def _test11(): # A reduced version of test10 ret = "" # GmmMgr setup num_states = 2 dimension = 2 models = [] for i in xrange(num_states): dm = DummyModel(dimension, 1.0) models.append(dm) gmm_mgr = GmmMgr(models) gb = GraphBuilder() node_id0 = gb.new_node((0, 0)) node_id1 = gb.new_node((1, 1)) node_id2 = gb.new_node((2, 1)) node_id3 = gb.new_node((3, 1)) node_id4 = gb.new_node((4, 2)) # The topology here is slightly complex than the previous example arc_id = gb.new_arc(node_id0, node_id1) arc_id = gb.new_arc(node_id1, node_id4) arc_id = gb.new_arc(node_id0, node_id2) arc_id = gb.new_arc(node_id2, node_id3) arc_id = gb.new_arc(node_id3, node_id4) arc_id = gb.new_arc(node_id2, node_id4) gr0 = FrozenGraph(gb) # Make two Hmms with 3 states and order 2 (self loop, forward 1) # The models in the middle are special and can skip. seed(0) hmm0 = make_forward_hmm(gmm_mgr, num_states, order=2, exact=False) hmm1 = Hmm(1) trans = array(((0.0, 0.5, 0.5), (0.0, 0.5, 0.5), (0.0, 0.0, 0.0))) hmm1.build_model(gmm_mgr, (0, ), 1, 1, trans) hmm2 = make_forward_hmm(gmm_mgr, num_states, order=2, exact=True) hmm_mgr = HmmMgr((hmm0, hmm1, hmm2)) spd = {} spd[(0, 1)] = (0.4, ) spd[(0, 2)] = (0.6, ) spd[(2, 3)] = (0.4, ) spd[(2, 4)] = (0.6, ) tg0 = TrainingGraph(gr0, hmm_mgr, split_prob_dict=spd) if do_display: tg0.dot_display() tg0.dot_display(expand_hmms=True) with DebugPrint("bwt_ctsh") if True else DebugPrint(): result_hmm = tg0.convert_to_standalone_hmm() ret += "\n\n========= TG CONVERTED TO Hmm =========\n\n" + result_hmm.to_string( full=True) return ret
def _test11(): # A reduced version of test10 ret = "" # GmmMgr setup num_states = 2 dimension = 2 models = [] for i in xrange(num_states): dm = DummyModel(dimension, 1.0) models.append(dm) gmm_mgr = GmmMgr(models) gb = GraphBuilder() node_id0 = gb.new_node((0, 0)) node_id1 = gb.new_node((1, 1)) node_id2 = gb.new_node((2, 1)) node_id3 = gb.new_node((3, 1)) node_id4 = gb.new_node((4, 2)) # The topology here is slightly complex than the previous example arc_id = gb.new_arc(node_id0, node_id1) arc_id = gb.new_arc(node_id1, node_id4) arc_id = gb.new_arc(node_id0, node_id2) arc_id = gb.new_arc(node_id2, node_id3) arc_id = gb.new_arc(node_id3, node_id4) arc_id = gb.new_arc(node_id2, node_id4) gr0 = FrozenGraph(gb) # Make two Hmms with 3 states and order 2 (self loop, forward 1) # The models in the middle are special and can skip. seed(0) hmm0 = make_forward_hmm(gmm_mgr, num_states, order=2, exact=False) hmm1 = Hmm(1) trans = array(((0.0, 0.5, 0.5), (0.0, 0.5, 0.5), (0.0, 0.0, 0.0))) hmm1.build_model(gmm_mgr, (0,), 1, 1, trans) hmm2 = make_forward_hmm(gmm_mgr, num_states, order=2, exact=True) hmm_mgr = HmmMgr((hmm0, hmm1, hmm2)) spd = {} spd[(0, 1)] = (0.4,) spd[(0, 2)] = (0.6,) spd[(2, 3)] = (0.4,) spd[(2, 4)] = (0.6,) tg0 = TrainingGraph(gr0, hmm_mgr, split_prob_dict=spd) if do_display: tg0.dot_display() tg0.dot_display(expand_hmms=True) with DebugPrint("bwt_ctsh") if True else DebugPrint(): result_hmm = tg0.convert_to_standalone_hmm() ret += "\n\n========= TG CONVERTED TO Hmm =========\n\n" + result_hmm.to_string(full=True) return ret
def test5(num_obs, do_display=False): # A test in which one of the HMMs has a transition from an input directly to # an output, so it can behave as an epsilon. This node is between two other # nodes in a linear arrangement. # Data generator setup and data generation dimension = 2 obs_gen = make_data_generator(dimension) obs_list = [obs_gen.next() for i in xrange(num_obs)] # GmmMgr setup num_models = 20 models = make_standard_gmms(dimension, num_models) gmm_mgr1 = GmmMgr(models[0:10]) gmm_mgr2 = GmmMgr(models[10:20]) # Hmm setup # Make two Hmms with 2 states and order 2 (self loop, forward 1) The model # in the middle is special in that it can skip directly from the input state # to the output state. seed(0) num_states = 2 hmm0 = make_forward_hmm(gmm_mgr1, num_states, 2, exact=False) hmm1 = Hmm(1) trans = array(((0.0, 0.5, 0.5), (0.0, 0.5, 0.5), (0.0, 0.0, 0.0))) hmm1.build_model(gmm_mgr1, (0, ), 1, 1, trans) hmm2 = make_forward_hmm(gmm_mgr1, num_states, 2, exact=False) hmm_mgr = HmmMgr((hmm0, hmm1, hmm2)) # TrainingGraph setup gb = GraphBuilder() node_id0 = gb.new_node((0, 0)) node_id1 = gb.new_node((1, 1)) # node_id2 = gb.new_node((2,2)) arc_id = gb.new_arc(node_id0, node_id1) # arc_id = gb.new_arc(node_id1, node_id2) gr0 = FrozenGraph(gb) tg0 = TrainingGraph(gr0, hmm_mgr, split_prob_dict=dict()) if do_display: tg0.dot_display() tg0.dot_display(expand_hmms=True) valid, ret = validate_training_graph(tg0, gmm_mgr1, hmm_mgr, obs_list, 1, gmm_mgr2) return ret
def test5(num_obs, do_display=False): # A test in which one of the HMMs has a transition from an input directly to # an output, so it can behave as an epsilon. This node is between two other # nodes in a linear arrangement. # Data generator setup and data generation dimension = 2 obs_gen = make_data_generator(dimension) obs_list = [obs_gen.next() for i in xrange(num_obs)] # GmmMgr setup num_models = 20 models = make_standard_gmms(dimension, num_models) gmm_mgr1 = GmmMgr(models[0:10]) gmm_mgr2 = GmmMgr(models[10:20]) # Hmm setup # Make two Hmms with 2 states and order 2 (self loop, forward 1) The model # in the middle is special in that it can skip directly from the input state # to the output state. seed(0) num_states = 2 hmm0 = make_forward_hmm(gmm_mgr1, num_states, 2, exact=False) hmm1 = Hmm(1) trans = array(((0.0, 0.5, 0.5), (0.0, 0.5, 0.5), (0.0, 0.0, 0.0))) hmm1.build_model(gmm_mgr1, (0,), 1, 1, trans) hmm2 = make_forward_hmm(gmm_mgr1, num_states, 2, exact=False) hmm_mgr = HmmMgr((hmm0, hmm1, hmm2)) # TrainingGraph setup gb = GraphBuilder() node_id0 = gb.new_node((0, 0)) node_id1 = gb.new_node((1, 1)) # node_id2 = gb.new_node((2,2)) arc_id = gb.new_arc(node_id0, node_id1) # arc_id = gb.new_arc(node_id1, node_id2) gr0 = FrozenGraph(gb) tg0 = TrainingGraph(gr0, hmm_mgr, split_prob_dict=dict()) if do_display: tg0.dot_display() tg0.dot_display(expand_hmms=True) valid, ret = validate_training_graph(tg0, gmm_mgr1, hmm_mgr, obs_list, 1, gmm_mgr2) return ret