def train(self):
self.logger.info(str(self.cfg))
config_path = os.path.join(self.cfg["checkpoint_path"], "config.json")
write_json(config_path, self.cfg)
batch_size = self.cfg["batch_size"]
epochs = self.cfg["epochs"]
train_path = self.cfg["train_set"]
valid_path = self.cfg["valid_set"]
self.n_gold_spans = count_gold_spans(valid_path)
if self.cfg["knn_sampling"] == "knn":
self.knn_ids = h5py.File(
os.path.join(self.cfg["raw_path"], "knn_ids.hdf5"), "r")
valid_batch_size = 1
shuffle = False
else:
valid_batch_size = batch_size
shuffle = True
valid_set = list(
self.batcher.batchnize_dataset(data=valid_path,
data_name="valid",
batch_size=valid_batch_size,
shuffle=shuffle))
best_f1 = -np.inf
init_lr = self.cfg["lr"]
self.log_trainable_variables()
self.logger.info("Start training...")
self._add_summary()
for epoch in range(1, epochs + 1):
self.logger.info('Epoch {}/{}:'.format(epoch, epochs))
train_set = self.batcher.batchnize_dataset(data=train_path,
data_name="train",
batch_size=batch_size,
shuffle=True)
_ = self.train_knn_epoch(train_set, "train")
if self.cfg["use_lr_decay"]: # learning rate decay
self.cfg["lr"] = max(
init_lr / (1.0 + self.cfg["lr_decay"] * epoch),
self.cfg["minimal_lr"])
eval_metrics = self.evaluate_knn_epoch(valid_set, "valid")
cur_valid_f1 = eval_metrics[0]
if cur_valid_f1 > best_f1:
best_f1 = cur_valid_f1
self.save_session(epoch)
self.logger.info(
'-- new BEST F1 on valid set: {:>7.2%}'.format(best_f1))
self.train_writer.close()
self.test_writer.close()
def save_span_representation(self, data_name, preprocessor):
self.logger.info(str(self.cfg))
########################
# Load validation data #
########################
valid_data = preprocessor.load_dataset(
self.cfg["data_path"],
keep_number=True,
lowercase=self.cfg["char_lowercase"])
valid_data = valid_data[:self.cfg["data_size"]]
dataset = preprocessor.build_dataset(valid_data, self.word_dict,
self.char_dict, self.tag_dict)
self.logger.info("Valid sentences: {:>7}".format(len(dataset)))
#############
# Main loop #
#############
start_time = time.time()
gold_labels = {}
fout_path = os.path.join(self.cfg["checkpoint_path"],
"%s.span_reps.hdf5" % data_name)
fout = h5py.File(fout_path, 'w')
print("PREDICTION START")
for record, data in zip(valid_data, dataset):
valid_sent_id = record["sent_id"]
batch = self.make_one_batch_for_target(data, valid_sent_id)
if (valid_sent_id + 1) % 100 == 0:
print("%d" % (valid_sent_id + 1), flush=True, end=" ")
##############
# Prediction #
##############
feed_dict = self._get_feed_dict(batch)
span_reps = self.sess.run([self.span_rep], feed_dict)[0][0]
span_tags = batch["tags"][0]
assert len(span_reps) == len(span_tags)
##################
# Add the result #
##################
fout.create_dataset(name='{}'.format(valid_sent_id),
dtype='float32',
data=span_reps)
gold_labels[valid_sent_id] = [
self.rev_tag_dict[int(tag)] for tag in span_tags
]
fout.close()
path = os.path.join(self.cfg["checkpoint_path"],
"%s.gold_labels.json" % data_name)
write_json(path, gold_labels)
self.logger.info("-- Time: %f seconds\nFINISHED." %
(time.time() - start_time))
def process_data(config):
train_file = os.path.join(config["raw_path"], "train.txt")
dev_file = os.path.join(config["raw_path"], "dev.txt")
ref_file = os.path.join(config["raw_path"], "ref.txt")
asr_file = os.path.join(config["raw_path"], "asr.txt")
if not os.path.exists(config["save_path"]):
os.makedirs(config["save_path"])
# build vocabulary
word_vocab, char_vocab = build_vocab_list([train_file],
config["min_word_count"],
config["min_char_count"],
config["max_vocab_size"])
if not config["use_pretrained"]:
word_dict, char_dict = build_vocabulary(word_vocab, char_vocab)
else:
glove_path = config["glove_path"].format(config["glove_name"],
config["emb_dim"])
glove_vocab = load_glove_vocab(glove_path, config["glove_name"])
glove_vocab = glove_vocab & {word.lower() for word in glove_vocab}
word_vocab = [word for word in word_vocab if word in glove_vocab]
word_dict, char_dict = build_vocabulary(word_vocab, char_vocab)
tmp_word_dict = word_dict.copy()
del tmp_word_dict[UNK], tmp_word_dict[NUM], tmp_word_dict[END]
vectors = filter_glove_emb(tmp_word_dict, glove_path,
config["glove_name"], config["emb_dim"])
np.savez_compressed(config["pretrained_emb"], embeddings=vectors)
# create indices dataset
punct_dict = dict([(punct, idx)
for idx, punct in enumerate(PUNCTUATION_VOCABULARY)])
train_set = build_dataset([train_file], word_dict, char_dict, punct_dict,
config["max_sequence_len"])
dev_set = build_dataset([dev_file], word_dict, char_dict, punct_dict,
config["max_sequence_len"])
ref_set = build_dataset([ref_file], word_dict, char_dict, punct_dict,
config["max_sequence_len"])
asr_set = build_dataset([asr_file], word_dict, char_dict, punct_dict,
config["max_sequence_len"])
vocab = {
"word_dict": word_dict,
"char_dict": char_dict,
"tag_dict": punct_dict
}
# write to file
write_json(config["vocab"], vocab)
write_json(config["train_set"], train_set)
write_json(config["dev_set"], dev_set)
write_json(config["ref_set"], ref_set)
write_json(config["asr_set"], asr_set)
def train(self):
self.logger.info(str(self.cfg))
write_json(os.path.join(self.cfg["checkpoint_path"], "config.json"),
self.cfg)
batch_size = self.cfg["batch_size"]
epochs = self.cfg["epochs"]
train_path = self.cfg["train_set"]
valid_path = self.cfg["valid_set"]
self.n_gold_spans = count_gold_spans(valid_path)
valid_set = list(
self.batcher.batchnize_dataset(valid_path,
batch_size,
shuffle=True))
best_f1 = -np.inf
init_lr = self.cfg["lr"]
self.log_trainable_variables()
self.logger.info("Start training...")
self._add_summary()
for epoch in range(1, epochs + 1):
self.logger.info('Epoch {}/{}:'.format(epoch, epochs))
train_set = self.batcher.batchnize_dataset(train_path,
batch_size,
shuffle=True)
_ = self.train_epoch(train_set)
if self.cfg["use_lr_decay"]: # learning rate decay
self.cfg["lr"] = max(
init_lr / (1.0 + self.cfg["lr_decay"] * epoch),
self.cfg["minimal_lr"])
eval_metrics = self.evaluate_epoch(valid_set, "valid")
cur_valid_f1 = eval_metrics[0]
if cur_valid_f1 > best_f1:
best_f1 = cur_valid_f1
self.save_session(epoch)
self.logger.info(
"-- new BEST F1 on valid set: {:>7.2%}".format(best_f1))
self.train_writer.close()
self.test_writer.close()
def process_data(dataset, config):
dataset["train"] = data_formatted(dataset["train"])
dataset["dev"] = data_formatted(dataset["dev"])
dataset["test"] = data_formatted(dataset["test"])
train_data = load_dataset(dataset["train"], config["task_name"])
dev_data = load_dataset(dataset["dev"], config["task_name"])
test_data = load_dataset(dataset["test"], config["task_name"])
if not os.path.exists(config["save_path"]):
os.makedirs(config["save_path"])
# build vocabulary
if not config["use_pretrained"]:
word_dict = build_word_vocab([train_data, dev_data, test_data])
else:
glove_path = config["glove_path"].format(config["glove_name"],
config["emb_dim"])
glove_vocab = load_glove_vocab(glove_path, config["glove_name"])
word_dict = build_word_vocab_pretrained(
[train_data, dev_data, test_data], glove_vocab)
vectors = filter_glove_emb(word_dict, glove_path, config["glove_name"],
config["emb_dim"])
np.savez_compressed(config["pretrained_emb"], embeddings=vectors)
tag_dict = build_tag_vocab([train_data, dev_data, test_data],
config["task_name"])
# build char dict
train_data = load_dataset(dataset["train"],
config["task_name"],
keep_number=True,
lowercase=config["char_lowercase"])
dev_data = load_dataset(dataset["dev"],
config["task_name"],
keep_number=True,
lowercase=config["char_lowercase"])
test_data = load_dataset(dataset["test"],
config["task_name"],
keep_number=True,
lowercase=config["char_lowercase"])
char_dict = build_char_vocab([train_data, dev_data, test_data])
# create indices dataset
train_set = build_dataset(train_data, word_dict, char_dict, tag_dict)
dev_set = build_dataset(dev_data, word_dict, char_dict, tag_dict)
test_set = build_dataset(test_data, word_dict, char_dict, tag_dict)
vocab = {
"word_dict": word_dict,
"char_dict": char_dict,
"tag_dict": tag_dict
}
write_json(os.path.join(config["save_path"], "vocab.json"), vocab)
write_json(os.path.join(config["save_path"], "train.json"), train_set)
write_json(os.path.join(config["save_path"], "dev.json"), dev_set)
write_json(os.path.join(config["save_path"], "test.json"), test_set)
return (train_set, dev_set, test_set, vocab)
def eval(self, preprocessor):
self.logger.info(str(self.cfg))
data = preprocessor.load_dataset(self.cfg["data_path"],
keep_number=True,
lowercase=self.cfg["char_lowercase"])
data = data[:self.cfg["data_size"]]
dataset = preprocessor.build_dataset(data, self.word_dict,
self.char_dict, self.tag_dict)
write_json(os.path.join(self.cfg["save_path"], "tmp.json"), dataset)
self.n_gold_spans = count_gold_spans(
os.path.join(self.cfg["save_path"], "tmp.json"))
self.logger.info("Target data: %s sentences" % len(dataset))
del dataset
batches = list(
self.batcher.batchnize_dataset(os.path.join(
self.cfg["save_path"], "tmp.json"),
batch_size=self.cfg["batch_size"],
shuffle=True))
self.logger.info("Target data: %s batches" % len(batches))
_ = self.evaluate_epoch(batches, "valid")
def process_data(config):
# load raw data
train_data = load_dataset(os.path.join(config["raw_path"], "train.txt"))
dev_data = load_dataset(os.path.join(config["raw_path"], "valid.txt"))
test_data = load_dataset(os.path.join(config["raw_path"], "test.txt"))
# build vocabulary
word_dict, char_dict, _ = build_vocab([train_data, dev_data])
*_, tag_dict = build_vocab([train_data, dev_data, test_data])
# create indices dataset
train_set = build_dataset(train_data, word_dict, char_dict, tag_dict)
dev_set = build_dataset(dev_data, word_dict, char_dict, tag_dict)
test_set = build_dataset(test_data, word_dict, char_dict, tag_dict)
vocab = {
"word_dict": word_dict,
"char_dict": char_dict,
"tag_dict": tag_dict
}
# write to file
if not os.path.exists(config["save_path"]):
os.makedirs(config["save_path"])
write_json(os.path.join(config["save_path"], "vocab.json"), vocab)
write_json(os.path.join(config["save_path"], "train.json"), train_set)
write_json(os.path.join(config["save_path"], "dev.json"), dev_set)
write_json(os.path.join(config["save_path"], "test.json"), test_set)
def preprocess(self):
config = self.config
os.makedirs(config["save_path"], exist_ok=True)
# List[{'words': List[str], 'tags': List[str]}]
train_data = self.load_dataset(os.path.join(config["raw_path"],
"train.json"),
keep_number=False,
lowercase=True)
valid_data = self.load_dataset(os.path.join(config["raw_path"],
"valid.json"),
keep_number=False,
lowercase=True)
train_data = train_data[:config["data_size"]]
valid_data = valid_data[:config["data_size"]]
# build vocabulary
if config["use_pretrained"]:
glove_path = self.config["glove_path"].format(
config["glove_name"], config["emb_dim"])
glove_vocab = self.load_glove_vocab(glove_path,
config["glove_name"])
word_dict = self.build_word_vocab_pretrained(
[train_data, valid_data], glove_vocab)
vectors = self.filter_glove_emb(word_dict, glove_path,
config["glove_name"],
config["emb_dim"])
np.savez_compressed(config["pretrained_emb"], embeddings=vectors)
else:
word_dict = self.build_word_vocab([train_data, valid_data])
# build tag dict
tag_dict = self.build_tag_vocab([train_data, valid_data])
# build char dict
train_data = self.load_dataset(os.path.join(config["raw_path"],
"train.json"),
keep_number=True,
lowercase=config["char_lowercase"])
valid_data = self.load_dataset(os.path.join(config["raw_path"],
"valid.json"),
keep_number=True,
lowercase=config["char_lowercase"])
train_data = train_data[:config["data_size"]]
valid_data = valid_data[:config["data_size"]]
char_dict = self.build_char_vocab([train_data])
# create indices dataset
# List[{'words': List[str], 'chars': List[List[str]], 'tags': List[str]}]
train_set = self.build_dataset(train_data, word_dict, char_dict,
tag_dict)
valid_set = self.build_dataset(valid_data, word_dict, char_dict,
tag_dict)
vocab = {
"word_dict": word_dict,
"char_dict": char_dict,
"tag_dict": tag_dict
}
print("Train Sents: %d" % len(train_set))
print("Valid Sents: %d" % len(valid_set))
# write to file
write_json(os.path.join(config["save_path"], "vocab.json"), vocab)
write_json(os.path.join(config["save_path"], "train.json"), train_set)
write_json(os.path.join(config["save_path"], "valid.json"), valid_set)
def save_nearest_spans(self, data_name, preprocessor, print_knn):
self.logger.info(str(self.cfg))
########################
# Load validation data #
########################
valid_data = preprocessor.load_dataset(
self.cfg["data_path"],
keep_number=True,
lowercase=self.cfg["char_lowercase"])
valid_data = valid_data[:self.cfg["data_size"]]
dataset = preprocessor.build_dataset(valid_data, self.word_dict,
self.char_dict, self.tag_dict)
dataset_path = os.path.join(self.cfg["save_path"], "tmp.json")
write_json(dataset_path, dataset)
self.logger.info("Valid sentences: {:>7}".format(len(dataset)))
self.n_gold_spans = count_gold_spans(dataset_path)
######################
# Load training data #
######################
train_sents = load_json(self.cfg["train_set"])
if self.cfg["knn_sampling"] == "random":
train_sent_ids = [sent_id for sent_id in range(len(train_sents))]
else:
train_sent_ids = None
train_data = preprocessor.load_dataset(os.path.join(
self.cfg["raw_path"], "train.json"),
keep_number=True,
lowercase=False)
self.logger.info("Train sentences: {:>7}".format(len(train_sents)))
#############
# Main loop #
#############
correct = 0
p_total = 0
start_time = time.time()
file_path = os.path.join(self.cfg["checkpoint_path"],
"%s.nearest_spans.txt" % data_name)
fout_txt = open(file_path, "w")
print("PREDICTION START")
for record, data in zip(valid_data, dataset):
valid_sent_id = record["sent_id"]
batch = self.make_one_batch_for_target(data, valid_sent_id)
if (valid_sent_id + 1) % 100 == 0:
print("%d" % (valid_sent_id + 1), flush=True, end=" ")
#####################
# Sentence sampling #
#####################
batch, sampled_sent_ids = self._make_batch_and_sample_sent_ids(
batch, record, train_sents, train_sent_ids)
##############
# Prediction #
##############
feed_dict = self._get_feed_dict(batch)
batch_sims, batch_preds = self.sess.run(
[self.similarity, self.predicts], feed_dict)
crr_i, p_total_i = count_gold_and_system_outputs(
batch["tags"], batch_preds, NULL_LABEL_ID)
correct += crr_i
p_total += p_total_i
####################
# Write the result #
####################
self._write_predictions(fout_txt, record)
self._write_nearest_spans(fout_txt, record, train_data,
sampled_sent_ids, batch_sims,
batch_preds, print_knn)
fout_txt.close()
p, r, f = f_score(correct, p_total, self.n_gold_spans)
self.logger.info("-- Time: %f seconds" % (time.time() - start_time))
self.logger.info(
"-- {} set\tF:{:>7.2%} P:{:>7.2%} ({:>5}/{:>5}) R:{:>7.2%} ({:>5}/{:>5})"
.format(data_name, f, p, correct, p_total, r, correct,
self.n_gold_spans))
def save_predicted_bio_tags(self, data_name, preprocessor):
self.logger.info(str(self.cfg))
########################
# Load validation data #
########################
valid_data = preprocessor.load_dataset(
self.cfg["data_path"],
keep_number=True,
lowercase=self.cfg["char_lowercase"])
valid_data = valid_data[:self.cfg["data_size"]]
dataset = preprocessor.build_dataset(valid_data, self.word_dict,
self.char_dict, self.tag_dict)
dataset_path = os.path.join(self.cfg["save_path"], "tmp.json")
write_json(dataset_path, dataset)
self.logger.info("Valid sentences: {:>7}".format(len(dataset)))
######################
# Load training data #
######################
train_sents = load_json(self.cfg["train_set"])
if self.cfg["knn_sampling"] == "random":
train_sent_ids = [sent_id for sent_id in range(len(train_sents))]
else:
train_sent_ids = None
self.logger.info("Train sentences: {:>7}".format(len(train_sents)))
#############
# Main loop #
#############
start_time = time.time()
path = os.path.join(self.cfg["checkpoint_path"],
"%s.bio.txt" % data_name)
fout_txt = open(path, "w")
print("PREDICTION START")
for record, data in zip(valid_data, dataset):
valid_sent_id = record["sent_id"]
batch = self.make_one_batch_for_target(data,
valid_sent_id,
add_tags=False)
if (valid_sent_id + 1) % 100 == 0:
print("%d" % (valid_sent_id + 1), flush=True, end=" ")
#####################
# Sentence sampling #
#####################
batch, sampled_sent_ids = self._make_batch_and_sample_sent_ids(
batch, record, train_sents, train_sent_ids)
###############
# KNN predict #
###############
feed_dict = self._get_feed_dict(batch)
proba = self.sess.run([self.marginal_proba], feed_dict)[0][0]
########################
# Make predicted spans #
########################
words = record["words"]
triples = greedy_search(proba,
n_words=len(words),
max_span_len=self.max_span_len,
null_label_id=NULL_LABEL_ID)
pred_bio_tags = span2bio(spans=triples,
n_words=len(words),
tag_dict=self.rev_tag_dict)
gold_bio_tags = span2bio(spans=record["tags"], n_words=len(words))
assert len(words) == len(pred_bio_tags) == len(gold_bio_tags)
####################
# Write the result #
####################
for word, gold_tag, pred_tag in zip(words, gold_bio_tags,
pred_bio_tags):
fout_txt.write("%s _ %s %s\n" % (word, gold_tag, pred_tag))
fout_txt.write("\n")
self.logger.info("-- Time: %f seconds\nFINISHED." %
(time.time() - start_time))
def save_predicted_spans(self, data_name, preprocessor):
self.logger.info(str(self.cfg))
########################
# Load validation data #
########################
valid_data = preprocessor.load_dataset(
self.cfg["data_path"],
keep_number=True,
lowercase=self.cfg["char_lowercase"])
valid_data = valid_data[:self.cfg["data_size"]]
dataset = preprocessor.build_dataset(valid_data, self.word_dict,
self.char_dict, self.tag_dict)
dataset_path = os.path.join(self.cfg["save_path"], "tmp.json")
write_json(dataset_path, dataset)
self.logger.info("Valid sentences: {:>7}".format(len(dataset)))
######################
# Load training data #
######################
train_sents = load_json(self.cfg["train_set"])
if self.cfg["knn_sampling"] == "random":
train_sent_ids = [sent_id for sent_id in range(len(train_sents))]
else:
train_sent_ids = None
self.logger.info("Train sentences: {:>7}".format(len(train_sents)))
#############
# Main loop #
#############
start_time = time.time()
results = []
print("PREDICTION START")
for record, data in zip(valid_data, dataset):
valid_sent_id = record["sent_id"]
batch = self.make_one_batch_for_target(data,
valid_sent_id,
add_tags=False)
if (valid_sent_id + 1) % 100 == 0:
print("%d" % (valid_sent_id + 1), flush=True, end=" ")
#####################
# Sentence sampling #
#####################
batch, sampled_sent_ids = self._make_batch_and_sample_sent_ids(
batch, record, train_sents, train_sent_ids)
###############
# KNN predict #
###############
feed_dict = self._get_feed_dict(batch)
batch_preds = self.sess.run([self.predicts], feed_dict)[0]
preds = batch_preds[0]
########################
# Make predicted spans #
########################
indx_i, indx_j = get_span_indices(n_words=len(record["words"]),
max_span_len=self.max_span_len)
assert len(preds) == len(indx_i) == len(indx_j)
pred_spans = [[
self.rev_tag_dict[pred_label_id],
int(i), int(j)
] for pred_label_id, i, j in zip(preds, indx_i, indx_j)
if pred_label_id != NULL_LABEL_ID]
##################
# Add the result #
##################
results.append({
"sent_id": valid_sent_id,
"words": record["words"],
"spans": pred_spans,
"train_sent_ids": sampled_sent_ids
})
path = os.path.join(self.cfg["checkpoint_path"],
"%s.predicted_spans.json" % data_name)
write_json(path, results)
self.logger.info("-- Time: %f seconds\nFINISHED." %
(time.time() - start_time))
def save_span_representation(self, data_name, preprocessor):
self.logger.info(str(self.cfg))
########################
# Load validation data #
########################
valid_data = preprocessor.load_dataset(
self.cfg["data_path"],
keep_number=True,
lowercase=self.cfg["char_lowercase"])
valid_data = valid_data[:self.cfg["data_size"]]
dataset = preprocessor.build_dataset(valid_data, self.word_dict,
self.char_dict, self.tag_dict)
dataset_path = os.path.join(self.cfg["save_path"], "tmp.json")
write_json(dataset_path, dataset)
self.logger.info("Valid sentences: {:>7}".format(len(dataset)))
#############
# Main loop #
#############
start_time = time.time()
results = []
fout_hdf5 = h5py.File(
os.path.join(self.cfg["checkpoint_path"],
"%s.span_reps.hdf5" % data_name), 'w')
print("PREDICTION START")
for record, data in zip(valid_data, dataset):
valid_sent_id = record["sent_id"]
batch = self.batcher.make_each_batch(
batch_words=[data["words"]],
batch_chars=[data["chars"]],
max_span_len=self.max_span_len,
batch_tags=[data["tags"]])
if (valid_sent_id + 1) % 100 == 0:
print("%d" % (valid_sent_id + 1), flush=True, end=" ")
#################
# Predict spans #
#################
feed_dict = self._get_feed_dict(batch)
preds, span_reps = self.sess.run([self.predicts, self.span_rep],
feed_dict=feed_dict)
golds = batch["tags"][0]
preds = preds[0]
span_reps = span_reps[0]
assert len(span_reps) == len(golds) == len(preds)
########################
# Make predicted spans #
########################
indx_i, indx_j = get_span_indices(n_words=len(record["words"]),
max_span_len=self.max_span_len)
assert len(preds) == len(indx_i) == len(indx_j)
pred_spans = [[self.rev_tag_dict[label_id],
int(i), int(j)]
for label_id, i, j in zip(preds, indx_i, indx_j)]
gold_spans = [[self.rev_tag_dict[label_id],
int(i), int(j)]
for label_id, i, j in zip(golds, indx_i, indx_j)]
####################
# Write the result #
####################
fout_hdf5.create_dataset(name='{}'.format(valid_sent_id),
dtype='float32',
data=span_reps)
results.append({
"sent_id": valid_sent_id,
"words": record["words"],
"gold_spans": gold_spans,
"pred_spans": pred_spans
})
fout_hdf5.close()
write_json(
os.path.join(self.cfg["checkpoint_path"],
"%s.spans.json" % data_name), results)
self.logger.info("-- Time: %f seconds\nFINISHED." %
(time.time() - start_time))
