def preprocess(self, requests):
""" Very basic preprocessing code - only tokenizes.
Extend with your own preprocessing steps as needed.
"""
input_batch = []
for idx, data in enumerate(requests):
input_text = data.get("data")
if input_text is None:
input_text = data.get("body")
if isinstance(input_text, (bytes, bytearray)):
input_text = input_text.decode('utf-8')
input_text = json.loads(input_text)
input_batch.extend(input_text)
eval_examples = get_examples_from_dialogues(input_batch,
user_first=False,
dialogue_level=False)
eval_features = self.processor.convert_examples_to_features(
eval_examples)
eval_data = WOSDataset(eval_features)
eval_sampler = SequentialSampler(eval_data)
eval_loader = DataLoader(
eval_data,
batch_size=1,
sampler=eval_sampler,
collate_fn=self.processor.collate_fn,
)
return eval_loader
# Slot Meta tokenizing for the decoder initial inputs
tokenized_slot_meta = []
for slot in slot_meta:
tokenized_slot_meta.append(
tokenizer.encode(slot.replace("-", " "), add_special_tokens=False)
)
# Model 선언
model = TRADE(args, tokenized_slot_meta)
model.set_subword_embedding(args.model_name_or_path) # Subword Embedding 초기화
print(f"Subword Embeddings is loaded from {args.model_name_or_path}")
model.to(device)
print("Model is initialized")
train_data = WOSDataset(train_features)
train_sampler = RandomSampler(train_data)
train_loader = DataLoader(
train_data,
batch_size=args.train_batch_size,
sampler=train_sampler,
collate_fn=processor.collate_fn,
)
print("# train:", len(train_data))
dev_data = WOSDataset(dev_features)
dev_sampler = SequentialSampler(dev_data)
dev_loader = DataLoader(
dev_data,
batch_size=args.eval_batch_size,
sampler=dev_sampler,
model_dir_path = os.path.dirname(args.model_dir)
eval_data = json.load(open(f"{args.data_dir}/eval_dials.json", "r"))
config = json.load(open(f"{model_dir_path}/exp_config.json", "r"))
config = argparse.Namespace(**config)
slot_meta = json.load(open(f"{model_dir_path}/slot_meta.json", "r"))
tokenizer = BertTokenizer.from_pretrained(config.model_name_or_path)
processor = TRADEPreprocessor(slot_meta, tokenizer)
eval_examples = get_examples_from_dialogues(
eval_data, user_first=False, dialogue_level=False
)
# Extracting Featrues
eval_features = processor.convert_examples_to_features(eval_examples)
eval_data = WOSDataset(eval_features)
eval_sampler = SequentialSampler(eval_data)
eval_loader = DataLoader(
eval_data,
batch_size=args.eval_batch_size,
sampler=eval_sampler,
collate_fn=processor.collate_fn,
)
print("# eval:", len(eval_data))
tokenized_slot_meta = []
for slot in slot_meta:
tokenized_slot_meta.append(
tokenizer.encode(slot.replace("-", " "), add_special_tokens=False)
)
def train(args):
# operation encoder domain encoder
op2id = OP_SET[args.op_code]
domain2id = DOMAIN2ID
args.n_op = len(op2id)
args.n_domain = get_domain_nums(domain2id)
args.update_id = op2id["update"]
# initialize tokenizer
tokenizer_module = getattr(import_module("transformers"),
f"{args.model_name}Tokenizer")
tokenizer = tokenizer_module.from_pretrained(args.pretrained_name_or_path)
tokenizer.add_special_tokens({
"additional_special_tokens": [NULL_TOKEN, SLOT_TOKEN],
'eos_token':
EOS_TOKEN
})
args.vocab_size = len(tokenizer)
# load data
slot_meta = json.load(open(os.path.join(args.data_dir, "slot_meta.json")))
train_data, dev_data, dev_labels = load_somdst_dataset(
os.path.join(args.data_dir, "train_dials.json"))
train_examples = get_somdst_examples_from_dialogues(
data=train_data, n_history=args.n_history)
dev_examples = get_somdst_examples_from_dialogues(data=dev_data,
n_history=args.n_history)
# preprocessing
preprocessor = SomDSTPreprocessor(slot_meta=slot_meta,
src_tokenizer=tokenizer,
max_seq_length=args.max_seq_length,
word_dropout=0.1,
domain2id=domain2id)
train_features = preprocessor.convert_examples_to_features(
train_examples, word_dropout=args.word_dropout)
dev_features = preprocessor.convert_examples_to_features(dev_examples,
word_dropout=0.0)
train_dataset = WOSDataset(features=train_features)
train_sampler = RandomSampler(train_dataset)
train_dataloader = DataLoader(
train_dataset,
sampler=train_sampler,
batch_size=args.train_batch_size,
collate_fn=preprocessor.collate_fn,
num_workers=args.num_workers,
)
# initialize model: update embedding size & initailize weights
model_config = BertConfig.from_pretrained(args.pretrained_name_or_path)
model_config.dropout = 0.1
model_config.vocab_size = len(tokenizer)
model = SomDST(model_config,
n_op=args.n_op,
n_domain=args.n_domain,
update_id=args.update_id)
# model.encoder.bert.embeddings.word_embeddings.weight.data[1].normal_(mean=0.0, std=0.02)
# model.encoder.bert.embeddings.word_embeddings.weight.data[2].normal_(mean=0.0, std=0.02)
# model.encoder.bert.embeddings.word_embeddings.weight.data[3].normal_(mean=0.0, std=0.02)
model.to(device)
print("Model is initialized")
# num_train_steps = int(len(train_features) / args.train_batch_size * args.epochs)
num_train_steps = len(train_dataloader) * args.epochs
# no_decay = ["bias", "LayerNorm.bias", "LayerNorm.weight"]
# enc_param_optimizer = list(model.encoder.named_parameters())
# enc_optimizer_grouped_parameters = [
# {
# "params": [
# p for n, p in enc_param_optimizer if not any(nd in n for nd in no_decay)
# ],
# "weight_decay": 0.01,
# },
# {
# "params": [
# p for n, p in enc_param_optimizer if any(nd in n for nd in no_decay)
# ],
# "weight_decay": 0.0,
# },
# ]
# initialize optimizer & scheduler
# enc_optimizer = AdamW(enc_optimizer_grouped_parameters, lr=args.enc_lr, eps=args.adam_epsilon)
enc_param_optimizer = list(model.encoder.parameters())
enc_optimizer = AdamW(enc_param_optimizer,
lr=args.enc_lr,
eps=args.adam_epsilon)
enc_scheduler = get_linear_schedule_with_warmup(
optimizer=enc_optimizer,
num_warmup_steps=0.1,
num_training_steps=num_train_steps)
dec_param_optimizer = list(model.decoder.parameters())
dec_optimizer = AdamW(dec_param_optimizer,
lr=args.dec_lr,
eps=args.adam_epsilon)
dec_scheduler = get_linear_schedule_with_warmup(
optimizer=dec_optimizer,
num_warmup_steps=0.1,
num_training_steps=num_train_steps)
criterion = nn.CrossEntropyLoss()
rng = random.Random(args.seed)
# save experiment settings
json.dump(
vars(args),
open(f"{args.model_dir}/{args.model_fold}/exp_config.json", "w"),
indent=2,
ensure_ascii=False,
)
json.dump(
slot_meta,
open(f"{args.model_dir}/{args.model_fold}/slot_meta.json", "w"),
indent=2,
ensure_ascii=False,
)
best_score = {
"epoch": 0,
"joint_acc": 0, #
"op_acc": 0,
"slot_acc": 0,
"slot_f1": 0,
"op_acc": 0,
"op_f1": 0,
}
for epoch in range(args.epochs):
print(f'Epoch #{epoch}')
batch_loss = []
model.train()
for step, batch in tqdm(enumerate(train_dataloader), desc='[Step]'):
batch = [
b.to(device) if not isinstance(b, int) else b for b in batch
]
(
input_ids,
input_mask,
segment_ids,
state_position_ids,
op_ids,
domain_ids,
gen_ids,
max_value,
max_update,
) = batch
# teacher forcing for generation(decoder)
teacher = gen_ids if rng.random(
) < args.teacher_forcing_ratio else None
domain_scores, state_scores, gen_scores = model(
input_ids=input_ids,
token_type_ids=segment_ids,
state_positions=state_position_ids,
attention_mask=input_mask,
max_value=max_value,
op_ids=op_ids,
max_update=max_update,
teacher=teacher,
)
loss_s = criterion(state_scores.view(-1, len(op2id)),
op_ids.view(-1))
loss_g = masked_cross_entropy_for_value(
logits=gen_scores.contiguous(),
target=gen_ids.contiguous(),
pad_idx=tokenizer.pad_token_id,
)
loss = loss_s + loss_g
if args.exclude_domain is not True:
loss_d = criterion(domain_scores.view(-1, args.n_domain),
domain_ids.view(-1))
loss = loss + loss_d
batch_loss.append(loss.item())
loss.backward()
enc_optimizer.step()
enc_scheduler.step()
dec_optimizer.step()
dec_scheduler.step()
model.zero_grad()
for learning_rate in enc_scheduler.get_lr():
wandb.log({"encoder_learning_rate": learning_rate})
for learning_rate in dec_scheduler.get_lr():
wandb.log({"decoder_learning_rate": learning_rate})
if step % 100 == 0:
if args.exclude_domain is not True:
print(
f"[{epoch+1}/{args.epochs}] [{step}/{len(train_dataloader)}] mean_loss : {np.mean(batch_loss):.3f}, state_loss : {loss_s.item():.3f}, gen_loss : {loss_g.item():.3f}, dom_loss : {loss_d.item():.3f}"
)
wandb.log({
"epoch": epoch,
"Train epoch loss": np.mean(batch_loss),
"Train epoch state loss": loss_s.item(),
"Train epoch generation loss": loss_g.item(),
"Train epoch domain loss": loss_d.item(),
})
else:
print(
f"[{epoch+1}/{args.epochs}] [{step}/{len(train_dataloader)}] mean_loss : {np.mean(batch_loss):.3f}, state_loss : {loss_s.item():.3f}, gen_loss : {loss_g.item():.3f}"
)
wandb.log({
"epoch": epoch,
"Train epoch loss": np.mean(batch_loss),
"Train epoch state loss": loss_s.item(),
"Train epoch generation loss": loss_g.item(),
})
batch_loss = []
# evaluation for each epoch
eval_res = model_evaluation(
model,
dev_features,
tokenizer,
slot_meta,
domain2id,
epoch + 1,
args.op_code,
)
if eval_res["joint_acc"] > best_score["joint_acc"]:
print("Update Best checkpoint!")
best_score = eval_res
best_checkpoint = best_score["epoch"]
wandb.log({
"epoch": best_score["epoch"],
"Best joint goal accuracy": best_score["joint_acc"],
"Best turn slot accuracy": best_score["slot_acc"],
"Best turn slot f1": best_score["slot_f1"],
"Best operation accucay": best_score["op_acc"],
"Best operation f1": best_score["op_f1"],
})
# save phase
model_to_save = model.module if hasattr(model, "module") else model
torch.save(
model_to_save.state_dict(),
f"{args.model_dir}/{args.model_fold}/model-best.bin",
)
print("Best Score : ", best_score)
print("\n")
print(f"Best checkpoint: {args.model_dir}/model-best.bin")