def _calculate_metrics(self, run_states):
total_infer = total_label = total_correct = loss_sum = 0
run_step = run_time_used = run_examples = 0
precision_sum = recall_sum = f1_score_sum = 0
for run_state in run_states:
loss_sum += np.mean(run_state.run_results[-1])
if self.add_crf:
precision_sum += np.mean(
run_state.run_results[0]) * run_state.run_examples
recall_sum += np.mean(
run_state.run_results[1]) * run_state.run_examples
f1_score_sum += np.mean(
run_state.run_results[2]) * run_state.run_examples
else:
np_labels = run_state.run_results[0]
np_infers = run_state.run_results[1]
np_lens = run_state.run_results[2]
label_num, infer_num, correct_num = chunk_eval(
np_labels, np_infers, np_lens, self.num_labels,
self.device_count)
total_infer += infer_num
total_label += label_num
total_correct += correct_num
run_examples += run_state.run_examples
run_step += run_state.run_step
run_time_used = time.time() - run_states[0].run_time_begin
run_speed = run_step / run_time_used
avg_loss = loss_sum / run_examples
if self.add_crf:
precision = precision_sum / run_examples
recall = recall_sum / run_examples
f1 = f1_score_sum / run_examples
else:
precision, recall, f1 = calculate_f1(total_label, total_infer,
total_correct)
# The first key will be used as main metrics to update the best model
scores = OrderedDict()
for metric in self.metrics_choices:
if metric == "precision":
scores["precision"] = precision
elif metric == "recall":
scores["recall"] = recall
elif metric == "f1":
scores["f1"] = f1
else:
raise ValueError("Not Support Metric: \"%s\"" % metric)
return scores, avg_loss, run_speed
def _calculate_metrics(self, run_states):
total_infer = total_label = total_correct = loss_sum = 0
run_step = run_time_used = run_examples = 0
for run_state in run_states:
loss_sum += np.mean(run_state.run_results[-1])
np_labels = run_state.run_results[0]
np_infers = run_state.run_results[1]
np_lens = run_state.run_results[2]
label_num, infer_num, correct_num = chunk_eval(
np_labels, np_infers, np_lens, self.num_labels,
self.device_count)
total_infer += infer_num
total_label += label_num
total_correct += correct_num
run_examples += run_state.run_examples
run_step += run_state.run_step
run_time_used = time.time() - run_states[0].run_time_begin
run_speed = run_step / run_time_used
avg_loss = loss_sum / run_examples
precision, recall, f1 = calculate_f1(total_label, total_infer,
total_correct)
return precision, recall, f1, avg_loss, run_speed
def finetune(args):
module = hub.Module(name="ernie", max_seq_len=args.max_seq_len)
# Use the appropriate tokenizer to preprocess the data set
tokenizer = hub.BertTokenizer(vocab_file=module.get_vocab_path())
dataset = hub.dataset.MSRA_NER(tokenizer=tokenizer,
max_seq_len=args.max_seq_len)
with fluid.dygraph.guard():
ts = TransformerSeqLabeling(num_classes=dataset.num_labels,
transformer=module)
adam = AdamOptimizer(learning_rate=1e-5,
parameter_list=ts.parameters())
state_dict_path = os.path.join(args.checkpoint_dir,
'dygraph_state_dict')
if os.path.exists(state_dict_path + '.pdparams'):
state_dict, _ = fluid.load_dygraph(state_dict_path)
ts.load_dict(state_dict)
loss_sum = total_infer = total_label = total_correct = cnt = 0
for epoch in range(args.num_epoch):
for batch_id, data in enumerate(
dataset.batch_records_generator(
phase="train",
batch_size=args.batch_size,
shuffle=True,
pad_to_batch_max_seq_len=False)):
batch_size = len(data["input_ids"])
input_ids = np.array(data["input_ids"]).astype(
np.int64).reshape([batch_size, -1, 1])
position_ids = np.array(data["position_ids"]).astype(
np.int64).reshape([batch_size, -1, 1])
segment_ids = np.array(data["segment_ids"]).astype(
np.int64).reshape([batch_size, -1, 1])
input_mask = np.array(data["input_mask"]).astype(
np.float32).reshape([batch_size, -1, 1])
labels = np.array(data["label"]).astype(np.int64).reshape(
-1, 1)
seq_len = np.array(data["seq_len"]).astype(np.int64).reshape(
-1, 1)
pred, ret_infers = ts(input_ids, position_ids, segment_ids,
input_mask)
loss = fluid.layers.cross_entropy(pred, to_variable(labels))
avg_loss = fluid.layers.mean(loss)
avg_loss.backward()
adam.minimize(avg_loss)
loss_sum += avg_loss.numpy() * labels.shape[0]
label_num, infer_num, correct_num = chunk_eval(
labels, ret_infers.numpy(), seq_len, dataset.num_labels, 1)
cnt += labels.shape[0]
total_infer += infer_num
total_label += label_num
total_correct += correct_num
if batch_id % args.log_interval == 0:
precision, recall, f1 = calculate_f1(
total_label, total_infer, total_correct)
print('epoch {}: loss {}, f1 {} recall {} precision {}'.
format(epoch, loss_sum / cnt, f1, recall, precision))
loss_sum = total_infer = total_label = total_correct = cnt = 0
if batch_id % args.save_interval == 0:
state_dict = ts.state_dict()
fluid.save_dygraph(state_dict, state_dict_path)
def finetune(args):
ernie = hub.Module(name="ernie", max_seq_len=args.max_seq_len)
with fluid.dygraph.guard():
dataset = hub.dataset.MSRA_NER()
ts = TransformerSequenceLabelLayer(
num_classes=dataset.num_labels, transformer=ernie)
adam = AdamOptimizer(learning_rate=1e-5, parameter_list=ts.parameters())
state_dict_path = os.path.join(args.checkpoint_dir,
'dygraph_state_dict')
if os.path.exists(state_dict_path + '.pdparams'):
state_dict, _ = fluid.load_dygraph(state_dict_path)
ts.load_dict(state_dict)
reader = hub.reader.SequenceLabelReader(
dataset=dataset,
vocab_path=ernie.get_vocab_path(),
max_seq_len=args.max_seq_len,
sp_model_path=ernie.get_spm_path(),
word_dict_path=ernie.get_word_dict_path())
train_reader = reader.data_generator(
batch_size=args.batch_size, phase='train')
loss_sum = total_infer = total_label = total_correct = cnt = 0
# 执行epoch_num次训练
for epoch in range(args.num_epoch):
# 读取训练数据进行训练
for batch_id, data in enumerate(train_reader()):
input_ids = np.array(data[0][0]).astype(np.int64)
position_ids = np.array(data[0][1]).astype(np.int64)
segment_ids = np.array(data[0][2]).astype(np.int64)
input_mask = np.array(data[0][3]).astype(np.float32)
labels = np.array(data[0][4]).astype(np.int64).reshape(-1, 1)
seq_len = np.squeeze(
np.array(data[0][5]).astype(np.int64), axis=1)
pred, ret_infers = ts(input_ids, position_ids, segment_ids,
input_mask)
loss = fluid.layers.cross_entropy(pred, to_variable(labels))
avg_loss = fluid.layers.mean(loss)
avg_loss.backward()
# 参数更新
adam.minimize(avg_loss)
loss_sum += avg_loss.numpy() * labels.shape[0]
label_num, infer_num, correct_num = chunk_eval(
labels, ret_infers.numpy(), seq_len, dataset.num_labels, 1)
cnt += labels.shape[0]
total_infer += infer_num
total_label += label_num
total_correct += correct_num
if batch_id % args.log_interval == 0:
precision, recall, f1 = calculate_f1(
total_label, total_infer, total_correct)
print('epoch {}: loss {}, f1 {} recall {} precision {}'.
format(epoch, loss_sum / cnt, f1, recall, precision))
loss_sum = total_infer = total_label = total_correct = cnt = 0
if batch_id % args.save_interval == 0:
state_dict = ts.state_dict()
fluid.save_dygraph(state_dict, state_dict_path)
