autodl-projects/exps/LFNA/lfna-debug-hpnet.py

229 lines
7.3 KiB
Python
Raw Normal View History

2021-05-12 09:45:45 +02:00
#####################################################
# Copyright (c) Xuanyi Dong [GitHub D-X-Y], 2021.04 #
#####################################################
2021-05-13 10:39:19 +02:00
# python exps/LFNA/lfna-debug-hpnet.py --env_version v1 --hidden_dim 16 --meta_batch 64 --device cuda
2021-05-12 09:45:45 +02:00
#####################################################
import sys, time, copy, torch, random, argparse
from tqdm import tqdm
from copy import deepcopy
from pathlib import Path
lib_dir = (Path(__file__).parent / ".." / ".." / "lib").resolve()
if str(lib_dir) not in sys.path:
sys.path.insert(0, str(lib_dir))
from procedures import prepare_seed, prepare_logger, save_checkpoint, copy_checkpoint
from log_utils import time_string
from log_utils import AverageMeter, convert_secs2time
from utils import split_str2indexes
from procedures.advanced_main import basic_train_fn, basic_eval_fn
from procedures.metric_utils import SaveMetric, MSEMetric, ComposeMetric
from datasets.synthetic_core import get_synthetic_env
from models.xcore import get_model
from xlayers import super_core, trunc_normal_
from lfna_utils import lfna_setup, train_model, TimeData
from lfna_models import HyperNet
def main(args):
logger, env_info, model_kwargs = lfna_setup(args)
dynamic_env = env_info["dynamic_env"]
2021-05-12 14:32:50 +02:00
model = get_model(**model_kwargs)
2021-05-12 09:45:45 +02:00
criterion = torch.nn.MSELoss()
shape_container = model.get_w_container().to_shape_container()
2021-05-13 10:39:19 +02:00
hypernet = HyperNet(
shape_container, args.hidden_dim, args.task_dim, len(dynamic_env)
)
2021-05-12 11:36:04 +02:00
hypernet = hypernet.to(args.device)
2021-05-13 05:40:04 +02:00
logger.log(
"{:} There are {:} weights in the base-model.".format(
time_string(), model.numel()
)
)
logger.log(
"{:} There are {:} weights in the meta-model.".format(
time_string(), hypernet.numel()
)
)
2021-05-13 10:39:19 +02:00
for i in range(len(dynamic_env)):
2021-05-13 05:40:04 +02:00
env_info["{:}-x".format(i)] = env_info["{:}-x".format(i)].to(args.device)
env_info["{:}-y".format(i)] = env_info["{:}-y".format(i)].to(args.device)
2021-05-13 10:39:19 +02:00
logger.log("{:} Convert to device-{:} done".format(time_string(), args.device))
2021-05-12 14:32:50 +02:00
2021-05-13 09:57:41 +02:00
optimizer = torch.optim.Adam(
hypernet.parameters(), lr=args.init_lr, weight_decay=1e-5, amsgrad=True
)
2021-05-12 09:45:45 +02:00
lr_scheduler = torch.optim.lr_scheduler.MultiStepLR(
optimizer,
milestones=[
int(args.epochs * 0.8),
int(args.epochs * 0.9),
],
gamma=0.1,
)
# LFNA meta-training
per_epoch_time, start_time = AverageMeter(), time.time()
2021-05-13 10:39:19 +02:00
last_success_epoch = 0
2021-05-12 09:45:45 +02:00
for iepoch in range(args.epochs):
need_time = "Time Left: {:}".format(
convert_secs2time(per_epoch_time.avg * (args.epochs - iepoch), True)
)
head_str = (
"[{:}] [{:04d}/{:04d}] ".format(time_string(), iepoch, args.epochs)
+ need_time
)
2021-05-13 10:39:19 +02:00
# One Epoch
loss_meter = AverageMeter()
for istep in range(args.per_epoch_step):
losses = []
for ibatch in range(args.meta_batch):
cur_time = random.randint(0, len(dynamic_env) - 1)
cur_container = hypernet(cur_time)
cur_x = env_info["{:}-x".format(cur_time)]
cur_y = env_info["{:}-y".format(cur_time)]
cur_dataset = TimeData(cur_time, cur_x, cur_y)
preds = model.forward_with_container(cur_dataset.x, cur_container)
optimizer.zero_grad()
loss = criterion(preds, cur_dataset.y)
losses.append(loss)
final_loss = torch.stack(losses).mean()
final_loss.backward()
optimizer.step()
lr_scheduler.step()
loss_meter.update(final_loss.item())
success, best_score = hypernet.save_best(-loss_meter.avg)
2021-05-13 10:07:54 +02:00
if success:
logger.log("Achieve the best with best_score = {:.3f}".format(best_score))
2021-05-13 10:39:19 +02:00
last_success_epoch = iepoch
if iepoch - last_success_epoch >= args.early_stop_thresh:
2021-05-13 10:07:54 +02:00
logger.log("Early stop at {:}".format(iepoch))
break
2021-05-13 10:39:19 +02:00
logger.log(
head_str
+ " meta-loss: {:.4f} ({:.4f}) :: lr={:.5f}, batch={:}".format(
loss_meter.avg,
loss_meter.val,
min(lr_scheduler.get_last_lr()),
len(losses),
2021-05-12 09:45:45 +02:00
)
2021-05-13 10:39:19 +02:00
)
2021-05-12 09:45:45 +02:00
2021-05-13 10:39:19 +02:00
save_checkpoint(
{
"hypernet": hypernet.state_dict(),
"lr_scheduler": lr_scheduler.state_dict(),
"iepoch": iepoch,
},
logger.path("model"),
logger,
)
2021-05-12 09:45:45 +02:00
per_epoch_time.update(time.time() - start_time)
start_time = time.time()
print(model)
print(hypernet)
2021-05-13 09:57:41 +02:00
hypernet.load_best()
2021-05-12 09:45:45 +02:00
2021-05-12 14:32:50 +02:00
w_container_per_epoch = dict()
2021-05-13 10:39:19 +02:00
for idx in range(0, env_info["total"]):
2021-05-12 14:32:50 +02:00
future_x = env_info["{:}-x".format(idx)]
future_y = env_info["{:}-y".format(idx)]
2021-05-13 09:57:41 +02:00
future_container = hypernet(idx)
2021-05-12 14:32:50 +02:00
w_container_per_epoch[idx] = future_container.no_grad_clone()
with torch.no_grad():
future_y_hat = model.forward_with_container(
future_x, w_container_per_epoch[idx]
)
future_loss = criterion(future_y_hat, future_y)
logger.log("meta-test: [{:03d}] -> loss={:.4f}".format(idx, future_loss.item()))
save_checkpoint(
{"w_container_per_epoch": w_container_per_epoch},
logger.path(None) / "final-ckp.pth",
logger,
)
2021-05-12 09:45:45 +02:00
logger.log("-" * 200 + "\n")
logger.close()
if __name__ == "__main__":
parser = argparse.ArgumentParser("Use the data in the past.")
parser.add_argument(
"--save_dir",
type=str,
2021-05-13 10:39:19 +02:00
default="./outputs/lfna-synthetic/lfna-debug-hpnet",
2021-05-12 09:45:45 +02:00
help="The checkpoint directory.",
)
parser.add_argument(
"--env_version",
type=str,
required=True,
help="The synthetic enviornment version.",
)
parser.add_argument(
"--hidden_dim",
type=int,
required=True,
help="The hidden dimension.",
)
#####
parser.add_argument(
"--init_lr",
type=float,
2021-05-13 10:39:19 +02:00
default=0.01,
2021-05-12 09:45:45 +02:00
help="The initial learning rate for the optimizer (default is Adam)",
)
parser.add_argument(
"--meta_batch",
type=int,
default=64,
help="The batch size for the meta-model",
)
2021-05-13 10:39:19 +02:00
parser.add_argument(
"--early_stop_thresh",
type=int,
default=100,
help="The maximum epochs for early stop.",
)
2021-05-12 09:45:45 +02:00
parser.add_argument(
"--epochs",
type=int,
default=2000,
help="The total number of epochs.",
)
2021-05-13 10:39:19 +02:00
parser.add_argument(
"--per_epoch_step",
type=int,
default=20,
help="The total number of epochs.",
)
2021-05-12 11:36:04 +02:00
parser.add_argument(
"--device",
type=str,
default="cpu",
help="",
)
2021-05-12 09:45:45 +02:00
# Random Seed
parser.add_argument("--rand_seed", type=int, default=-1, help="manual seed")
args = parser.parse_args()
if args.rand_seed is None or args.rand_seed < 0:
args.rand_seed = random.randint(1, 100000)
assert args.save_dir is not None, "The save dir argument can not be None"
2021-05-13 10:39:19 +02:00
args.task_dim = args.hidden_dim
2021-05-12 09:45:45 +02:00
args.save_dir = "{:}-{:}-d{:}".format(
args.save_dir, args.env_version, args.hidden_dim
)
main(args)