if (args.task == 'abs'):
if (args.mode == 'train'):
train_abs(args, device_id)
elif (args.mode == 'validate'):
validate_abs(args, device_id)
elif (args.mode == 'lead'):
baseline(args, cal_lead=True)
elif (args.mode == 'oracle'):
baseline(args, cal_oracle=True)
if (args.mode == 'test'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_abs(args, device_id, cp, step)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
elif (args.task == 'ext'):
if (args.mode == 'train'):
train_ext(args, device_id)
elif (args.mode == 'validate'):
validate_ext(args, device_id)
if (args.mode == 'test'):
cp = args.test_from
if (args.mode == 'train'):
train_abs(args, device_id, tokenizer)
elif (args.mode == 'validate'):
validate_abs(args, device_id, tokenizer)
elif (args.mode == 'lead'):
baseline(args, cal_lead=True)
elif (args.mode == 'oracle'):
baseline(args, cal_oracle=True)
if (args.mode == 'test'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_abs(args, device_id, cp, step, tokenizer)
elif (args.mode == 'dev'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
dev_abs(args, device_id, cp, step, tokenizer)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step, tokenizer)
args = parser.parse_args()
config_path = args.config_path
with open(config_path, "r") as f:
config = yaml.safe_load(f)
config["train"]["gpu_ranks"] = [
int(i) for i in range(len(config["train"]["visible_gpus"].split(",")))
]
config["train"]["world_size"] = len(config["train"]["gpu_ranks"])
os.environ["CUDA_VISIBLE_DEVICES"] = config["train"]["visible_gpus"]
init_logger(config["train"]["log_file"])
DEVICE = "cpu" if config["train"]["visible_gpus"] == "-1" else "cuda"
DEVICE_ID = 0 if DEVICE == "cuda" else -1
kwargs_train = config["train"]
kwargs_test = {**kwargs_train, **config["test"]}
TrainArgs = recordtype("TrainArgs", kwargs_train)
TestArgs = recordtype("TestArgs", kwargs_test)
train_args = TrainArgs(**kwargs_train)
test_args = TestArgs(**kwargs_test)
if args.mode == "train":
train_abs(args=train_args, device_id=DEVICE_ID)
elif args.mode == "test":
cp = test_args.test_from
try:
STEP = int(cp.split(".")[-2].split("_")[-1])
except ValueError:
STEP = 0
test_abs(args=test_args, device_id=DEVICE_ID, pt=cp, step=STEP)
else:
raise ValueError("Unknown argument")
if (args.task == 'abs'):
if (args.mode == 'train'):
train_abs(args, device_id)
elif (args.mode == 'validate'):
validate_abs(args, device_id)
elif (args.mode == 'lead'):
baseline(args, cal_lead=True)
elif (args.mode == 'oracle'):
baseline(args, cal_oracle=True)
if (args.mode == 'test'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_abs(args, device_id, cp, step, calculate_rouge=args.dont_calculate_rouge)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
elif (args.task == 'ext'):
if (args.mode == 'train'):
train_ext(args, device_id)
elif (args.mode == 'validate'):
validate_ext(args, device_id)
if (args.mode == 'test'):
cp = args.test_from
def run(args, hpspace):
if (args.task == 'abs'):
if (args.mode == 'train'):
if (args.hyperopt):
t = time.time()
newT = str(t).split(".")
args.model_path = "../models/" + newT[0]
args.log_file = "../logs/abs_bert_abs_" + newT[0]
args.lr_bert = hpspace['lr_bert']
args.lr_dec = hpspace['lr_dec']
args.accum_count = int(hpspace['accum_count'])
args.beta1 = hpspace['beta1']
args.beta2 = hpspace['beta2']
args.visible_gpus = '0'
args.bert_model = '..temp/bert-base-danish-uncased-v2'
args.vocab = '..temp/bert-base-danish-uncased-v2'
train_stats = train_abs(args, device_id)
x = train_stats.x
ppl = train_stats.perplexity
acc = train_stats.acc
print(x)
return {
'loss': x,
'eval_time': time.time(),
'status': STATUS_OK,
'other_stuff': {
'ppl': ppl,
'acc': acc
}
}
elif (args.mode == 'validate'):
validate_abs(args, device_id)
elif (args.mode == 'lead'):
baseline(args, cal_lead=True)
elif (args.mode == 'oracle'):
baseline(args, cal_oracle=True)
if (args.mode == 'test'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_abs(args, device_id, cp, step)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
elif (args.task == 'ext'):
if (args.mode == 'train'):
train_ext(args, device_id)
elif (args.mode == 'validate'):
validate_ext(args, device_id)
if (args.mode == 'test'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_ext(args, device_id, cp, step)
elif (args.mode == 'test_text'):
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
step = 0
test_text_abs(args, device_id, cp, step)
if (args.mode == "sent_label"):
step = 0
sent_label_ext(args, device_id)
os.makedirs(args.model_path, exist_ok=True)
os.makedirs(args.log_path, exist_ok=True)
os.makedirs(args.result_path, exist_ok=True)
# Train/valid/test
if args.task == 'abs':
if args.mode == 'train':
init_logger(os.path.join(args.log_path, 'train.log'))
train_abs(args, DEVICE_ID)
elif args.mode == 'validate':
init_logger(os.path.join(args.log_path, 'valid.log'))
validate_abs(args, DEVICE_ID)
elif args.mode == 'test':
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
print("Not correct model name (EX: model_step_1200.pt)")
init_logger(
os.path.join(args.log_path, 'test.' + str(step) + '.log'))
test_abs(args, DEVICE_ID, cp, step)
elif args.mode == 'test_text':
cp = args.test_from
try:
step = int(cp.split('.')[-2].split('_')[-1])
except:
print("Not correct model name (EX: model_step_1200.pt)")
init_logger(
os.path.join(args.log_path, 'test_text.' + str(step) + '.log'))
test_text_abs(args, DEVICE_ID, cp, step)