poisson_hmm_compute_loglikelihood.py

    def compute_loglikelihood(self):
        #Init the list of loglikelihhod values, one value for each y observation
        ll = []
        for t in range(len(self.y)):
            prob_y_t = 0
            mu_t = 0
            for j in range(self.k_regimes):
                #To use the law of total probability, uncomment this row and comment out the next
                # two rows
                #prob_y_t += poisson.pmf(y[t], mu[t][j]) * self.delta_matrix[t][j]
                #Calculate the Poisson mean mu_t as an expectation over all Markov state
                # probabilities
                mu_t += self.mu_matrix[t][j] * self.delta_matrix[t][j]
                prob_y_t += poisson.pmf(self.y[t], mu_t)
            #This is a bit of a kludge. If the likelihood turns out to be real tiny, fix it to
            # the EPS value for the machine
            if prob_y_t < self.EPS:
                prob_y_t = self.EPS
            #Push the LL into the list of LLs
            ll.append(math.log(prob_y_t))
        ll = np.array(ll)
        return ll