def get_lex_repr(self, word):
"""
Get representation for lexical feature
"""
if not self.embed_lex: ## n-hot representation
n_hot = np.zeros(len(self.dictionary_values))
values = is_in_dict(word, self.dictionary)
if values:
for v in values:
n_hot[self.dictionary_values.index(v)] = 1.0
return n_hot
else:
lex_feats = []
for property in self.dictionary_values:
values = is_in_dict(word, self.dictionary)
if values:
if property in values:
lex_feats.append(
self.lembeds[self.l2i[property]].npvalue())
else:
lex_feats.append(self.lembeds[self.l2i[UNK]].npvalue())
else:
lex_feats.append(
self.lembeds[self.l2i[UNK]].npvalue()) # unknown word
return np.concatenate(lex_feats)
def save_wordlex_map(self, out_filename):
"""
save final word-to-lexicon-embedding map to file
:param out_filename: filename
"""
# construct reverse mapping using word embeddings
i2wl = {self.w2i[w]: w for w in self.w2i.keys()}
OUT = open(out_filename+".wlmap.emb","w")
for word_id in i2wl.keys():
word = i2wl[word_id]
lex_feats = []
for property in self.dictionary_values:
values = is_in_dict(word, self.dictionary)
if values:
if property in values:
lex_feats.append(property)
else:
lex_feats.append(UNK)
else:
lex_feats.append(UNK) # unknown word
OUT.write("{} {}\n".format(word," ".join([str(x) for x in lex_feats])))
OUT.close()
def save_wordlex_map(self, out_filename):
"""
save final word-to-lexicon-embedding map to file
:param out_filename: filename
"""
# construct reverse mapping using word embeddings
i2wl = {self.w2i[w]: w for w in self.w2i.keys()}
OUT = open(out_filename + ".wlmap.emb", "w")
for word_id in i2wl.keys():
word = i2wl[word_id]
lex_feats = []
for property in self.dictionary_values:
values = is_in_dict(word, self.dictionary)
if values:
if property in values:
lex_feats.append(property)
else:
lex_feats.append(UNK)
else:
lex_feats.append(UNK) # unknown word
OUT.write("{} {}\n".format(word,
" ".join([str(x) for x in lex_feats])))
OUT.close()
def get_lex_repr(self, word):
"""
Get representation for lexical feature
"""
if not self.embed_lex: ## n-hot representation
n_hot = np.zeros(len(self.dictionary_values))
values = is_in_dict(word, self.dictionary)
if values:
for v in values:
n_hot[self.dictionary_values.index(v)] = 1.0
return n_hot
else:
lex_feats = []
for property in self.dictionary_values:
values = is_in_dict(word, self.dictionary)
if values:
if property in values:
lex_feats.append(self.lembeds[self.l2i[property]].npvalue())
else:
lex_feats.append(self.lembeds[self.l2i[UNK]].npvalue())
else:
lex_feats.append(self.lembeds[self.l2i[UNK]].npvalue()) # unknown word
return np.concatenate(lex_feats)