text stringlengths 1 93.6k |
|---|
sample = [sample_ids[idx] for sample_ids in samples]
|
if data_utils.EOS_ID in sample:
|
sample = sample[:sample.index(data_utils.EOS_ID)]
|
ans = " ".join([tf.compat.as_str(rev_vocab[output]) for output in sample])
|
if grammar_critic.check_ans(sentence.split(), ans):
|
sample_correct += 1
|
if sentence.strip() not in hist_samples:
|
hist_samples[sentence.strip()] = []
|
if (ans,'T') not in hist_samples[sentence.strip()]:
|
hist_samples[sentence.strip()].append((ans,'T'))
|
coverage += 1.0 / grammar_critic.possible_ans_num(sentence.split())
|
sample_count += 1
|
print('argmax accuracy: {}'.format(float(argmax_correct) / argmax_count))
|
print('sample accuracy: {}'.format(float(sample_correct) / sample_count))
|
print('sample coverage: {}'.format(coverage / len(tmp_encoder_inputs)))
|
print('number of inputs (regardless to batch):{}'.format(len(tmp_encoder_inputs)))
|
for key, value in hist_samples.items():
|
fout.write("{}:{}\n".format(key, value))
|
else:
|
sys.stdout.write('> ')
|
sys.stdout.flush()
|
sentence = sys.stdin.readline()
|
while sentence:
|
if sentence.strip() == 'exit()':
|
break
|
# step
|
token_ids = data_utils.sentence_to_token_ids(tf.compat.as_bytes(sentence), vocab, normalize_digits=False)
|
encoder_pad = [data_utils.PAD_ID] * (_buckets[-1][0] - len(token_ids))
|
encoder_lens = [len(token_ids)]
|
token_ids = list(reversed(token_ids)) + encoder_pad
|
encoder_inputs = []
|
for idx in token_ids:
|
encoder_inputs.append([idx])
|
decoder_inputs = [[data_utils.GO_ID]]
|
if FLAGS.test_type == 'realtime_argmax':#TODO: check WHO don't have argmax?
|
outs, log_prob = model.dynamic_decode(sess, encoder_inputs, encoder_lens, \
|
decoder_inputs)
|
#outs = [int(np.argmax(logit, axis=1)) for logit in outs]
|
outputs = [output_ids[0] for output_ids in outs]
|
#prob_outputs = [prob[0] for prob in probs]
|
if data_utils.EOS_ID in outputs:
|
idx = outputs.index(data_utils.EOS_ID)
|
#outputs = outputs[:outputs.index(data_utils.EOS_ID)]
|
outputs = outputs[:idx]
|
#prob_outputs = prob_outputs[:idx]
|
print(" ".join([tf.compat.as_str(rev_vocab[output]) for output in outputs]))
|
#print(prob_outputs)
|
print("log prob %.6f" % log_prob[0])
|
elif FLAGS.test_type == 'realtime_beam_search':
|
outs, log_prob = beam_search(encoder_inputs, encoder_lens)
|
for b in range(len(outs[0])):
|
outputs = [output_ids[b] for output_ids in outs]
|
if data_utils.EOS_ID in outputs:
|
outputs = outputs[1:outputs.index(data_utils.EOS_ID)]
|
print(" ".join([tf.compat.as_str(rev_vocab[output]) for output in outputs]))
|
print("log prob %.6f" % log_prob[b])
|
elif FLAGS.test_type == 'realtime_MMI':
|
outs, log_prob = beam_search(encoder_inputs, encoder_lens)
|
lm_log_prob, lens = MMI(outs)
|
MMI_scores = np.array(log_prob) - 0.5 * np.array(lm_log_prob) + 0.5 * np.array(lens)
|
MMI_idx = np.argsort(MMI_scores)
|
for b in [MMI_idx[-1]]:#range(len(outs[0])):
|
outputs = [output_ids[b] for output_ids in outs]
|
if data_utils.EOS_ID in outputs:
|
outputs = outputs[1:outputs.index(data_utils.EOS_ID)]
|
print(" ".join([tf.compat.as_str(rev_vocab[output]) for output in outputs]))
|
print("log prob %.6f" % log_prob[b])
|
print("general log prob(w/o condition) %.6f" % lm_log_prob[b])
|
print("sentence lenth %d" % lens[b])
|
elif FLAGS.test_type == 'realtime_sample':
|
samples, log_prob = model.dynamic_decode(sess, encoder_inputs, encoder_lens, \
|
decoder_inputs, mode='sample')
|
sample = [sample_ids[0] for sample_ids in samples]
|
if data_utils.EOS_ID in sample:
|
sample = sample[:sample.index(data_utils.EOS_ID)]
|
print(sample)
|
print(" ".join([tf.compat.as_str(rev_vocab[output]) for output in sample]))
|
print("log prob %.6f" % log_prob)
|
sys.stdout.write('> ')
|
sys.stdout.flush()
|
sentence = sys.stdin.readline()
|
# <FILESEP>
|
from objection_engine.beans.comment import Comment
|
from telegram import Update
|
import re
|
class Message:
|
def __init__(self, update: Update, updater):
|
if (len(update.message.photo) > 0):
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.