def load_best(experiment, parts):
log_data = parse(log(experiment))
best_step, _ = max_key(scores(log_data, parts))
parameters = load_parameters(experiment)
MODEL_CLASS = load_model(parameters['MODEL_PREFIX'])
TOKENIZER = MODEL_CLASS[0].from_pretrained(parameters['MODEL_NAME'])
CONFIG = MODEL_CLASS[1].from_pretrained(parameters['MODEL_NAME'],
num_labels=3)
MODEL = MODEL_CLASS[2].from_pretrained(parameters['MODEL_NAME'],
config=CONFIG)
MODEL.load_state_dict(
torch.load('logs/{}/model_{}.torch'.format(experiment, best_step),
map_location='cpu'))
return best_step, log_data[best_step], MODEL, TOKENIZER
with open(output_dir + '/parameters.json', 'w+') as config_file:
json.dump({
'MODEL_PREFIX': MODEL_PREFIX,
'MODEL_NAME': MODEL_NAME,
'DATASET': DATASET,
'LANGS': LANGS,
'TRAIN_BATCH_SIZE': TRAIN_BATCH_SIZE,
'ACCUMULATION_STEPS': ACCUMULATION_STEPS,
'LEARN_RATE': LEARN_RATE,
'EPOCHS': EPOCHS,
'WARMUP_STEPS': WARMUP_STEPS,
'SEQUENCE_LENGTH': SEQUENCE_LENGTH,
}, config_file, sort_keys=True, indent=4, separators=(',', ': '))
# Load and initialize model
MODEL_CLASS = load_model(MODEL_PREFIX)
TOKENIZER = MODEL_CLASS[0].from_pretrained(MODEL_NAME)
CONFIG = MODEL_CLASS[1].from_pretrained(MODEL_NAME, num_labels=3)
MODEL = MODEL_CLASS[2].from_pretrained(MODEL_NAME, config=CONFIG)
# Load training data
train_dataset = dataset(
tokenize(chain(*(load_semeval(DATASET, 'train', lang) for lang in LANGS)), TOKENIZER, SEQUENCE_LENGTH))
train_sampler = RandomSampler(train_dataset)
train_dataset = DataLoader(train_dataset, sampler=train_sampler, batch_size=TRAIN_BATCH_SIZE, drop_last=True)
# Run Training
training(
train_dataset,
val_datasets(TOKENIZER, SEQUENCE_LENGTH),
MODEL,