def _build_confusion_network(self, sampled_da_items):
confusion_net = DialogueActConfusionNetwork()
for da_items, probs in sampled_da_items:
for dai, prob in zip(da_items, probs):
confusion_net.add_merge(prob, dai)
return confusion_net
def _build_confusion_network(self, sampled_da_items):
'''Build confusion network from a list containing DialgoueActItem and their observation probability.'''
confusion_net = DialogueActConfusionNetwork()
for da_items, probs in sampled_da_items:
for dai, prob in zip(da_items, probs):
confusion_net.add_merge(prob, dai)
return confusion_net
def test_get_prob(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.2, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.1, DialogueActItem(dai='inform(food=russian)'))
self.assertAlmostEqual(dacn._get_prob([0, 1, 1]), 0.2 * 0.3 * 0.9)
self.assertAlmostEqual(dacn._get_prob([0, 0, 0]), 0.2 * 0.7 * 0.1)
def test_prune(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.00005, DialogueActItem(dai='inform(food=russian)'))
# Russian food should be pruned.
self.assertEqual(len(dacn), 3)
dacn.prune()
self.assertEqual(len(dacn), 2)
self.assertTrue(not DialogueActItem(
dai='inform(food=russian)') in dacn)
def context_resolution(self, user_da, system_da):
"""Resolves and converts meaning of some user dialogue acts
given the context."""
old_user_da = deepcopy(user_da)
new_user_da = DialogueActConfusionNetwork()
if isinstance(system_da, DialogueAct):
for system_dai in system_da:
for prob, user_dai in user_da:
new_user_dai = None
if system_dai.dat == "confirm" and user_dai.dat == "affirm":
new_user_dai = DialogueActItem("inform",
system_dai.name,
system_dai.value)
elif system_dai.dat == "confirm" and user_dai.dat == "negate":
new_user_dai = DialogueActItem("deny", system_dai.name,
system_dai.value)
elif system_dai.dat == "request" and user_dai.dat == "inform" and \
user_dai.name in self.ontology['context_resolution'] and \
system_dai.name in self.ontology['context_resolution'][user_dai.name] and \
user_dai.value == "dontcare":
new_user_dai = DialogueActItem("inform",
system_dai.name,
system_dai.value)
elif system_dai.dat == "request" and user_dai.dat == "inform" and \
user_dai.name in self.ontology['context_resolution'] and \
system_dai.name in self.ontology['context_resolution'][user_dai.name] and \
self.ontology.slot_has_value(system_dai.name, user_dai.value):
new_user_dai = DialogueActItem("inform",
system_dai.name,
user_dai.value)
elif system_dai.dat == "request" and system_dai.name != "" and \
user_dai.dat == "affirm" and self.ontology.slot_is_binary(system_dai.name):
new_user_dai = DialogueActItem("inform",
system_dai.name, "true")
elif system_dai.dat == "request" and system_dai.name != "" and \
user_dai.dat == "negate" and self.ontology.slot_is_binary(system_dai.name):
new_user_dai = DialogueActItem("inform",
system_dai.name,
"false")
if new_user_dai:
new_user_da.add(prob, new_user_dai)
old_user_da.extend(new_user_da)
return old_user_da
def test_sort(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(1.0, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.00005, DialogueActItem(dai='inform(food=russian)'))
dacn.sort()
cn = list(dacn)
self.assertEqual(cn[0][1], DialogueActItem(dai='inform(food=czech)'))
self.assertEqual(cn[1][1], DialogueActItem(dai='inform(food=chinese)'))
self.assertEqual(cn[2][1], DialogueActItem(dai='inform(food=russian)'))
def test_prune(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.00005, DialogueActItem(dai='inform(food=russian)'))
# Russian food should be pruned.
self.assertEqual(len(dacn), 3)
dacn.prune()
self.assertEqual(len(dacn), 2)
self.assertTrue(not DialogueActItem(dai='inform(food=russian)') in dacn)
def test_add_merge(self):
dai = DialogueActItem(dai='inform(food=chinese)')
dacn = DialogueActConfusionNetwork()
dacn.add_merge(0.5, dai, combine='add')
self.assertEqual(dacn._get_prob([0]), 0.5)
dacn.add_merge(0.5, dai, combine='add')
self.assertEqual(dacn._get_prob([0]), 1.0)
def test_get_prob(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.2, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.1, DialogueActItem(dai='inform(food=russian)'))
self.assertAlmostEqual(dacn._get_prob([0, 1, 1]), 0.2 * 0.3 * 0.9)
self.assertAlmostEqual(dacn._get_prob([0, 0, 0]), 0.2 * 0.7 * 0.1)
def test_sort(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(1.0, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.00005, DialogueActItem(dai='inform(food=russian)'))
dacn.sort()
cn = list(dacn)
self.assertEqual(cn[0][1], DialogueActItem(dai='inform(food=czech)'))
self.assertEqual(cn[1][1], DialogueActItem(dai='inform(food=chinese)'))
self.assertEqual(cn[2][1], DialogueActItem(dai='inform(food=russian)'))
def test_add_merge(self):
dai = DialogueActItem(dai='inform(food=chinese)')
dacn = DialogueActConfusionNetwork()
dacn.add_merge(0.5, dai, combine='add')
self.assertEqual(dacn._get_prob([0]), 0.5)
dacn.add_merge(0.5, dai, combine='add')
self.assertEqual(dacn._get_prob([0]), 1.0)
def _resolve_user_da_in_context(self, user_da, system_da):
"""Resolves and converts meaning of some user dialogue acts
given the context."""
old_user_da = deepcopy(user_da)
new_user_da = DialogueActConfusionNetwork()
if isinstance(system_da, DialogueAct):
for system_dai in system_da:
for prob, user_dai in user_da:
new_user_dai = None
if system_dai.dat == "confirm" and user_dai.dat == "affirm":
new_user_dai = DialogueActItem("inform", system_dai.name, system_dai.value)
elif system_dai.dat == "confirm" and user_dai.dat == "negate":
new_user_dai = DialogueActItem("deny", system_dai.name, system_dai.value)
elif system_dai.dat == "request" and user_dai.dat == "inform" and \
user_dai.name in self.ontology['context_resolution'] and \
system_dai.name in self.ontology['context_resolution'][user_dai.name] and \
user_dai.value == "dontcare":
new_user_dai = DialogueActItem("inform", system_dai.name, system_dai.value)
elif system_dai.dat == "request" and user_dai.dat == "inform" and \
user_dai.name in self.ontology['context_resolution'] and \
system_dai.name in self.ontology['context_resolution'][user_dai.name] and \
self.ontology.slot_has_value(system_dai.name, user_dai.value):
new_user_dai = DialogueActItem("inform", system_dai.name, user_dai.value)
elif system_dai.dat == "request" and system_dai.name != "" and \
user_dai.dat == "affirm" and self.ontology.slot_is_binary(system_dai.name):
new_user_dai = DialogueActItem("inform", system_dai.name, "true")
elif system_dai.dat == "request" and system_dai.name != "" and \
user_dai.dat == "negate" and self.ontology.slot_is_binary(system_dai.name):
new_user_dai = DialogueActItem("inform", system_dai.name, "false")
if new_user_dai:
new_user_da.add(prob, new_user_dai)
old_user_da.merge(new_user_da, combine='max')
return old_user_da
def _infer_last_talked_about_slots(self, user_da, system_da):
"""This adds dialogue act items to support inference of the last slots the user talked about."""
old_user_da = deepcopy(user_da)
new_user_da = DialogueActConfusionNetwork()
colliding_slots = {}
done_slots = set()
for prob, user_dai in user_da:
new_user_dais = []
lta_tsvs = self.ontology.last_talked_about(user_dai.dat, user_dai.name, user_dai.value)
for name, value in lta_tsvs:
new_user_dais.append(DialogueActItem("inform", name, value))
if name in done_slots:
if not name in colliding_slots:
colliding_slots[name] = set()
colliding_slots[name].add(value)
else:
done_slots.add(name)
if new_user_dais:
for nudai in new_user_dais:
if not nudai in new_user_da:
new_user_da.add(prob, nudai)
# In case of collisions, prefer the current last talked about values if it is one of the colliding values.
# If there is a collision and the current last talked about value is not among the colliding values, do not
# consider the colliding DA's at all.
invalid_das = set()
for prob, da in set(new_user_da):
if da.name in colliding_slots and self[da.name].mpv() in colliding_slots[da.name]:
if not da.value == self[da.name].mpv():
invalid_das.add(da)
elif da.name in colliding_slots:
invalid_das.add(da)
for invalid_da in invalid_das:
new_user_da.remove(invalid_da)
old_user_da.merge(new_user_da, combine='max')
return old_user_da
def last_talked_about(self, user_da, system_da):
"""This adds dialogue act items to support inference of the last slots the user talked about."""
old_user_da = deepcopy(user_da)
new_user_da = DialogueActConfusionNetwork()
for prob, user_dai in user_da:
new_user_dais = []
lta_tsvs = self.ontology.last_talked_about(user_dai.dat, user_dai.name, user_dai.value)
for name, value in lta_tsvs:
new_user_dais.append(DialogueActItem("inform", name, value))
if new_user_dais:
for nudai in new_user_dais:
new_user_da.add(prob, nudai)
old_user_da.extend(new_user_da)
return old_user_da
def parse_nblist(self, obs, *args, **kwargs):
"""
Parses an observation featuring an utterance n-best list using the
parse_1_best method.
Arguments:
obs -- a dictionary of observations
:: observation type -> observed value
where observation type is one of values for `obs_type' used in
`ft_props', and observed value is the corresponding observed
value for the input
args -- further positional arguments that should be passed to the
`parse_1_best' method call
kwargs -- further keyword arguments that should be passed to the
`parse_1_best' method call
"""
nblist = obs['utt_nbl']
if len(nblist) == 0:
return DialogueActConfusionNetwork()
obs_wo_nblist = copy.deepcopy(obs)
del obs_wo_nblist['utt_nbl']
dacn_list = []
for prob, utt in nblist:
if "_other_" == utt:
dacn = DialogueActConfusionNetwork()
dacn.add(1.0, DialogueActItem("other"))
elif "_silence_" == utt:
dacn = DialogueActConfusionNetwork()
dacn.add(1.0, DialogueActItem("silence"))
else:
obs_wo_nblist['utt'] = utt
dacn = self.parse_1_best(obs_wo_nblist, *args, **kwargs)
dacn_list.append((prob, dacn))
dacn = merge_slu_confnets(dacn_list)
dacn.prune()
dacn.sort()
return dacn
def last_talked_about(self, user_da, system_da):
"""This adds dialogue act items to support inference of the last slots the user talked about."""
old_user_da = deepcopy(user_da)
new_user_da = DialogueActConfusionNetwork()
for prob, user_dai in user_da:
new_user_dais = []
lta_tsvs = self.ontology.last_talked_about(user_dai.dat,
user_dai.name,
user_dai.value)
for name, value in lta_tsvs:
new_user_dais.append(DialogueActItem("inform", name, value))
if new_user_dais:
for nudai in new_user_dais:
new_user_da.add(prob, nudai)
old_user_da.extend(new_user_da)
return old_user_da
def test_get_platform_res_da(self):
hdc_policy = self._build_policy()
state = DeterministicDiscriminativeDialogueState(self.cfg, self.ontology)
system_input = DialogueActConfusionNetwork()
res = hdc_policy.get_da(state)
user_input = DialogueActConfusionNetwork()
user_input.add(1.0, DialogueActItem(dai='info(task=find_platform)'))
user_input.add(1.0, DialogueActItem(dai='inform(from_stop=Praha)'))
user_input.add(1.0, DialogueActItem(dai='inform(to_stop=Brno)'))
state.update(user_input, system_input)
res = hdc_policy.get_da(state)
self.assert_('inform(not_supported)' in res)
def main():
# initialize tracker and state
slots = ["food", "location"]
tr = DSTCTracker(slots)
state = DSTCState(slots)
state.pprint()
# try to update state with some information
print '---'
cn = DialogueActConfusionNetwork()
cn.add(0.3, DialogueActItem("inform", "food", "chinese"))
cn.add(0.1, DialogueActItem("inform", "food", "indian"))
tr.update_state(state, cn)
state.pprint()
# try to deny some information
print '---'
cn.add(0.9, DialogueActItem("deny", "food", "chinese"))
cn.add(0.1, DialogueActItem("deny", "food", "indian"))
tr.update_state(state, cn)
state.pprint()
def main():
# initialize tracker and state
slots = ["food", "location"]
tr = DSTCTracker(slots)
state = DSTCState(slots)
state.pprint()
# try to update state with some information
print '---'
cn = DialogueActConfusionNetwork()
cn.add(0.3, DialogueActItem("inform", "food", "chinese"))
cn.add(0.1, DialogueActItem("inform", "food", "indian"))
tr.update_state(state, cn)
state.pprint()
# try to deny some information
print '---'
cn.add(0.9, DialogueActItem("deny", "food", "chinese"))
cn.add(0.1, DialogueActItem("deny", "food", "indian"))
tr.update_state(state, cn)
state.pprint()
def _infer_last_talked_about_slots(self, user_da, system_da):
"""This adds dialogue act items to support inference of the last slots the user talked about."""
old_user_da = deepcopy(user_da)
new_user_da = DialogueActConfusionNetwork()
colliding_slots = {}
done_slots = set()
for prob, user_dai in user_da:
new_user_dais = []
lta_tsvs = self.ontology.last_talked_about(user_dai.dat,
user_dai.name,
user_dai.value)
for name, value in lta_tsvs:
new_user_dais.append(DialogueActItem("inform", name, value))
if name in done_slots:
if not name in colliding_slots:
colliding_slots[name] = set()
colliding_slots[name].add(value)
else:
done_slots.add(name)
if new_user_dais:
for nudai in new_user_dais:
if not nudai in new_user_da:
new_user_da.add(prob, nudai)
# In case of collisions, prefer the current last talked about values if it is one of the colliding values.
# If there is a collision and the current last talked about value is not among the colliding values, do not
# consider the colliding DA's at all.
invalid_das = set()
for prob, da in set(new_user_da):
if da.name in colliding_slots and self[
da.name].mpv() in colliding_slots[da.name]:
if not da.value == self[da.name].mpv():
invalid_das.add(da)
elif da.name in colliding_slots:
invalid_das.add(da)
for invalid_da in invalid_das:
new_user_da.remove(invalid_da)
old_user_da.merge(new_user_da, combine='max')
return old_user_da
def test_switching_tasks(self):
hdc_policy = self._build_policy()
self.mox.StubOutWithMock(hdc_policy.weather, 'get_weather')
self.mox.StubOutWithMock(hdc_policy, 'get_directions')
hdc_policy.weather.get_weather(city=u'Praha',
daily=False,
lat=u'50.0755381',
lon=u'14.4378005',
time=None).AndReturn(None)
hdc_policy.get_directions(mox.IgnoreArg(),
check_conflict=True).AndReturn([DialogueActItem(dai="inform(time=10:00)")])
self.mox.ReplayAll()
state = DeterministicDiscriminativeDialogueState(self.cfg, self.ontology)
system_input = DialogueActConfusionNetwork()
res = hdc_policy.get_da(state)
# User says she wants weather so the task should be weather.
user_input = self._build_user_input("inform(task=weather)")
state.update(user_input, system_input)
res = hdc_policy.get_da(state)
self.assertEqual(state['lta_task'].mpv(), 'weather')
# User wants to find a connection so the task should be find_connection.
user_input = self._build_user_input(u"inform(task=find_connection)",
u"inform(to_stop=Malostranská)",
u"inform(from_stop=Anděl)")
state.update(user_input, system_input)
res = hdc_policy.get_da(state)
self.assertEqual(state['lta_task'].mpv(), 'find_connection')
self.mox.VerifyAll()
def parse_nblist(self, obs, *args, **kwargs):
"""
Parses an observation featuring an utterance n-best list using the
parse_1_best method.
Arguments:
obs -- a dictionary of observations
:: observation type -> observed value
where observation type is one of values for `obs_type' used in
`ft_props', and observed value is the corresponding observed
value for the input
args -- further positional arguments that should be passed to the
`parse_1_best' method call
kwargs -- further keyword arguments that should be passed to the
`parse_1_best' method call
"""
nblist = obs['utt_nbl']
if len(nblist) == 0:
return DialogueActConfusionNetwork()
obs_wo_nblist = copy.deepcopy(obs)
del obs_wo_nblist['utt_nbl']
dacn_list = []
for prob, utt in nblist:
if "_other_" == utt:
dacn = DialogueActConfusionNetwork()
dacn.add(1.0, DialogueActItem("other"))
elif "_silence_" == utt:
dacn = DialogueActConfusionNetwork()
dacn.add(1.0, DialogueActItem("silence"))
else:
obs_wo_nblist['utt'] = utt
dacn = self.parse_1_best(obs_wo_nblist, *args, **kwargs)
dacn_list.append((prob, dacn))
dacn = merge_slu_confnets(dacn_list)
dacn.prune()
dacn.sort()
return dacn
def test_session_logger(self):
cfg = Config.load_configs(config=CONFIG_DICT, use_default=False)
sl = SessionLogger()
# test 3 calls at once
for i in range(3):
sess_dir = "./%d" % i
if not os.path.isdir(sess_dir):
os.mkdir(sess_dir)
sl.session_start(sess_dir)
sl.config('config = ' + unicode(cfg))
sl.header(cfg['Logging']["system_name"], cfg['Logging']["version"])
sl.input_source("voip")
sl.dialogue_rec_start(None, "both_complete_dialogue.wav")
sl.dialogue_rec_start("system", "system_complete_dialogue.wav")
sl.dialogue_rec_start("user", "user_complete_dialogue.wav")
sl.dialogue_rec_end("both_complete_dialogue.wav")
sl.dialogue_rec_end("system_complete_dialogue.wav")
sl.dialogue_rec_end("user_complete_dialogue.wav")
sl.turn("system")
sl.dialogue_act("system", "hello()")
sl.text("system", "Hello.")
sl.rec_start("system", "system1.wav")
sl.rec_end("system1.wav")
sl.turn("user")
sl.rec_start("user", "user1.wav")
sl.rec_end("user1.wav")
A1, A2, A3 = 0.90, 0.05, 0.05
B1, B2, B3 = 0.70, 0.20, 0.10
C1, C2, C3 = 0.80, 0.10, 0.10
asr_confnet = UtteranceConfusionNetwork()
asr_confnet.add([[A1, "want"], [A2, "has"], [A3, 'ehm']])
asr_confnet.add([[B1, "Chinese"], [B2, "English"], [B3, 'cheap']])
asr_confnet.add([[C1, "restaurant"], [C2, "pub"], [C3, 'hotel']])
asr_confnet.merge()
asr_confnet.normalise()
asr_confnet.sort()
asr_nblist = asr_confnet.get_utterance_nblist()
sl.asr("user", "user1.wav", asr_nblist, asr_confnet)
slu_confnet = DialogueActConfusionNetwork()
slu_confnet.add(0.7, DialogueActItem('hello'))
slu_confnet.add(0.6, DialogueActItem('thankyou'))
slu_confnet.add(0.4, DialogueActItem('restart'))
slu_confnet.add(0.1, DialogueActItem('bye'))
slu_confnet.merge()
slu_confnet.normalise()
slu_confnet.sort()
slu_nblist = slu_confnet.get_da_nblist()
sl.slu("user", "user1.wav", slu_nblist, slu_confnet)
sl.turn("system")
sl.dialogue_act("system", "thankyou()")
sl.text("system", "Thank you.", cost = 1.0)
sl.rec_start("system", "system2.wav")
sl.rec_end("system2.wav")
sl.barge_in("system", tts_time = True)
sl.turn("user")
sl.rec_start("user", "user2.wav")
sl.rec_end("user2.wav")
sl.hangup("user")
def test_session_logger(self):
cfg = Config.load_configs(config=CONFIG_DICT, use_default=False)
sl = SessionLogger()
# test 3 calls at once
for i in range(3):
sess_dir = "./%d" % i
if not os.path.isdir(sess_dir):
os.mkdir(sess_dir)
sl.session_start(sess_dir)
sl.config('config = ' + unicode(cfg))
sl.header(cfg['Logging']["system_name"], cfg['Logging']["version"])
sl.input_source("voip")
sl.dialogue_rec_start(None, "both_complete_dialogue.wav")
sl.dialogue_rec_start("system", "system_complete_dialogue.wav")
sl.dialogue_rec_start("user", "user_complete_dialogue.wav")
sl.dialogue_rec_end("both_complete_dialogue.wav")
sl.dialogue_rec_end("system_complete_dialogue.wav")
sl.dialogue_rec_end("user_complete_dialogue.wav")
sl.turn("system")
sl.dialogue_act("system", "hello()")
sl.text("system", "Hello.")
sl.rec_start("system", "system1.wav")
sl.rec_end("system1.wav")
sl.turn("user")
sl.rec_start("user", "user1.wav")
sl.rec_end("user1.wav")
A1, A2, A3 = 0.90, 0.05, 0.05
B1, B2, B3 = 0.70, 0.20, 0.10
C1, C2, C3 = 0.80, 0.10, 0.10
asr_confnet = UtteranceConfusionNetwork()
asr_confnet.add([[A1, "want"], [A2, "has"], [A3, 'ehm']])
asr_confnet.add([[B1, "Chinese"], [B2, "English"], [B3, 'cheap']])
asr_confnet.add([[C1, "restaurant"], [C2, "pub"], [C3, 'hotel']])
asr_confnet.merge()
asr_confnet.normalise()
asr_confnet.sort()
asr_nblist = asr_confnet.get_utterance_nblist()
sl.asr("user", "user1.wav", asr_nblist, asr_confnet)
slu_confnet = DialogueActConfusionNetwork()
slu_confnet.add(0.7, DialogueActItem('hello'))
slu_confnet.add(0.6, DialogueActItem('thankyou'))
slu_confnet.add(0.4, DialogueActItem('restart'))
slu_confnet.add(0.1, DialogueActItem('bye'))
slu_confnet.merge()
slu_confnet.normalise()
slu_confnet.sort()
slu_nblist = slu_confnet.get_da_nblist()
sl.slu("user", "user1.wav", slu_nblist, slu_confnet)
sl.turn("system")
sl.dialogue_act("system", "thankyou()")
sl.text("system", "Thank you.", cost=1.0)
sl.rec_start("system", "system2.wav")
sl.rec_end("system2.wav")
sl.barge_in("system", tts_time=True)
sl.turn("user")
sl.rec_start("user", "user2.wav")
sl.rec_end("user2.wav")
sl.hangup("user")
def test_merge_slu_confnets(self):
confnet1 = DialogueActConfusionNetwork()
confnet1.add(0.7, DialogueActItem('hello'))
confnet1.add(0.2, DialogueActItem('bye'))
confnet2 = DialogueActConfusionNetwork()
confnet2.add(0.6, DialogueActItem('hello'))
confnet2.add(0.3, DialogueActItem('restart'))
confnets = [[0.7, confnet1], [0.3, confnet2]]
merged_confnets = merge_slu_confnets(confnets)
correct_merged_confnet = DialogueActConfusionNetwork()
correct_merged_confnet.add_merge(0.7 * 0.7,
DialogueActItem('hello'),
combine='add')
correct_merged_confnet.add_merge(0.7 * 0.2,
DialogueActItem('bye'),
combine='add')
correct_merged_confnet.add_merge(0.3 * 0.6,
DialogueActItem('hello'),
combine='add')
correct_merged_confnet.add_merge(0.3 * 0.3,
DialogueActItem('restart'),
combine='add')
s = []
s.append("")
s.append("Merged confnets:")
s.append(unicode(merged_confnets))
s.append("")
s.append("Correct merged results:")
s.append(unicode(correct_merged_confnet))
s.append("")
print '\n'.join(s)
self.assertEqual(unicode(merged_confnets),
unicode(correct_merged_confnet))
def main():
from alex.utils.config import Config
from alex.utils.caminfodb import CamInfoDb
# This implicitly loads also the default config.
cfg = Config.load_configs(['resources/lz.cfg'], project_root=True)
db_cfg = cfg['DM']["PUfalRuleDM"]['db_cfg'] # database provider
db = CamInfoDb(db_cfg)
pdm = PRuleDM(cfg, db)
pdm.new_dialogue()
pdm.da_out()
# user's input
cn = DialogueActConfusionNetwork()
cn.add(0.7, DialogueActItem(dai="inform(food=chinese)"))
cn.add(0.2, DialogueActItem(dai="inform(food=indian)"))
cn.add(0.5, DialogueActItem(dai="inform(food=chinese)"))
cn.add(0.1, DialogueActItem(dai="inform(food=czech)"))
cn.add(0.1, DialogueActItem(dai="confirm(food=czech)"))
cn.add(0.6, DialogueActItem(dai="request(phone)"))
cn.add(0.3, DialogueActItem(dai="reset()"))
cn.add(0.3, DialogueActItem(dai="asdf()"))
cn.add(0.3, DialogueActItem(dai="reset()"))
print cn
pdm.da_in(cn)
pdm.da_out()
cn = DialogueActConfusionNetwork()
cn.add(0.99, DialogueActItem(dai="confirm(food=indian)"))
print cn
pdm.da_in(cn)
pdm.da_out()
cn = DialogueActConfusionNetwork()
cn.add(0.77, DialogueActItem(dai="reqalts()"))
print cn
pdm.da_in(cn)
pdm.da_out()
cn = DialogueActConfusionNetwork()
cn.add(0.77, DialogueActItem(dai="reqalts()"))
print cn
pdm.da_in(cn)
pdm.da_out()
cn = DialogueActConfusionNetwork()
cn.add(0.99, DialogueActItem(dai="confirm(food=indian)"))
print cn
pdm.da_in(cn)
pdm.da_out()
cn = DialogueActConfusionNetwork()
cn.add(0.99, DialogueActItem(dai="request(name)"))
cn.add(0.99, DialogueActItem(dai="request(food)"))
print cn
pdm.da_in(cn)
pdm.da_out()
cn = DialogueActConfusionNetwork()
cn.add(0.99, DialogueActItem(dai="bye()"))
print cn
pdm.da_in(cn)
pdm.da_out()
def parse_X(self, utterance, verbose=False):
if verbose:
print '=' * 120
print 'Parsing X'
print '-' * 120
print unicode(utterance)
if self.preprocessing:
utterance = self.preprocessing.normalise(utterance)
utterance_fvcs = self.get_fvc(utterance)
if verbose:
print unicode(utterance)
print unicode(utterance_fvcs)
da_confnet = DialogueActConfusionNetwork()
for clser in self.trained_classifiers:
if verbose:
print "Using classifier: ", unicode(clser)
if self.parsed_classifiers[clser].value and self.parsed_classifiers[
clser].value.startswith('CL_'):
# process abstracted classifiers
for f, v, c in utterance_fvcs:
cc = "CL_" + c.upper()
if self.parsed_classifiers[clser].value == cc:
#print clser, f, v, c
classifiers_features = self.get_features(
utterance, (f, v, cc), utterance_fvcs)
classifiers_inputs = np.zeros(
(1, len(self.classifiers_features_mapping[clser])))
classifiers_inputs[
0] = classifiers_features.get_feature_vector(
self.classifiers_features_mapping[clser])
#if verbose:
# print classifiers_features
# print self.classifiers_features_mapping[clser]
p = self.trained_classifiers[clser].predict_proba(
classifiers_inputs)
if verbose:
print ' Probability:', p
dai = DialogueActItem(
self.parsed_classifiers[clser].dat,
self.parsed_classifiers[clser].name, v)
da_confnet.add_merge(p[0][1], dai, combine='max')
else:
# process concrete classifiers
classifiers_features = self.get_features(
utterance, (None, None, None), utterance_fvcs)
classifiers_inputs = np.zeros(
(1, len(self.classifiers_features_mapping[clser])))
classifiers_inputs[
0] = classifiers_features.get_feature_vector(
self.classifiers_features_mapping[clser])
#if verbose:
# print classifiers_features
# print self.classifiers_features_mapping[clser]
p = self.trained_classifiers[clser].predict_proba(
classifiers_inputs)
if verbose:
print ' Probability:', p
dai = self.parsed_classifiers[clser]
da_confnet.add_merge(p[0][1], dai, combine='max')
da_confnet.sort().prune()
return da_confnet
def parse_1_best(self, obs, verbose=False, *args, **kwargs):
"""Parse an utterance into a dialogue act.
:rtype DialogueActConfusionNetwork
"""
utterance = obs['utt']
if isinstance(utterance, UtteranceHyp):
# Parse just the utterance and ignore the confidence score.
utterance = utterance.utterance
if verbose:
print 'Parsing utterance "{utt}".'.format(utt=utterance)
res_cn = DialogueActConfusionNetwork()
dict_da = self.utt2da.get(unicode(utterance), None)
if dict_da:
for dai in DialogueAct(dict_da):
res_cn.add(1.0, dai)
return res_cn
utterance = self.preprocessing.normalise_utterance(utterance)
abutterance, category_labels = self.abstract_utterance(utterance)
if verbose:
print 'After preprocessing: "{utt}".'.format(utt=abutterance)
print category_labels
self.parse_non_speech_events(utterance, res_cn)
utterance = utterance.replace_all(['_noise_'], '').replace_all(['_laugh_'], '').replace_all(['_ehm_hmm_'], '').replace_all(['_inhale_'], '')
abutterance = abutterance.replace_all(['_noise_'], '').replace_all(['_laugh_'], '').replace_all(['_ehm_hmm_'], '').replace_all(['_inhale_'], '')
abutterance = self.handle_false_abstractions(abutterance)
category_labels.add('CITY')
category_labels.add('VEHICLE')
category_labels.add('NUMBER')
if len(res_cn) == 0:
if 'STOP' in category_labels:
self.parse_stop(abutterance, res_cn)
if 'CITY' in category_labels:
self.parse_city(abutterance, res_cn)
if 'NUMBER' in category_labels:
self.parse_number(abutterance)
if any([word.startswith("TIME") for word in abutterance]):
category_labels.add('TIME')
if 'TIME' in category_labels:
self.parse_time(abutterance, res_cn)
if 'DATE_REL' in category_labels:
self.parse_date_rel(abutterance, res_cn)
if 'AMPM' in category_labels:
self.parse_ampm(abutterance, res_cn)
if 'VEHICLE' in category_labels:
self.parse_vehicle(abutterance, res_cn)
if 'TASK' in category_labels:
self.parse_task(abutterance, res_cn)
self.parse_meta(utterance, res_cn)
res_cn.merge()
return res_cn
def test_get_best_nonnull_da(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.2, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.1, DialogueActItem(dai='inform(food=russian)'))
da_nn = dacn.get_best_nonnull_da()
self.assertEqual(len(da_nn), 1)
self.assertEqual(da_nn.dais[0],
DialogueActItem(dai='inform(food=czech)'))
dacn = DialogueActConfusionNetwork()
dacn.add(0.075, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='null()'))
dacn.add(0.15, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.075, DialogueActItem(dai='inform(food=russian)'))
da_nn = dacn.get_best_nonnull_da()
self.assertEqual(len(da_nn), 1)
self.assertEqual(da_nn.dais[0],
DialogueActItem(dai='inform(food=czech)'))
def _build_user_input(self, *args):
user_input = DialogueActConfusionNetwork()
for arg in args:
user_input.add(1.0, DialogueActItem(dai=arg))
return user_input
def process_pending_commands(self):
"""Process all pending commands.
Available commands:
stop() - stop processing and exit the process
flush() - flush input buffers.
Now it only flushes the input connection.
Return True if the process should terminate.
"""
while self.commands.poll():
command = self.commands.recv()
if self.cfg['DM']['debug']:
self.cfg['Logging']['system_logger'].debug(command)
if isinstance(command, Command):
if command.parsed['__name__'] == 'stop':
return True
if command.parsed['__name__'] == 'flush':
# discard all data in in input buffers
while self.slu_hypotheses_in.poll():
data_in = self.slu_hypotheses_in.recv()
self.dm.end_dialogue()
self.commands.send(Command("flushed()", 'DM', 'HUB'))
return False
if command.parsed['__name__'] == 'new_dialogue':
self.epilogue_state = None
self.dm.new_dialogue()
self.cfg['Logging']['session_logger'].turn("system")
self.dm.log_state()
# I should generate the first DM output
da = self.dm.da_out()
if self.cfg['DM']['debug']:
s = []
s.append("DM Output")
s.append("-"*60)
s.append(unicode(da))
s.append("")
s = '\n'.join(s)
self.cfg['Logging']['system_logger'].debug(s)
self.cfg['Logging']['session_logger'].dialogue_act("system", da)
self.commands.send(DMDA(da, 'DM', 'HUB'))
return False
if command.parsed['__name__'] == 'end_dialogue':
self.dm.end_dialogue()
return False
if command.parsed['__name__'] == 'timeout':
# check whether there is a looong silence
# if yes then inform the DM
silence_time = command.parsed['silence_time']
cn = DialogueActConfusionNetwork()
cn.add(1.0, DialogueActItem('silence','time', silence_time))
# process the input DA
self.dm.da_in(cn)
self.cfg['Logging']['session_logger'].turn("system")
self.dm.log_state()
if self.epilogue_state and float(silence_time) > 5.0:
# a user was silent for too long, therefore hung up
self.cfg['Logging']['session_logger'].dialogue_act("system", self.epilogue_da)
self.commands.send(DMDA(self.epilogue_da, 'DM', 'HUB'))
self.commands.send(Command('hangup()', 'DM', 'HUB'))
else:
da = self.dm.da_out()
if self.cfg['DM']['debug']:
s = []
s.append("DM Output")
s.append("-"*60)
s.append(unicode(da))
s.append("")
s = '\n'.join(s)
self.cfg['Logging']['system_logger'].debug(s)
self.cfg['Logging']['session_logger'].dialogue_act("system", da)
self.commands.send(DMDA(da, 'DM', 'HUB'))
if da.has_dat("bye"):
self.commands.send(Command('hangup()', 'DM', 'HUB'))
return False
return False
def test_get_da_nblist(self):
# Simple case with one good hypothesis.
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.05, DialogueActItem(dai='inform(food=russian)'))
nblist = dacn.get_da_nblist()
best_da = nblist.get_best_da()
expected_da = DialogueAct(da_str='inform(food=czech)')
self.assertEqual(best_da, expected_da)
# More good hypotheses
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
nblist = dacn.get_da_nblist()
best_da = nblist.get_best_da()
expected_da = DialogueAct(da_str='inform(food=czech)&inform(food=russian)')
self.assertEqual(best_da, expected_da)
def test_get_da_nblist(self):
# Simple case with one good hypothesis.
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.05, DialogueActItem(dai='inform(food=russian)'))
nblist = dacn.get_da_nblist()
best_da = nblist.get_best_da()
expected_da = DialogueAct(da_str='inform(food=czech)')
self.assertEqual(best_da, expected_da)
# More good hypotheses
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
nblist = dacn.get_da_nblist()
best_da = nblist.get_best_da()
expected_da = DialogueAct(
da_str='inform(food=czech)&inform(food=russian)')
self.assertEqual(best_da, expected_da)
def test_get_best_da_hyp(self):
# Test case when only one dai should be included in the hyp.
dacn = DialogueActConfusionNetwork()
dacn.add(0.2, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.1, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(use_log=False)
self.assertAlmostEqual(best_hyp.prob, 0.8 * 0.7 * 0.9)
self.assertEqual(len(best_hyp.da), 1)
# Test case when 2 dais should be included in the hyp.
dacn = DialogueActConfusionNetwork()
dacn.add(0.1, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(use_log=False)
self.assertAlmostEqual(best_hyp.prob, 0.9 * 0.7 * 0.9)
self.assertEqual(len(best_hyp.da), 2)
# Test the case with logarithms.
dacn = DialogueActConfusionNetwork()
dacn.add(0.1, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(use_log=True)
self.assertAlmostEqual(best_hyp.prob, math.log(0.9 * 0.7 * 0.9))
self.assertEqual(len(best_hyp.da), 2)
# Test the case with manual thresholds.
dacn = DialogueActConfusionNetwork()
dacn.add(0.1, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(
use_log=True, threshold=0.1, thresholds={
DialogueActItem(dai='inform(food=chinese)'): 0.5,
DialogueActItem(dai='inform(food=czech)'): 0.9,
DialogueActItem(dai='inform(food=russian)'): 0.5
})
# Test food=czech should NOT be included.
self.assertAlmostEqual(best_hyp.prob, math.log(0.9 * 0.3 * 0.9))
self.assertEqual(len(best_hyp.da), 1)
self.assertTrue(not DialogueActItem(dai='inform(food=czech)') in best_hyp.da)
dacn = DialogueActConfusionNetwork()
dacn.add(0.1, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(
use_log=True, threshold=0.1, thresholds={
DialogueActItem(dai='inform(food=chinese)'): 0.5,
DialogueActItem(dai='inform(food=czech)'): 0.5,
DialogueActItem(dai='inform(food=russian)'): 0.5
})
# Test food=czech should be included.
self.assertAlmostEqual(best_hyp.prob, math.log(0.9 * 0.7 * 0.9))
self.assertEqual(len(best_hyp.da), 2)
self.assertTrue(DialogueActItem(dai='inform(food=czech)') in best_hyp.da)
def test_get_best_nonnull_da(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.2, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.1, DialogueActItem(dai='inform(food=russian)'))
da_nn = dacn.get_best_nonnull_da()
self.assertEqual(len(da_nn), 1)
self.assertEqual(da_nn.dais[0], DialogueActItem(dai='inform(food=czech)'))
dacn = DialogueActConfusionNetwork()
dacn.add(0.075, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='null()'))
dacn.add(0.15, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.075, DialogueActItem(dai='inform(food=russian)'))
da_nn = dacn.get_best_nonnull_da()
self.assertEqual(len(da_nn), 1)
self.assertEqual(da_nn.dais[0], DialogueActItem(dai='inform(food=czech)'))
def test_merge(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.00005, DialogueActItem(dai='inform(food=russian)'))
dacn.merge(dacn, combine='max')
# Russian food should be pruned.
dacn.sort().prune()
self.assertTrue(not DialogueActItem(
dai='inform(food=russian)') in dacn)
def test_merge_slu_confnets(self):
confnet1 = DialogueActConfusionNetwork()
confnet1.add(0.7, DialogueActItem('hello'))
confnet1.add(0.2, DialogueActItem('bye'))
confnet2 = DialogueActConfusionNetwork()
confnet2.add(0.6, DialogueActItem('hello'))
confnet2.add(0.3, DialogueActItem('restart'))
confnets = [[0.7, confnet1], [0.3, confnet2]]
merged_confnets = merge_slu_confnets(confnets)
correct_merged_confnet = DialogueActConfusionNetwork()
correct_merged_confnet.add_merge(0.7 * 0.7, DialogueActItem('hello'),
combine='add')
correct_merged_confnet.add_merge(0.7 * 0.2, DialogueActItem('bye'),
combine='add')
correct_merged_confnet.add_merge(0.3 * 0.6, DialogueActItem('hello'),
combine='add')
correct_merged_confnet.add_merge(0.3 * 0.3, DialogueActItem('restart'),
combine='add')
s = []
s.append("")
s.append("Merged confnets:")
s.append(unicode(merged_confnets))
s.append("")
s.append("Correct merged results:")
s.append(unicode(correct_merged_confnet))
s.append("")
self.assertEqual(unicode(merged_confnets), unicode(correct_merged_confnet))
def test_get_best_da_hyp(self):
# Test case when only one dai should be included in the hyp.
dacn = DialogueActConfusionNetwork()
dacn.add(0.2, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.1, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(use_log=False)
self.assertAlmostEqual(best_hyp.prob, 0.8 * 0.7 * 0.9)
self.assertEqual(len(best_hyp.da), 1)
# Test case when 2 dais should be included in the hyp.
dacn = DialogueActConfusionNetwork()
dacn.add(0.1, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(use_log=False)
self.assertAlmostEqual(best_hyp.prob, 0.9 * 0.7 * 0.9)
self.assertEqual(len(best_hyp.da), 2)
# Test the case with logarithms.
dacn = DialogueActConfusionNetwork()
dacn.add(0.1, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(use_log=True)
self.assertAlmostEqual(best_hyp.prob, math.log(0.9 * 0.7 * 0.9))
self.assertEqual(len(best_hyp.da), 2)
# Test the case with manual thresholds.
dacn = DialogueActConfusionNetwork()
dacn.add(0.1, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(
use_log=True,
threshold=0.1,
thresholds={
DialogueActItem(dai='inform(food=chinese)'): 0.5,
DialogueActItem(dai='inform(food=czech)'): 0.9,
DialogueActItem(dai='inform(food=russian)'): 0.5
})
# Test food=czech should NOT be included.
self.assertAlmostEqual(best_hyp.prob, math.log(0.9 * 0.3 * 0.9))
self.assertEqual(len(best_hyp.da), 1)
self.assertTrue(not DialogueActItem(
dai='inform(food=czech)') in best_hyp.da)
dacn = DialogueActConfusionNetwork()
dacn.add(0.1, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.7, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=russian)'))
best_hyp = dacn.get_best_da_hyp(
use_log=True,
threshold=0.1,
thresholds={
DialogueActItem(dai='inform(food=chinese)'): 0.5,
DialogueActItem(dai='inform(food=czech)'): 0.5,
DialogueActItem(dai='inform(food=russian)'): 0.5
})
# Test food=czech should be included.
self.assertAlmostEqual(best_hyp.prob, math.log(0.9 * 0.7 * 0.9))
self.assertEqual(len(best_hyp.da), 2)
self.assertTrue(
DialogueActItem(dai='inform(food=czech)') in best_hyp.da)
def parse_1_best(self, obs, verbose=False):
"""Parse an utterance into a dialogue act."""
utterance = obs['utt']
if isinstance(utterance, UtteranceHyp):
# Parse just the utterance and ignore the confidence score.
utterance = utterance.utterance
# print 'Parsing utterance "{utt}".'.format(utt=utterance)
if verbose:
print 'Parsing utterance "{utt}".'.format(utt=utterance)
if self.preprocessing:
# the text normalisation
utterance = self.preprocessing.normalise_utterance(utterance)
abutterance, category_labels = self.abstract_utterance(utterance)
if verbose:
print 'After preprocessing: "{utt}".'.format(utt=abutterance)
print category_labels
else:
category_labels = dict()
# handle false positive alarms of abstraction
abutterance = abutterance.replace(('STOP=Metra',), ('metra',))
abutterance = abutterance.replace(('STOP=Nádraží',), ('nádraží',))
abutterance = abutterance.replace(('STOP=SME',), ('sme',))
abutterance = abutterance.replace(('STOP=Bílá Hora', 'STOP=Železniční stanice',), ('STOP=Bílá Hora', 'železniční stanice',))
abutterance = abutterance.replace(('TIME=now','bych', 'chtěl'), ('teď', 'bych', 'chtěl'))
abutterance = abutterance.replace(('STOP=Čím','se'), ('čím', 'se',))
abutterance = abutterance.replace(('STOP=Lužin','STOP=Na Chmelnici',), ('STOP=Lužin','na','STOP=Chmelnici',))
abutterance = abutterance.replace(('STOP=Konečná','zastávka'), ('konečná', 'zastávka',))
abutterance = abutterance.replace(('STOP=Konečná','STOP=Anděl'), ('konečná', 'STOP=Anděl',))
abutterance = abutterance.replace(('STOP=Konečná stanice','STOP=Ládví'), ('konečná', 'stanice', 'STOP=Ládví',))
abutterance = abutterance.replace(('STOP=Výstupní', 'stanice', 'je'), ('výstupní', 'stanice', 'je'))
abutterance = abutterance.replace(('STOP=Nová','jiné'), ('nové', 'jiné',))
abutterance = abutterance.replace(('STOP=Nová','spojení'), ('nové', 'spojení',))
abutterance = abutterance.replace(('STOP=Nová','zadání'), ('nové', 'zadání',))
abutterance = abutterance.replace(('STOP=Nová','TASK=find_connection'), ('nový', 'TASK=find_connection',))
abutterance = abutterance.replace(('z','CITY=Liberk',), ('z', 'CITY=Liberec',))
abutterance = abutterance.replace(('do','CITY=Liberk',), ('do', 'CITY=Liberec',))
abutterance = abutterance.replace(('pauza','hrozně','STOP=Dlouhá',), ('pauza','hrozně','dlouhá',))
abutterance = abutterance.replace(('v','STOP=Praga',), ('v', 'CITY=Praha',))
abutterance = abutterance.replace(('na','STOP=Praga',), ('na', 'CITY=Praha',))
abutterance = abutterance.replace(('po','STOP=Praga', 'ale'), ('po', 'CITY=Praha',))
abutterance = abutterance.replace(('jsem','v','STOP=Metra',), ('jsem', 'v', 'VEHICLE=metro',))
category_labels.add('CITY')
category_labels.add('VEHICLE')
# print 'After preprocessing: "{utt}".'.format(utt=abutterance)
# print category_labels
res_cn = DialogueActConfusionNetwork()
self.parse_non_speech_events(utterance, res_cn)
if len(res_cn) == 0:
# remove non speech events, they are not relevant for SLU
abutterance = abutterance.replace_all('_noise_', '').replace_all('_laugh_', '').replace_all('_ehm_hmm_', '').replace_all('_inhale_', '')
if 'STOP' in category_labels:
self.parse_stop(abutterance, res_cn)
if 'CITY' in category_labels:
self.parse_city(abutterance, res_cn)
if 'TIME' in category_labels:
self.parse_time(abutterance, res_cn)
if 'DATE_REL' in category_labels:
self.parse_date_rel(abutterance, res_cn)
if 'AMPM' in category_labels:
self.parse_ampm(abutterance, res_cn)
if 'VEHICLE' in category_labels:
self.parse_vehicle(abutterance, res_cn)
if 'TASK' in category_labels:
self.parse_task(abutterance, res_cn)
self.parse_meta(utterance, res_cn)
res_cn.merge()
return res_cn
def test_normalise(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(1.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.00005, DialogueActItem(dai='inform(food=russian)'))
self.assertRaises(ConfusionNetworkException, dacn.normalise)
def parse_1_best(self, obs, verbose=False):
"""Parse an utterance into a dialogue act."""
utterance = obs['utt']
if isinstance(utterance, UtteranceHyp):
# Parse just the utterance and ignore the confidence score.
utterance = utterance.utterance
# print 'Parsing utterance "{utt}".'.format(utt=utterance)
if verbose:
print 'Parsing utterance "{utt}".'.format(utt=utterance)
if self.preprocessing:
# the text normalisation
utterance = self.preprocessing.normalise_utterance(utterance)
abutterance, category_labels = self.abstract_utterance(utterance)
if verbose:
print 'After preprocessing: "{utt}".'.format(utt=abutterance)
print category_labels
else:
category_labels = dict()
# handle false positive alarms of abstraction
abutterance = abutterance.replace(('STOP=Metra', ), ('metra', ))
abutterance = abutterance.replace(('STOP=Nádraží', ), ('nádraží', ))
abutterance = abutterance.replace(('STOP=SME', ), ('sme', ))
abutterance = abutterance.replace((
'STOP=Bílá Hora',
'STOP=Železniční stanice',
), (
'STOP=Bílá Hora',
'železniční stanice',
))
abutterance = abutterance.replace(('TIME=now', 'bych', 'chtěl'),
('teď', 'bych', 'chtěl'))
abutterance = abutterance.replace(('STOP=Čím', 'se'), (
'čím',
'se',
))
abutterance = abutterance.replace((
'STOP=Lužin',
'STOP=Na Chmelnici',
), (
'STOP=Lužin',
'na',
'STOP=Chmelnici',
))
abutterance = abutterance.replace(('STOP=Konečná', 'zastávka'), (
'konečná',
'zastávka',
))
abutterance = abutterance.replace(('STOP=Konečná', 'STOP=Anděl'), (
'konečná',
'STOP=Anděl',
))
abutterance = abutterance.replace(
('STOP=Konečná stanice', 'STOP=Ládví'), (
'konečná',
'stanice',
'STOP=Ládví',
))
abutterance = abutterance.replace(('STOP=Výstupní', 'stanice', 'je'),
('výstupní', 'stanice', 'je'))
abutterance = abutterance.replace(('STOP=Nová', 'jiné'), (
'nové',
'jiné',
))
abutterance = abutterance.replace(('STOP=Nová', 'spojení'), (
'nové',
'spojení',
))
abutterance = abutterance.replace(('STOP=Nová', 'zadání'), (
'nové',
'zadání',
))
abutterance = abutterance.replace(
('STOP=Nová', 'TASK=find_connection'), (
'nový',
'TASK=find_connection',
))
abutterance = abutterance.replace((
'z',
'CITY=Liberk',
), (
'z',
'CITY=Liberec',
))
abutterance = abutterance.replace((
'do',
'CITY=Liberk',
), (
'do',
'CITY=Liberec',
))
abutterance = abutterance.replace((
'pauza',
'hrozně',
'STOP=Dlouhá',
), (
'pauza',
'hrozně',
'dlouhá',
))
abutterance = abutterance.replace((
'v',
'STOP=Praga',
), (
'v',
'CITY=Praha',
))
abutterance = abutterance.replace((
'na',
'STOP=Praga',
), (
'na',
'CITY=Praha',
))
abutterance = abutterance.replace(('po', 'STOP=Praga', 'ale'), (
'po',
'CITY=Praha',
))
abutterance = abutterance.replace((
'jsem',
'v',
'STOP=Metra',
), (
'jsem',
'v',
'VEHICLE=metro',
))
category_labels.add('CITY')
category_labels.add('VEHICLE')
# print 'After preprocessing: "{utt}".'.format(utt=abutterance)
# print category_labels
res_cn = DialogueActConfusionNetwork()
self.parse_non_speech_events(utterance, res_cn)
if len(res_cn) == 0:
# remove non speech events, they are not relevant for SLU
abutterance = abutterance.replace_all('_noise_', '').replace_all(
'_laugh_', '').replace_all('_ehm_hmm_',
'').replace_all('_inhale_', '')
if 'STOP' in category_labels:
self.parse_stop(abutterance, res_cn)
if 'CITY' in category_labels:
self.parse_city(abutterance, res_cn)
if 'TIME' in category_labels:
self.parse_time(abutterance, res_cn)
if 'DATE_REL' in category_labels:
self.parse_date_rel(abutterance, res_cn)
if 'AMPM' in category_labels:
self.parse_ampm(abutterance, res_cn)
if 'VEHICLE' in category_labels:
self.parse_vehicle(abutterance, res_cn)
if 'TASK' in category_labels:
self.parse_task(abutterance, res_cn)
self.parse_meta(utterance, res_cn)
res_cn.merge()
return res_cn
def parse_1_best(self, obs, verbose=False, *args, **kwargs):
"""Parse an utterance into a dialogue act.
:rtype DialogueActConfusionNetwork
"""
utterance = obs['utt']
if isinstance(utterance, UtteranceHyp):
# Parse just the utterance and ignore the confidence score.
utterance = utterance.utterance
if verbose:
print 'Parsing utterance "{utt}".'.format(utt=utterance)
res_cn = DialogueActConfusionNetwork()
dict_da = self.utt2da.get(unicode(utterance), None)
if dict_da:
for dai in DialogueAct(dict_da):
res_cn.add(1.0, dai)
return res_cn
utterance = self.preprocessing.normalise_utterance(utterance)
abutterance, category_labels = self.abstract_utterance(utterance)
if verbose:
print 'After preprocessing: "{utt}".'.format(utt=abutterance)
print category_labels
self.parse_non_speech_events(utterance, res_cn)
utterance = utterance.replace_all(['_noise_'], '').replace_all(
['_laugh_'], '').replace_all(['_ehm_hmm_'],
'').replace_all(['_inhale_'], '')
abutterance = abutterance.replace_all(['_noise_'], '').replace_all(
['_laugh_'], '').replace_all(['_ehm_hmm_'],
'').replace_all(['_inhale_'], '')
abutterance = self.handle_false_abstractions(abutterance)
category_labels.add('CITY')
category_labels.add('VEHICLE')
category_labels.add('NUMBER')
if len(res_cn) == 0:
if 'STOP' in category_labels:
self.parse_stop(abutterance, res_cn)
if 'CITY' in category_labels:
self.parse_city(abutterance, res_cn)
if 'NUMBER' in category_labels:
self.parse_number(abutterance)
if any([word.startswith("TIME") for word in abutterance]):
category_labels.add('TIME')
if 'TIME' in category_labels:
self.parse_time(abutterance, res_cn)
if 'DATE_REL' in category_labels:
self.parse_date_rel(abutterance, res_cn)
if 'AMPM' in category_labels:
self.parse_ampm(abutterance, res_cn)
if 'VEHICLE' in category_labels:
self.parse_vehicle(abutterance, res_cn)
if 'TASK' in category_labels:
self.parse_task(abutterance, res_cn)
self.parse_meta(utterance, res_cn)
res_cn.merge()
return res_cn
def process_pending_commands(self):
"""Process all pending commands.
Available commands:
stop() - stop processing and exit the process
flush() - flush input buffers.
Now it only flushes the input connection.
Return True if the process should terminate.
"""
while self.commands.poll():
command = self.commands.recv()
if self.cfg['DM']['debug']:
self.cfg['Logging']['system_logger'].debug(command)
if isinstance(command, Command):
#Thanh:
if command.parsed['__name__'] == 'print_log_dir':
print '===***===session-log-dir:', command.source
if command.parsed['__name__'] == 'stop':
return True
if command.parsed['__name__'] == 'flush':
# discard all data in in input buffers
while self.slu_hypotheses_in.poll():
data_in = self.slu_hypotheses_in.recv()
self.dm.end_dialogue()
self.commands.send(Command("flushed()", 'DM', 'HUB'))
return False
#if command.parsed['__name__'] == 'prepare_new_dialogue':
#self.dm.new_dialogue()
if command.parsed['__name__'] == 'new_dialogue':
self.dm.new_dialogue()#thanh change???
self.epilogue_state = None
self.cfg['Logging']['session_logger'].turn("system")
self.dm.log_state()
# I should generate the first DM output
da = self.dm.da_out()
if self.cfg['DM']['debug']:
s = []
s.append("DM Output")
s.append("-"*60)
s.append(unicode(da))
s.append("")
s = '\n'.join(s)
self.cfg['Logging']['system_logger'].debug(s)
self.cfg['Logging']['session_logger'].dialogue_act("system", da)
self.commands.send(DMDA(da, 'DM', 'HUB'))
return False
if command.parsed['__name__'] == 'end_dialogue':
self.dm.end_dialogue()
return False
if command.parsed['__name__'] == 'timeout':
# check whether there is a looong silence
# if yes then inform the DM
silence_time = command.parsed['silence_time']
cn = DialogueActConfusionNetwork()
cn.add(1.0, DialogueActItem('silence','time', silence_time))
# process the input DA
self.dm.da_in(cn)
self.cfg['Logging']['session_logger'].turn("system")
self.dm.log_state()
print '----Time out: ', self.epilogue_state, silence_time
'''Thanh
if self.epilogue_state == 'give_code':
# an cant_apply act have been chosen
self.cfg['Logging']['session_logger'].dialogue_act("system", self.epilogue_da)
self.commands.send(DMDA(self.epilogue_da, 'DM', 'HUB'))
self.commands.send(Command('hangup()', 'DM', 'HUB'))
return False
#'''
if self.epilogue_state and float(silence_time) > 5.0:
if self.epilogue_state == 'final_question': # and self.final_question_repeated<16:
da = DialogueAct('say(text="{text}")'.format(text="Sorry, did you get the correct information?"))
#self.final_question_repeated += 1
self.cfg['Logging']['session_logger'].dialogue_act("system", da)
self.commands.send(DMDA(da, 'DM', 'HUB'))
else:
# a user was silent for too long, therefore hung up
self.cfg['Logging']['session_logger'].dialogue_act("system", self.epilogue_da)
self.commands.send(DMDA(self.epilogue_da, 'DM', 'HUB'))
self.commands.send(Command('hangup()', 'DM', 'HUB'))
else:
da = self.dm.da_out()
if self.cfg['DM']['debug']:
s = []
s.append("DM Output")
s.append("-"*60)
s.append(unicode(da))
s.append("")
s = '\n'.join(s)
self.cfg['Logging']['system_logger'].debug(s)
self.cfg['Logging']['session_logger'].dialogue_act("system", da)
self.commands.send(DMDA(da, 'DM', 'HUB'))
if da.has_dat("bye"):
self.commands.send(Command('hangup()', 'DM', 'HUB'))
return False
return False
def test_make_from_da(self):
da = DialogueAct('inform(food=czech)&inform(area=north)')
dacn = DialogueActConfusionNetwork.make_from_da(da)
self.assertEqual(dacn.get_best_da(), da)
def test_merge(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(0.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.00005, DialogueActItem(dai='inform(food=russian)'))
dacn.merge(dacn, combine='max')
# Russian food should be pruned.
dacn.sort().prune()
self.assertTrue(not DialogueActItem(dai='inform(food=russian)') in dacn)
def process_pending_commands(self):
"""Process all pending commands.
Available commands:
stop() - stop processing and exit the process
flush() - flush input buffers.
Now it only flushes the input connection.
Return True if the process should terminate.
"""
while self.commands.poll():
command = self.commands.recv()
if self.cfg['DM']['debug']:
self.cfg['Logging']['system_logger'].debug(command)
if isinstance(command, Command):
if command.parsed['__name__'] == 'stop':
return True
if command.parsed['__name__'] == 'flush':
# discard all data in in input buffers
while self.slu_hypotheses_in.poll():
data_in = self.slu_hypotheses_in.recv()
self.dm.end_dialogue()
self.commands.send(Command("flushed()", 'DM', 'HUB'))
return False
if command.parsed['__name__'] == 'new_dialogue':
self.epilogue_state = None
self.dm.new_dialogue()
self.cfg['Logging']['session_logger'].turn("system")
self.dm.log_state()
# I should generate the first DM output
da = self.dm.da_out()
if self.cfg['DM']['debug']:
s = []
s.append("DM Output")
s.append("-"*60)
s.append(unicode(da))
s.append("")
s = '\n'.join(s)
self.cfg['Logging']['system_logger'].debug(s)
self.cfg['Logging']['session_logger'].dialogue_act("system", da)
self.commands.send(DMDA(da, 'DM', 'HUB'))
return False
if command.parsed['__name__'] == 'end_dialogue':
self.dm.end_dialogue()
return False
if command.parsed['__name__'] == 'timeout':
# check whether there is a looong silence
# if yes then inform the DM
silence_time = command.parsed['silence_time']
cn = DialogueActConfusionNetwork()
cn.add(1.0, DialogueActItem('silence','time', silence_time))
# process the input DA
self.dm.da_in(cn)
self.cfg['Logging']['session_logger'].turn("system")
self.dm.log_state()
if self.epilogue_state and float(silence_time) > 5.0:
# a user was silent for too long, therefore hung up
self.cfg['Logging']['session_logger'].dialogue_act("system", self.epilogue_da)
self.commands.send(DMDA(self.epilogue_da, 'DM', 'HUB'))
self.commands.send(Command('hangup()', 'DM', 'HUB'))
else:
da = self.dm.da_out()
if self.cfg['DM']['debug']:
s = []
s.append("DM Output")
s.append("-"*60)
s.append(unicode(da))
s.append("")
s = '\n'.join(s)
self.cfg['Logging']['system_logger'].debug(s)
self.cfg['Logging']['session_logger'].dialogue_act("system", da)
self.commands.send(DMDA(da, 'DM', 'HUB'))
if da.has_dat("bye"):
self.commands.send(Command('hangup()', 'DM', 'HUB'))
return False
return False
def test_make_from_da(self):
da = DialogueAct('inform(food=czech)&inform(area=north)')
dacn = DialogueActConfusionNetwork.make_from_da(da)
self.assertEqual(dacn.get_best_da(), da)
def parse_X(self, utterance, verbose=False):
if verbose:
print '='*120
print 'Parsing X'
print '-'*120
print unicode(utterance)
if self.preprocessing:
utterance = self.preprocessing.normalise(utterance)
utterance_fvcs = self.get_fvc(utterance)
if verbose:
print unicode(utterance)
print unicode(utterance_fvcs)
da_confnet = DialogueActConfusionNetwork()
for clser in self.trained_classifiers:
if verbose:
print "Using classifier: ", unicode(clser)
if self.parsed_classifiers[clser].value and self.parsed_classifiers[clser].value.startswith('CL_'):
# process abstracted classifiers
for f, v, c in utterance_fvcs:
cc = "CL_" + c.upper()
if self.parsed_classifiers[clser].value == cc:
#print clser, f, v, c
classifiers_features = self.get_features(utterance, (f, v, cc), utterance_fvcs)
classifiers_inputs = np.zeros((1, len(self.classifiers_features_mapping[clser])))
classifiers_inputs[0] = classifiers_features.get_feature_vector(self.classifiers_features_mapping[clser])
#if verbose:
# print classifiers_features
# print self.classifiers_features_mapping[clser]
p = self.trained_classifiers[clser].predict_proba(classifiers_inputs)
if verbose:
print ' Probability:', p
dai = DialogueActItem(self.parsed_classifiers[clser].dat, self.parsed_classifiers[clser].name, v)
da_confnet.add_merge(p[0][1], dai, combine='max')
else:
# process concrete classifiers
classifiers_features = self.get_features(utterance, (None, None, None), utterance_fvcs)
classifiers_inputs = np.zeros((1, len(self.classifiers_features_mapping[clser])))
classifiers_inputs[0] = classifiers_features.get_feature_vector(self.classifiers_features_mapping[clser])
#if verbose:
# print classifiers_features
# print self.classifiers_features_mapping[clser]
p = self.trained_classifiers[clser].predict_proba(classifiers_inputs)
if verbose:
print ' Probability:', p
dai = self.parsed_classifiers[clser]
da_confnet.add_merge(p[0][1], dai, combine='max')
da_confnet.sort().prune()
return da_confnet
def test_normalise(self):
dacn = DialogueActConfusionNetwork()
dacn.add(0.05, DialogueActItem(dai='inform(food=chinese)'))
dacn.add(1.9, DialogueActItem(dai='inform(food=czech)'))
dacn.add(0.00005, DialogueActItem(dai='inform(food=russian)'))
self.assertRaises(ConfusionNetworkException, dacn.normalise)
