def _eval(mymodel,
args,
data_loader,
data_set=None,
RELOAD_MODEL=None,
use_cuda=False):
old_model_path = os.path.join(args.result_dir, RELOAD_MODEL)
if RELOAD_MODEL is not None and os.path.exists(old_model_path):
mymodel.load_model(old_model_path)
LOGGER.info("Reload model successfully~")
else:
LOGGER.info(
f'There is no such file in {old_model_path}, End evaluation')
return
LOGGER.info(f'reload model {RELOAD_MODEL}')
hyper_param = {
'batch_size': 100,
'issave': True,
'result_dir': args.result_dir
}
timer = Timer()
timer.set(args.time_budget)
with timer.time_limit('eval'):
mymodel.eval_model(data_loader, data_set, hyper_param, rebuild=True)
def _train(mymodel, args, data_loader, train_dataset=None, eval_dataset=None, RELOAD_MODEL=None, use_cuda=False):
if RELOAD_MODEL is None:
LOGGER.info(f'Rebuild model')
else:
old_model_path = os.path.join(args.result_dir, RELOAD_MODEL)
if os.path.exists(old_model_path):
mymodel.load_model(old_model_path)
LOGGER.info("Reload model successfully~")
else:
LOGGER.info(f'There is no such file in {old_model_path}, Rebuild model')
##TODO:
BATCH_SIZE = 8 if data_loader.sentence_max_len > 200 else 16
if use_cuda:
train_param = {
'EPOCH': 30, #45 TODO:15
'batch_size': BATCH_SIZE, #512 TODO:64
'learning_rate_bert': 5e-5,
'learning_rate_upper': 1e-3, #TODO:1e-3
'bert_finetune': True,
'visualize_length': 20, #10
'isshuffle': True,
'result_dir': args.result_dir,
'model_name':'model_test.p'
}
else:
train_param = {
'EPOCH': CPU_EPOCH, #45
'batch_size': CPU_BATCHSIZE, #512
'learning_rate_bert': 5e-5,
'learning_rate_upper': 1e-3,
'bert_finetune': False,
'visualize_length': CPU_VISUAL, #10
'isshuffle': True,
'result_dir': args.result_dir,
'model_name':'model_test.p'
}
timer = Timer()
timer.set(args.time_budget)
loss_record = None
with timer.time_limit('training'):
loss_record, score_record = mymodel.train_model(data_loader, hyper_param=train_param, train_dataset=train_dataset, eval_dataset=eval_dataset)
loss_record = np.array(loss_record)
loss_save_path = os.path.join(args.result_dir, 'loss_train.txt')
loss_img_path = os.path.join(args.result_dir, 'loss.png')
np.savetxt(loss_save_path, loss_record)
score_record = np.array(score_record)
score_save_path = os.path.join(args.result_dir, 'score_train.txt')
score_img_path = os.path.join(args.result_dir, 'score.png')
np.savetxt(score_save_path, score_record) ### (epochs, 3)
loss = np.loadtxt(loss_save_path).reshape(1, -1)
score = np.loadtxt(score_save_path).T
plot_img(loss, loss_img_path)
plot_img(score, score_img_path)
