pohmm.py 文件源码

def _compute_log_likelihood(self, obs, pstates_idx):

        q = np.zeros(shape=(len(obs), self.n_hidden_states, self.n_features))

        for col, (feature_name, feature_distr) in enumerate(zip(self.emission_name, self.emission_distr)):
            if feature_distr == 'normal':
                mu = self.emission[feature_name]['mu'][pstates_idx]
                sigma = self.emission[feature_name]['sigma'][pstates_idx]
                for j in range(self.n_hidden_states):
                    q[:, j, col] = np.log(
                        np.maximum(MIN_PROBA, stats.norm.pdf(obs[:, col], loc=mu[:, j], scale=sigma[:, j])))
            if feature_distr == 'lognormal':
                logmu = self.emission[feature_name]['logmu'][pstates_idx]
                logsigma = self.emission[feature_name]['logsigma'][pstates_idx]
                for j in range(self.n_hidden_states):
                    q[:, j, col] = np.log(np.maximum(MIN_PROBA,
                                                     stats.lognorm.pdf(obs[:, col], logsigma[:, j], loc=0,
                                                                       scale=np.exp(logmu[:, j]))))

        q = q.sum(axis=2)
        return q