_weight    =  0,
                                covars_prior    =  0.01,
                                covars_weight   =  1,
                                algorithm       = 'viterbi',
                                random_state    =  None,
                                n_iter          = 10,
                                tol             =  0.01,
                                verbose         =  False,
                                params          = 'stmc',
                                init_params     = 'stmc'
                                )

Does the same explanation also hold for the parameters startprob_prior and startbrob_?

user3666197
  • 1
  • 6
  • 50
  • 92
mrt
  • 339
  • 1
  • 2
  • 14
  • Can you post the documentation? – wwii Nov 01 '17 at 16:19
  • Only the latter is a parameter, the former is an attribute (same applies to startprob). Look up the basic concept of hmm's (it's somewhat an initial-guess; but theory might be more complex). (i'm also unsure about keeping the sklearn tag as this was removed a long time ago from sklearn; **edit** op removed sklearn-tag) – sascha Nov 01 '17 at 16:23
  • 1
    @wwii here is the documentation: http://hmmlearn.readthedocs.io/en/latest/api.html#gaussianhmm – mrt Nov 01 '17 at 16:24
  • Neither `transmat_` nor `startbrob_ ` are parameters of `GaussianHMM`. – Stop harming Monica Nov 01 '17 at 16:34

1 Answers1

2

In my understanding, transmat_prior is the initial value for the transition matrix that you can specify (it will be used for initializing the iterative parameter estimation algorithm). It is a parameter of the class.

transmat_ is an attribute of an object of the class GaussianHMM, it gives the values of the transition matrix after training. It is not something you input yourself but a result of the estimation process.

Eskapp
  • 3,419
  • 2
  • 22
  • 39