def __init__(self, config):
self.logger = ModelLogger(config, dirname=config['dir'])
self.dirname = self.logger.dirname
cuda = config['cuda']
cuda_id = config['cuda_id']
if not cuda:
self.device = torch.device('cpu')
else:
self.device = torch.device('cuda' if cuda_id < 0 else 'cuda:%d' %
cuda_id)
datasets = prepare_datasets(config)
train_set = datasets['train']
dev_set = datasets['dev']
test_set = datasets['test']
# Evaluation Metrics:
self._train_loss = AverageMeter()
self._train_f1 = AverageMeter()
self._train_em = AverageMeter()
self._dev_f1 = AverageMeter()
self._dev_em = AverageMeter()
if train_set:
self.train_loader = DataLoader(train_set,
batch_size=config['batch_size'],
shuffle=config['shuffle'],
collate_fn=lambda x: x,
pin_memory=True)
self._n_train_batches = len(train_set) // config['batch_size']
else:
self.train_loader = None
if dev_set:
self.dev_loader = DataLoader(dev_set,
batch_size=config['batch_size'],
shuffle=False,
collate_fn=lambda x: x,
pin_memory=True)
self._n_dev_batches = len(dev_set) // config['batch_size']
else:
self.dev_loader = None
if test_set:
self.test_loader = DataLoader(test_set,
batch_size=config['batch_size'],
shuffle=False,
collate_fn=lambda x: x,
pin_memory=True)
self._n_test_batches = len(test_set) // config['batch_size']
self._n_test_examples = len(test_set)
else:
self.test_loader = None
self._n_train_examples = 0
self.model = Model(config, train_set)
self.tokenizer = BertTokenizer.from_pretrained("bert-large-uncased")
self.model.network = self.model.network.to(self.device)
self.config = config
self.is_test = False
def __init__(self, config):
self.config = config
tokenizer_model = MODELS[config['model_name']]
self.train_loader, self.dev_loader, tokenizer = prepare_datasets(
config, tokenizer_model)
if config['cuda']:
self.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
def __init__(self, config):
self.config = config
tokenizer_model = MODELS[config['model_name']]
self.train_loader, self.dev_loader, tokenizer = prepare_datasets(
config, tokenizer_model)
self._n_dev_batches = len(
self.dev_loader.dataset) // config['batch_size']
self._n_train_batches = len(
self.train_loader.dataset) // config['batch_size']
if config['cuda']:
self.device = torch.device(
'cuda' if torch.cuda.is_available() else 'cpu')
else:
self.device = torch.device('cpu')
print("use device: ", self.device)
self._train_loss = AverageMeter()
self._train_f1 = AverageMeter()
self._train_em = AverageMeter()
self._dev_f1 = AverageMeter()
self._dev_em = AverageMeter()
self.model = Model(config, MODELS[config['model_name']], self.device,
tokenizer).to(self.device)
t_total = len(
self.train_loader
) // config['gradient_accumulation_steps'] * config['max_epochs']
self.optimizer = AdamW(self.model.parameters(),
lr=config['lr'],
eps=config['adam_epsilon'])
self.optimizer.zero_grad()
self._n_train_examples = 0
self._epoch = self._best_epoch = 0
self._best_f1 = 0
self._best_em = 0
self.restored = False
if config['pretrained_dir'] is not None:
if config['mode'] == 'train':
self.restore()
else:
self.load_model()
def __init__(self, config):
self.logger = ModelLogger(config,
dirname=config['dir'],
pretrained=config['pretrained'])
self.dirname = self.logger.dirname
#self.dirname=config["pretrained_model"]
cuda = config['cuda']
cuda_id = config['cuda_id']
if not cuda:
self.device = torch.device('cpu')
else:
self.device = torch.device('cuda' if cuda_id < 0 else 'cuda:%d' %
cuda_id)
#データの読み込み
datasets = prepare_datasets(config)
train_set = datasets['train']
dev_set = datasets['dev']
test_set = datasets['test']
# Evaluation Metrics:
self._train_loss = AverageMeter()
self._train_f1 = AverageMeter()
self._train_em = AverageMeter()
self._dev_f1 = AverageMeter()
self._dev_em = AverageMeter()
#データのロード
if train_set:
self.train_loader = DataLoader(train_set,
batch_size=config['batch_size'],
shuffle=config['shuffle'],
collate_fn=lambda x: x,
pin_memory=True)
self._n_train_batches = len(train_set) // config['batch_size']
else:
self.train_loader = None
if dev_set:
self.dev_loader = DataLoader(dev_set,
batch_size=config['batch_size'],
shuffle=False,
collate_fn=lambda x: x,
pin_memory=True)
self._n_dev_batches = len(dev_set) // config['batch_size']
else:
self.dev_loader = None
if test_set:
self.test_loader = DataLoader(test_set,
batch_size=config['batch_size'],
shuffle=False,
collate_fn=lambda x: x,
pin_memory=True)
self._n_test_batches = len(test_set) // config['batch_size']
self._n_test_examples = len(test_set)
else:
self.test_loader = None
#モデルの用意
self._n_train_examples = 0
self.model = Model(config, train_set)
self.model.network = self.model.network.to(self.device)
self.config = self.model.config
self.is_test = False
self.textlogger = get_logger("log.txt")