def get_copy_dict(vocab):
lines = du.read_data(filepaths['copypp'])
ids = du.convert_to_token_ids(lines, vocab)
copy_dict = {}
for line in ids:
copy_dict[line[0]] = line[1:]
return copy_dict
def predict():
tf.reset_default_graph()
with tf.Session() as sess:
model = load_model(sess)
while True:
sys.stdout.write("Please input a sentence...\n")
sys.stdout.flush()
sentence = sys.stdin.readline()
sentence = sentence.strip().split()
# [B = 1, T]
ids = du.convert_to_token_ids([sentence], vocab_dict)
encoder_inputs = list_transpose(ids)
# outputs是一个包含bucket[bucket[idx]][1]个元素的list，每个元素是一个batch_size * vocab_size的np.array
predict_ids = model.predict(sess, encoder_inputs)
predict_words = [
vocab_list[id[0]] for id in predict_ids[0].tolist()
]
# 截断<EOS>以后的部分
truncate = len(predict_words)
for i in range(truncate):
if predict_words[i] == '<EOS>':
truncate = i
break
predict_sentence = ' '.join(predict_words[:truncate])
print('Predicted paraphrase: %s' % predict_sentence)
def predict_a_lot():
tf.reset_default_graph()
with tf.Session() as sess:
model = load_model(sess)
while True:
sentence = input()
if len(sentence.strip().split()) == 0:
continue
# print(sentence)
sentence = sentence.strip().split()
# [B = 1, T]
ids = du.convert_to_token_ids([sentence], vocab_dict)
encoder_inputs = list_transpose(ids)
# outputs是一个包含bucket[bucket[idx]][1]个元素的list，每个元素是一个batch_size * vocab_size的np.array
predict_ids = model.predict(sess, encoder_inputs)
predict_words = [
vocab_list[id[0]] for id in predict_ids[0].tolist()
]
# 截断<EOS>以后的部分
truncate = len(predict_words)
for i in range(truncate):
if predict_words[i] == '<EOS>':
truncate = i
break
predict_sentence = ' '.join(predict_words[:truncate])
print(predict_sentence)
def predict():
tf.reset_default_graph()
with tf.Session() as sess:
model = load_model(sess, forward_only=True)
batch_size = model.batch_size = 1
while True:
sys.stdout.write("Please input a sentence...\n")
sys.stdout.flush()
sentence = sys.stdin.readline()
sentence = sentence.strip().split()
ids_tmp = du.convert_to_token_ids([sentence], vocab_dict)
ids = ids_tmp[0]
length = len(ids)
bucket_idx = -1
for (i, (bucket_size, _)) in enumerate(buckets):
if (length < bucket_size):
bucket_idx = i
break
if length > buckets[-1][0]:
print("Sorry, this sentence is too long :-(")
else:
enc_inputs_all = [[] for _ in range(len(buckets))]
dec_inputs_all = [[] for _ in range(len(buckets))]
enc_inputs_all[bucket_idx] = [ids]
dec_inputs_all[bucket_idx] = [[]]
encoder_inputs, decoder_inputs, target_weights = get_batch(
enc_inputs_all, dec_inputs_all, bucket_idx, batch_size)
# outputs是一个包含bucket[bucket[idx]][1]个元素的list，每个元素是一个batch_size * vocab_size的np.array
outputs = model.step(sess,
encoder_inputs,
decoder_inputs,
target_weights,
bucket_idx,
forward_only=True)
# 使用greedy decoder
predict_ids = [
int(np.argmax(output, axis=1)) for output in outputs
]
predict_words = [vocab_list[id] for id in predict_ids]
# 截断<EOS>以后的部分
truncate = len(predict_words)
for i in range(truncate):
if predict_words[i] == '<EOS>':
truncate = i
break
predict_sentence = ' '.join(predict_words[:truncate])
print('Predicted paraphrase: %s' % predict_sentence)
