def distribution_hmm_modular(fm_cube, N, M, pseudo, order, gap, reverse, num_examples): from modshogun import StringWordFeatures, StringCharFeatures, CUBE from modshogun import HMM, BW_NORMAL charfeat=StringCharFeatures(CUBE) charfeat.set_features(fm_cube) feats=StringWordFeatures(charfeat.get_alphabet()) feats.obtain_from_char(charfeat, order-1, order, gap, reverse) hmm=HMM(feats, N, M, pseudo) hmm.train() hmm.baum_welch_viterbi_train(BW_NORMAL) num_examples=feats.get_num_vectors() num_param=hmm.get_num_model_parameters() for i in range(num_examples): for j in range(num_param): hmm.get_log_derivative(j, i) best_path=0 best_path_state=0 for i in range(num_examples): best_path+=hmm.best_path(i) for j in range(N): best_path_state+=hmm.get_best_path_state(i, j) lik_example = hmm.get_log_likelihood() lik_sample = hmm.get_log_likelihood_sample() return lik_example, lik_sample, hmm