def distance(self, iset, oset):
"""Computes minimum distance.
This method computes, for a pair of input/output strings or acceptors, the
minimum edit distance according to the underlying edit transducer.
Args:
iset: input string or acceptor.
oset: output string or acceptor.
Returns:
Minimum edit distance according to the edit transducer.
"""
lattice = self._create_lattice(iset, oset)
# The shortest cost from all final states to the start state is
# equivalent to the cost of the shortest path.
return float(shortestdistance(lattice, reverse=True)[0])
def distance(self, iexpr: pynini.FstLike, oexpr: pynini.FstLike) -> float:
"""Computes minimum distance.
This method computes, for a pair of input/output strings or acceptors, the
minimum edit distance according to the underlying edit transducer.
Args:
iexpr: input string or acceptor.
oexpr: output string or acceptor.
Returns:
Minimum edit distance according to the edit transducer.
"""
lattice = self.create_lattice(iexpr, oexpr)
# The shortest cost from all final states to the start state is
# equivalent to the cost of the shortest path.
return float(
pynini.shortestdistance(lattice, reverse=True)[lattice.start()])
def disambiguate(self, verbalization):
normalized = {}
replacement_dicts = {}
for verbal_arr in verbalization.paths:
shortest_path = self._language_model_scoring(verbal_arr)
normalized_text = shortest_path.stringify(
token_type=self.word_symbols)
normalized[normalized_text] = pn.shortestdistance(shortest_path)
replacement_dicts[normalized_text] = copy.deepcopy(
self.replacement_dict)
best_normalized = self._lowest_cost(normalized)
#print("Best normalized: " + best_normalized)
if best_normalized in replacement_dicts:
best_normalized = self._insert_original_words(
best_normalized, replacement_dicts[best_normalized])
if self.oov_queue:
#TODO: is this sufficient for multiple verbalizations?
best_normalized = self._insert_original_oov(
best_normalized, self.oov_queue)
self.oov_queue = None
return best_normalized
def best(self):
return pow(e, -float(shortestdistance(self._best)[-1]))
def total_weight(self, fst: pynini.Fst) -> float:
return float(pynini.shortestdistance(fst, reverse=True)[fst.start()])