def __call__(self, raw_article_sents, raw_clusters):
self._net.eval()
n_art = len(raw_article_sents)
articles = conver2id(UNK, self._word2id, raw_article_sents)
clusters = (conver2id(UNK, self._word2id, raw_clusters[0]),
raw_clusters[1], raw_clusters[2])
article = pad_batch_tensorize(articles, PAD, cuda=False,
max_num=5).to(self._device)
clusters = (pad_batch_tensorize(clusters[0],
PAD,
cuda=False,
max_num=4).to(self._device),
pad_batch_tensorize(clusters[1],
PAD,
cuda=False,
max_num=4).to(self._device),
pad_batch_tensorize(clusters[2],
PAD,
cuda=False,
max_num=4).to(self._device))
if raw_clusters == []:
print(clusters)
indices = self._net.extract([article],
clusters,
k=min(n_art, self._max_ext))
return indices
def __call__(self, raw_article_sents, raw_query):
self._net.eval()
n_art = len(raw_article_sents)
articles = conver2id(UNK, self._word2id, raw_article_sents)
queries = conver2id(UNK, self._word2id, raw_query)
article = pad_batch_tensorize(articles, PAD, cuda=False
).to(self._device)
query = pad_batch_tensorize(queries, PAD, cuda=False
).to(self._device)
indices = self._net.extract([article], k=min(n_art, self._max_ext), queries=[query])
return indices
def __call__(self, raw_article_sents):
self._net.eval()
n_art = len(raw_article_sents)
articles = conver2id(UNK, self._word2id, raw_article_sents)
article = pad_batch_tensorize(articles, PAD, cuda=False
).to(self._device)
indices = self._net.extract([article], k=min(n_art, self._max_ext))
return indices
def _prepro(self, raw_article_sents):
ext_word2id = dict(self._word2id)
ext_id2word = dict(self._id2word)
for raw_words in raw_article_sents:
for w in raw_words:
if not w in ext_word2id:
ext_word2id[w] = len(ext_word2id)
ext_id2word[len(ext_id2word)] = w
articles = conver2id(UNK, self._word2id, raw_article_sents)
art_lens = [len(art) for art in articles]
article = pad_batch_tensorize(articles, PAD, cuda=False).to(self._device)
extend_arts = conver2id(UNK, ext_word2id, raw_article_sents)
extend_art = pad_batch_tensorize(extend_arts, PAD, cuda=False).to(self._device)
extend_vsize = len(ext_word2id)
dec_args = (article, art_lens, extend_art, extend_vsize,
START, END, UNK, self._max_len)
return dec_args, ext_id2word
def __call__(self, raw_article_sents, sent_labels):
self._net.eval()
n_art = len(raw_article_sents)
articles = conver2id(UNK, self._word2id, raw_article_sents)
article = pad_batch_tensorize(articles, PAD,
cuda=False).to(self._device)
indices = self._net.extract([article], [sent_labels])
return indices
def _prepro(self, raw_article_sents):
ext_word2id = dict(self._word2id)
ext_id2word = dict(self._id2word)
for raw_words in raw_article_sents:
for w in raw_words:
if not w in ext_word2id:
ext_word2id[w] = len(ext_word2id)
ext_id2word[len(ext_id2word)] = w
articles = conver2id(UNK, self._word2id, raw_article_sents)
art_lens = [len(art) for art in articles]
article = pad_batch_tensorize(articles, PAD, cuda=False
).to(self._device)
extend_arts = conver2id(UNK, ext_word2id, raw_article_sents)
extend_art = pad_batch_tensorize(extend_arts, PAD, cuda=False
).to(self._device)
extend_vsize = len(ext_word2id)
dec_args = (article, art_lens, extend_art, extend_vsize,
START, END, UNK, self._max_len)
return dec_args, ext_id2word
def __call__(self, raw_article_sents, raw_abs_sents=None):
if self.net_type == 'ml_rnn_extractor':
articles = conver2id(UNK, self._word2id, raw_article_sents)
article = pad_batch_tensorize(articles, PAD, cuda=False
).to(self._device)
elif self.net_type == 'ml_trans_rnn_extractor':
# print([" ".join(r) for r in raw_article_sents])
# print([" ".join(r) for r in raw_abs_sents])
article = myextract.get_batch_trans([([" ".join(r) for r in raw_article_sents], [" ".join(r) for r in raw_abs_sents])])
return article
def __call__(self, raw_article_sents):
self._net.eval()
n_art = len(raw_article_sents)
articles = conver2id(UNK, self._word2id, raw_article_sents)
article = pad_batch_tensorize(articles, PAD, cuda=False, max_num=5
).to(self._device)
if not self.force_ext:
indices = self._net.extract([article], k=min(n_art, self._max_ext), force_ext=self.force_ext)
else:
indices = self._net.extract([article], k=min(n_art, self._max_ext))
return indices
def _prepro(self, raw_article_sents):
ext_word2id = dict(self._word2id)
ext_id2word = dict(self._id2word)
articles = conver2id(UNK, self._word2id, raw_article_sents)
art_lens = [len(art) for art in articles]
article = pad_batch_tensorize(articles, PAD,
cuda=False).to(self._device)
dec_args = (article, art_lens, START, END, UNK, self._max_len)
return dec_args, ext_id2word
def __call__(self, raw_article_sents):
self._net.eval()
n_art = len(raw_article_sents)
if self._emb_type == 'W2V':
articles = conver2id(UNK, self._word2id, raw_article_sents)
else:
articles = [self._tokenizer.convert_tokens_to_ids(sentence)
for sentence in raw_article_sents]
article = pad_batch_tensorize(articles, PAD, cuda=False
).to(self._device)
indices = self._net.extract([article], k=min(n_art, self._max_ext))
return indices
def __call__(self, raw_article_sents):
self._net.eval()
n_art = len(raw_article_sents)
if self._net_type == 'ml_trans_rnn_extractor':
n_art = len(raw_article_sents[0])
batch = myextract.get_batch_trans([raw_article_sents])
indices = self._net.extract(batch, k=min(n_art, self._max_ext))
return indices, batch
else:
articles = conver2id(UNK, self._word2id, raw_article_sents)
article = pad_batch_tensorize(articles, PAD, cuda=False
).to(self._device)
indices = self._net.extract([article], k=min(n_art, self._max_ext))
return indices
def __call__(self, raw_article_sents):
articles = conver2id(UNK, self._word2id, raw_article_sents)
article = pad_batch_tensorize(articles, PAD,
cuda=False).to(self._device)
return article
def __call__(self, raw_article_sents):
articles = conver2id(UNK, self._word2id, raw_article_sents)
article = pad_batch_tensorize(articles, PAD, cuda=False
).to(self._device)
return article
