def decode(self):
start = time.time()
counter = 0
batch = self.batcher.next_batch()
while batch is not None: # and counter <= 100 # 11490
# Run beam search to get best Hypothesis
best_summary = self.beam_search(batch)
# Extract the output ids from the hypothesis and convert back to words
output_ids = [int(t) for t in best_summary.tokens[1:]]
decoded_words = data.outputids2words(
output_ids, self.vocab,
(batch.art_oovs[0] if config.pointer_gen else None))
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index(data.STOP_DECODING)
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
original_abstract_sents = batch.original_abstracts_sents[0]
write_for_rouge(original_abstract_sents, decoded_words, counter,
self._rouge_ref_dir, self._rouge_dec_dir)
counter += 1
if counter % 10 == 0:
print('%d example in %d sec' % (counter, time.time() - start))
start = time.time()
batch = self.batcher.next_batch()
print("Decoder has finished reading dataset for single_pass.")
print("Now starting ROUGE eval...")
results_dict = rouge_eval(self._rouge_ref_dir, self._rouge_dec_dir)
rouge_log(results_dict, self._decode_dir)
def decode(self):
start = time.time()
counter = 0
bleu_scores = []
batch = self.batcher.next_batch()
while batch is not None:
# Run beam search to get best Hypothesis
best_summary = self.beam_search(batch)
import pdb
pdb.set_trace()
# Extract the output ids from the hypothesis and convert back to words
output_ids = [int(t) for t in best_summary.tokens[1:]]
decoded_words = data.outputids2words(
output_ids, self.vocab,
(batch.art_oovs[0] if config.pointer_gen else None))
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index(data.STOP_DECODING)
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
original_abstracts = batch.original_abstracts_sents[0]
reference = original_abstracts[0].strip().split()
bleu = nltk.translate.bleu_score.sentence_bleu([reference],
decoded_words,
weights=(0.5, 0.5))
bleu_scores.append(bleu)
write_for_rouge(original_abstracts, decoded_words, counter,
self._rouge_ref_dir, self._rouge_dec_dir)
counter += 1
if counter % 1000 == 0:
print('%d example in %d sec' % (counter, time.time() - start))
start = time.time()
batch = self.batcher.next_batch()
print('Average BLEU score:', np.mean(bleu_scores))
# uncomment this if you successfully install `pyrouge`
print("Decoder has finished reading dataset for single_pass.")
print("Now starting ROUGE eval...")
results_dict = rouge_eval(self._rouge_ref_dir, self._rouge_dec_dir)
rouge_log(results_dict, self._decode_dir)
import sys
from utils import write_for_rouge, rouge_eval, rouge_log
decode_dir = sys.argv[1]
print("Decoder has finished reading dataset for single_pass.")
print("Now starting ROUGE eval...")
results_dict = rouge_eval(decode_dir + 'rouge_ref',
decode_dir + 'rouge_dec_dir')
rouge_log(results_dict, decode_dir + 'rouge_dec_dir')
def decode(self, data_iter):
start = time.time()
counter = 0
for batch in data_iter:
summ, summ_tag, article, article_extend, summ_lens, article_lens, oov_nums = batch
summ, summ_tag, article, article_extend, summ_lens, article_lens, oov_nums = [
data.repeat(config.beam_size, 1) for data in batch
]
summ_lens, article_lens, oov_nums = summ_lens.squeeze(
), article_lens.squeeze(), oov_nums.squeeze()
if use_cuda and torch.cuda.is_available():
summ, article, article_extend = to_cuda(data=(summ, article,
article_extend))
airticle_words = ids2words(article[0].squeeze()[1:], self.id2word)
# Run beam search to get best Hypothesis
best_summary = self.beam_search(enc_inputs=article,
enc_lens=article_lens,
enc_inputs_extend=article_extend,
oov_nums=oov_nums,
dec_lens=summ_lens)
# Extract the output ids from the hypothesis and convert back to words
output_ids = [int(t) for t in best_summary.tokens[1:]]
# decoded_words = data.outputids2words(output_ids, self.vocab,
# (batch.art_oovs[0] if config.use_pgen else None))
decoded_words = ids2words(output_ids, self.id2word)
# Remove the [STOP] token from decoded_words, if necessary
try:
fst_stop_idx = decoded_words.index('<end>')
decoded_words = decoded_words[:fst_stop_idx]
except ValueError:
decoded_words = decoded_words
original_abstract_sents = ids2words(
summ[0].squeeze()[1:],
self.id2word) # summary repeat beam_size time
decoded_words = "".join(decoded_words)
original_abstract_sents = "".join(original_abstract_sents)
rouge = Rouge()
rouge_score = rouge.get_scores(
decoded_words[:len(original_abstract_sents)],
original_abstract_sents)
rouge_score = rouge.get_scores(str(output_ids),
str(summ[0].squeeze()[1:]))
rouge_1 = rouge_score[0]["rouge-1"]
rouge_2 = rouge_score[0]["rouge-2"]
rouge_l = rouge_score[0]["rouge-l"]
# pyrouge don't support chinese, need to use token
# decoded_words = [str(ids) for ids in output_ids]
# original_abstract_sents = [str(ids) for ids in summ[0].squeeze()[1:].tolist()]
# write_for_rouge(original_abstract_sents, decoded_words, counter,
# self._rouge_ref_dir, self._rouge_dec_dir)
counter += 1
if counter % 1000 == 0:
print('%d example in %d sec' % (counter, time.time() - start))
start = time.time()
print("Decoder has finished reading dataset for single_pass.")
print("Now starting ROUGE eval...")
results_dict = rouge_eval(self._rouge_ref_dir, self._rouge_dec_dir)
rouge_log(results_dict, self._decode_dir)
from utils import rouge_eval, rouge_log
rouge_ref_dir = 'logs/pku_all_vocab/decode_test_50maxenc_4beam_6mindec_50maxdec_ckpt-123230/reference_tmp'
rouge_dec_dir = 'logs/pku_all_vocab/decode_test_50maxenc_4beam_6mindec_50maxdec_ckpt-123230/decoded_tmp'
print("Decoder has finished reading dataset for single_pass.")
print("Now starting ROUGE eval...")
results_dict = rouge_eval(rouge_ref_dir, rouge_dec_dir)
rouge_log(
results_dict,
'logs/pku_all_vocab/decode_test_50maxenc_4beam_6mindec_50maxdec_ckpt-123230/'
)
