def decode(self, src_sentence, trgt_sentence):
self.trgt_sentence = trgt_sentence + [utils.EOS_ID]
self.initialize_predictor(src_sentence)
hypo = PartialHypothesis(self.get_predictor_states(), self.calculate_stats)
while hypo.get_last_word() != utils.EOS_ID:
self._expand_hypo(hypo)
hypo.score = self.get_adjusted_score(hypo)
self.add_full_hypo(hypo.generate_full_hypothesis())
return self.get_full_hypos_sorted()
def decode(self, src_sentence, seed=0):
self.initialize_predictor(src_sentence)
hypos = [
PartialHypothesis(copy.deepcopy(self.get_predictor_states()))
for i in range(self.nbest)
]
t = 0
while hypos and t < self.max_len:
next_hypos = []
for sen_seed, hypo in enumerate(hypos):
if hypo.get_last_word() == utils.EOS_ID:
hypo.score = self.get_adjusted_score(hypo)
self.add_full_hypo(hypo.generate_full_hypothesis())
else:
self._expand_hypo(hypo, seed=seed + sen_seed)
next_hypos.append(hypo)
hypos = next_hypos
t += 1
for hypo in hypos:
hypo.score = self.get_adjusted_score(hypo)
self.add_full_hypo(hypo.generate_full_hypothesis())
return self.get_full_hypos_sorted()
def decode(self, src_sentence):
"""Decodes a single source sentence using A* search. """
self.initialize_predictor(src_sentence)
self.lower_bound = self.get_empty_hypo(
) if self.use_lower_bound else None
self.cur_capacity = self.capacity
open_set = MinMaxHeap(
reserve=self.capacity) if self.capacity > 0 else []
self.push(
open_set, 0.0,
PartialHypothesis(self.get_predictor_states(),
self.calculate_stats))
while open_set:
c, hypo = self.pop(open_set)
if hypo.get_last_word() == utils.EOS_ID: # Found best hypothesis
hypo.score = self.get_adjusted_score(hypo)
self.add_full_hypo(hypo.generate_full_hypothesis())
if len(self.full_hypos
) == self.nbest: # if we have enough hypos
return self.get_full_hypos_sorted()
self.cur_capacity -= 1
continue
if len(hypo) == self.max_len: #discard and continue
continue
for next_hypo in self._expand_hypo(hypo, self.capacity):
score = self.get_adjusted_score(next_hypo)
self.push(open_set, score, next_hypo)
if not self.full_hypos:
self.add_full_hypo(self.lower_bound.generate_full_hypothesis())
return self.get_full_hypos_sorted()
def decode(self, src_sentence):
self.initialize_predictor(src_sentence)
hypothesis = PartialHypothesis(self.get_predictor_states())
while hypothesis.get_last_word() != utils.EOS_ID and len(
hypothesis) < self.max_len:
ids, posterior, original_posterior = self.apply_predictor(
hypothesis if self.gumbel else None, 1)
trgt_word = ids[0]
if self.gumbel:
hypothesis.base_score += original_posterior[0]
hypothesis.score_breakdown.append(original_posterior[0])
else:
hypothesis.score += posterior[0]
hypothesis.score_breakdown.append(posterior[0])
hypothesis.trgt_sentence.append(trgt_word)
self.consume(trgt_word)
self.add_full_hypo(hypothesis.generate_full_hypothesis())
return self.full_hypos
def initialize_order_ds(self):
self.queues = [MinMaxHeap() for k in range(self.max_len + 1)]
self.queue_order = PointerQueue([0.0], reserve=self.max_len)
self.time_sync = defaultdict(lambda: self.beam
if self.beam > 0 else utils.INF)
# Initialize BOS hypothesis
self.queues[0].insert(
(0.0, PartialHypothesis(self.get_predictor_states())))
self.time_sync[0] = 1
def _get_initial_hypos(self):
"""Get the list of initial ``PartialHypothesis``. """
bos_hypo = PartialHypothesis(self.get_predictor_states())
hypos = self._expand_hypo(bos_hypo, self.beam_size)
inds = list(np.cumsum(self.group_sizes))
return [hypos[a:b] for a, b in zip([0] + inds[:-1], inds)]
def _get_initial_hypos(self):
"""Get the list of initial ``PartialHypothesis``. """
return [PartialHypothesis(self.get_predictor_states())]
def _get_initial_hypos(self):
"""Get the list of initial ``PartialHypothesis``. """
return [[PartialHypothesis(copy.deepcopy(self.get_predictor_states()),
self.calculate_stats)] for i in range(self.num_groups)]