def train(path):
# test saving pipeline
save_pipe(path)
# Trainer
trainer = Trainer(model=model,
train_data=train_data,
dev_data=dev_data,
loss=ParserLoss(),
metrics=ParserMetric(),
metric_key='UAS',
**train_args.data,
optimizer=fastNLP.Adam(**optim_args.data),
save_path=path)
# model.word_embedding = torch.nn.Embedding.from_pretrained(embed, freeze=False)
model.word_embedding.padding_idx = word_v.padding_idx
model.word_embedding.weight.data[word_v.padding_idx].fill_(0)
model.pos_embedding.padding_idx = pos_v.padding_idx
model.pos_embedding.weight.data[pos_v.padding_idx].fill_(0)
# try:
# ModelLoader.load_pytorch(model, "./save/saved_model.pkl")
# print('model parameter loaded!')
# except Exception as _:
# print("No saved model. Continue.")
# pass
# Start training
trainer.train()
print("Training finished!")
# save pipeline
save_pipe(path)
print('pipe saved')
def test_train2(self):
model = BiaffineParser(init_embed=(VOCAB_SIZE, 10),
pos_vocab_size=VOCAB_SIZE, pos_emb_dim=10,
rnn_hidden_size=16,
arc_mlp_size=10,
label_mlp_size=10,
num_label=NUM_CLS, encoder='transformer')
ds = prepare_parser_data()
RUNNER.run_model(model, ds, loss=ParserLoss(), metrics=ParserMetric())
def train(path):
# test saving pipeline
save_pipe(path)
embed = EmbedLoader.fast_load_embedding(model_args['word_emb_dim'],
emb_file_name, word_v)
embed = torch.tensor(embed, dtype=torch.float32)
# embed = EmbedLoader.fast_load_embedding(emb_dim=model_args['word_emb_dim'], emb_file=emb_file_name, vocab=word_v)
# embed = torch.tensor(embed, dtype=torch.float32)
# model.word_embedding = torch.nn.Embedding.from_pretrained(embed, freeze=True)
model.word_embedding.padding_idx = word_v.padding_idx
model.word_embedding.weight.data[word_v.padding_idx].fill_(0)
model.pos_embedding.padding_idx = pos_v.padding_idx
model.pos_embedding.weight.data[pos_v.padding_idx].fill_(0)
class MyCallback(Callback):
def on_step_end(self, optimizer):
step = self.trainer.step
# learning rate decay
if step > 0 and step % 1000 == 0:
for pg in optimizer.param_groups:
pg['lr'] *= 0.93
print('decay lr to {}'.format(
[pg['lr'] for pg in optimizer.param_groups]))
if step == 3000:
# start training embedding
print('start training embedding at {}'.format(step))
model = self.trainer.model
for m in model.modules():
if isinstance(m, torch.nn.Embedding):
m.weight.requires_grad = True
# Trainer
trainer = Trainer(model=model,
train_data=train_data,
dev_data=dev_data,
loss=ParserLoss(),
metrics=ParserMetric(),
metric_key='UAS',
**train_args.data,
optimizer=fastNLP.Adam(**optim_args.data),
save_path=path,
callbacks=[MyCallback()])
# Start training
try:
trainer.train()
print("Training finished!")
finally:
# save pipeline
save_pipe(path)
print('pipe saved')
def test_train(self):
ds, v1, v2, v3 = init_data()
model = BiaffineParser(word_vocab_size=len(v1),
word_emb_dim=30,
pos_vocab_size=len(v2),
pos_emb_dim=30,
num_label=len(v3))
trainer = fastNLP.Trainer(model=model,
train_data=ds,
dev_data=ds,
loss=ParserLoss(),
metrics=ParserMetric(),
metric_key='UAS',
n_epochs=10,
use_cuda=False,
use_tqdm=False)
trainer.train(load_best_model=False)