def single_eval(args=None):
if args is None:
args = command_parser.parse_arguments()
create_shared_model = model_class(args.model)
init_agent = agent_class(args.agent_type)
args.phase = 'eval'
args.episode_type = 'TestValEpisode'
args.test_or_val = 'test'
# if args.num_category != 60:
# args.detection_feature_file_name = 'det_feature_{}_categories.hdf5'.format(args.num_category)
start_time = time.time()
local_start_time_str = time.strftime('%Y_%m_%d_%H_%M_%S',
time.localtime(start_time))
tb_log_dir = args.log_dir + "/" + args.title + '_' + args.phase + '_' + local_start_time_str
log_writer = SummaryWriter(log_dir=tb_log_dir)
checkpoint = args.load_model
model = os.path.join(args.save_model_dir, checkpoint)
args.load_model = model
# run eval on model
# args.test_or_val = "val"
main_eval(args, create_shared_model, init_agent)
def main():
args = flag_parser.parse_arguments()
create_shared_model = model_class(args.model)
init_agent = agent_class(args.agent_type)
args.episode_type = "TestValEpisode"
args.test_or_val = "test"
# Get all valid saved_models for the given title and sort by train_ep.
checkpoints = [(f, f.split("_")) for f in os.listdir(args.save_model_dir)]
checkpoints = [
(f, int(s[-3])) for (f, s) in checkpoints
if len(s) >= 4 and f.startswith(args.title) and f.endswith('dat')
]
checkpoints.sort(key=lambda x: x[1])
best_model_on_val = None
best_performance_on_val = 0.0
for (f, train_ep) in tqdm(checkpoints, desc="Checkpoints."):
model = os.path.join(args.save_model_dir, f)
args.load_model = model
# run eval on model
args.test_or_val = "test"
main_eval(args, create_shared_model, init_agent)
# check if best on val.
with open(args.results_json, "r") as f:
results = json.load(f)
if results["success"] > best_performance_on_val:
best_model_on_val = model
best_performance_on_val = results["success"]
args.test_or_val = "test"
args.load_model = best_model_on_val
main_eval(args, create_shared_model, init_agent)
with open(args.results_json, "r") as f:
results = json.load(f)
print(
tabulate(
[
["SPL >= 1:", results["GreaterThan/1/spl"]],
["Success >= 1:", results["GreaterThan/1/success"]],
["SPL >= 5:", results["GreaterThan/5/spl"]],
["Success >= 5:", results["GreaterThan/5/success"]],
],
headers=["Metric", "Result"],
tablefmt="orgtbl",
))
print("Best model:", args.load_model)
def main():
setproctitle.setproctitle("Train/Test Manager")
args = flag_parser.parse_arguments()
if args.model == "BaseModel" or args.model == "GCN":
args.learned_loss = False
args.num_steps = 50
target = nonadaptivea3c_val if args.eval else nonadaptivea3c_train
# else:
# args.learned_loss = True
# args.num_steps = 6
# target = savn_val if args.eval else savn_train
create_shared_model = model_class(args.model)
init_agent = agent_class(args.agent_type)
optimizer_type = optimizer_class(args.optimizer)
# print('shared model created')
if args.eval:
main_eval(args, create_shared_model, init_agent)
return
start_time = time.time()
local_start_time_str = time.strftime("%Y-%m-%d_%H:%M:%S",
time.localtime(start_time))
np.random.seed(args.seed)
torch.manual_seed(args.seed)
random.seed(args.seed)
# print('seeding done')
if args.log_dir is not None:
tb_log_dir = args.log_dir + "/" + args.title + "-" + local_start_time_str
log_writer = SummaryWriter(log_dir=tb_log_dir)
else:
log_writer = SummaryWriter(comment=args.title)
if args.gpu_ids == -1:
args.gpu_ids = [-1]
else:
# print('something to do with cuda')
torch.cuda.manual_seed(args.seed)
mp.set_start_method("spawn")
shared_model = create_shared_model(args)
train_total_ep = 0
n_frames = 0
if shared_model is not None:
# print('shared model is being created')
shared_model.share_memory()
optimizer = optimizer_type(
filter(lambda p: p.requires_grad, shared_model.parameters()), args)
optimizer.share_memory()
print(shared_model)
# print('!!!!!!!!!!!!')
else:
assert (args.agent_type == "RandomNavigationAgent"
), "The model is None but agent is not random agent"
optimizer = None
processes = []
end_flag = mp.Value(ctypes.c_bool, False)
train_res_queue = mp.Queue()
for rank in range(0, args.workers):
print('Process {} being created'.format(rank))
p = mp.Process(
target=target,
args=(
rank,
args,
create_shared_model,
shared_model,
init_agent,
optimizer,
train_res_queue,
end_flag,
),
)
p.start()
processes.append(p)
time.sleep(0.1)
print("Train agents created.")
train_thin = args.train_thin
train_scalars = ScalarMeanTracker()
try:
while train_total_ep < args.max_ep:
print('total train ep: {} of {}'.format(train_total_ep,
args.max_ep))
print('Cuda available: {}'.format(torch.cuda.is_available()))
train_result = train_res_queue.get()
print('Got the train result from the queue')
train_scalars.add_scalars(train_result)
train_total_ep += 1
n_frames += train_result["ep_length"]
if (train_total_ep % train_thin) == 0:
log_writer.add_scalar("n_frames", n_frames, train_total_ep)
tracked_means = train_scalars.pop_and_reset()
for k in tracked_means:
log_writer.add_scalar(k + "/train", tracked_means[k],
train_total_ep)
if (train_total_ep % args.ep_save_freq) == 0:
print(n_frames)
if not os.path.exists(args.save_model_dir):
os.makedirs(args.save_model_dir)
state_to_save = shared_model.state_dict()
save_path = os.path.join(
args.save_model_dir,
"{0}_{1}_{2}_{3}.dat".format(args.title, n_frames,
train_total_ep,
local_start_time_str),
)
torch.save(state_to_save, save_path)
finally:
log_writer.close()
end_flag.value = True
for p in processes:
time.sleep(0.1)
p.join()
def main():
setproctitle.setproctitle("Train/Test Manager")
args = flag_parser.parse_arguments()
if args.model == "BaseModel" or args.model == "GCN_MLP" or args.model == "GCN" or args.model == "GCN_GRU":
args.learned_loss = False
args.num_steps = 50
target = nonadaptivea3c_val if args.eval else nonadaptivea3c_train
create_shared_model = model_class(args.model)
init_agent = agent_class(args.agent_type)
optimizer_type = optimizer_class(args.optimizer)
if args.eval:
main_eval(args, create_shared_model, init_agent)
return
model_to_open = args.load_model
if model_to_open != "":
shared_model = create_shared_model(args)
optimizer = optimizer_type(
filter(lambda p: p.requires_grad, shared_model.parameters()), args)
saved_state = torch.load(model_to_open,
map_location=lambda storage, loc: storage)
shared_model.load_state_dict(saved_state['model'])
optimizer.load_state_dict(saved_state['optimizer'])
optimizer.share_memory()
train_total_ep = saved_state['train_total_ep']
n_frames = saved_state['n_frames']
else:
shared_model = create_shared_model(args)
train_total_ep = 0
n_frames = 0
if shared_model is not None:
shared_model.share_memory()
optimizer = optimizer_type(
filter(lambda p: p.requires_grad, shared_model.parameters()),
args)
optimizer.share_memory()
print(shared_model)
else:
assert (args.agent_type == "RandomNavigationAgent"
), "The model is None but agent is not random agent"
optimizer = None
processes = []
end_flag = mp.Value(ctypes.c_bool, False)
train_res_queue = mp.Queue()
start_time = time.time()
local_start_time_str = time.strftime("%Y-%m-%d_%H:%M:%S",
time.localtime(start_time))
np.random.seed(args.seed)
torch.manual_seed(args.seed)
random.seed(args.seed)
if args.log_dir is not None:
tb_log_dir = args.log_dir + "/" + args.title + "-" + local_start_time_str
log_writer = SummaryWriter(log_dir=tb_log_dir)
else:
log_writer = SummaryWriter(comment=args.title)
if args.gpu_ids == -1:
args.gpu_ids = [-1]
else:
torch.cuda.manual_seed(args.seed)
for rank in range(0, args.workers):
p = mp.Process(
target=target,
args=(
rank,
args,
create_shared_model,
shared_model,
init_agent,
optimizer,
train_res_queue,
end_flag,
),
)
p.start()
processes.append(p)
time.sleep(0.1)
print("Train agents created.")
train_thin = args.train_thin
train_scalars = ScalarMeanTracker()
print(train_total_ep)
print(optimizer)
try:
while train_total_ep < args.max_ep:
train_result = train_res_queue.get()
train_scalars.add_scalars(train_result)
train_total_ep += 1
n_frames += train_result["ep_length"]
if (train_total_ep % train_thin) == 0:
log_writer.add_scalar("n_frames", n_frames, train_total_ep)
tracked_means = train_scalars.pop_and_reset()
for k in tracked_means:
log_writer.add_scalar(k + "/train", tracked_means[k],
train_total_ep)
if (train_total_ep % args.ep_save_freq) == 0:
print(n_frames)
if not os.path.exists(args.save_model_dir):
os.makedirs(args.save_model_dir)
state_to_save = shared_model.state_dict()
save_path = os.path.join(
args.save_model_dir,
"{0}_{1}_{2}_{3}.dat".format(args.title, n_frames,
train_total_ep,
local_start_time_str),
)
save_dict = {
'model': state_to_save,
'train_total_ep': train_total_ep,
'optimizer': optimizer.state_dict(),
'n_frames': n_frames
}
torch.save(save_dict, save_path)
#torch.save(state_to_save, save_path)
finally:
log_writer.close()
end_flag.value = True
for p in processes:
time.sleep(0.1)
p.join()
def main():
# 设置进程名称
setproctitle.setproctitle("Train/Test Manager")
# 获取命令行参数
args = flag_parser.parse_arguments()
if args.model == "SAVN":
args.learned_loss = True
args.num_steps = 6
target = savn_val if args.eval else savn_train
else:
args.learned_loss = False
args.num_steps = args.max_episode_length
target = nonadaptivea3c_val if args.eval else nonadaptivea3c_train
# 检查pinned_scene 和 data_source 是否冲突
if args.data_source == "ithor" and args.pinned_scene == True:
raise Exception(
"Cannot set pinned_scene to true when using ithor dataset")
# 获取模型对象类别， 未创建对象 e.g. <class 'models.basemodel.BaseModel'>
create_shared_model = model_class(args.model)
# 获取agent类别，未创建对象 default <class 'agents.navigation_agent.NavigationAgent'>
init_agent = agent_class(args.agent_type)
# 获取优化器对象类别，未创建对象 default <class 'optimizers.shared_adam.SharedAdam'>
optimizer_type = optimizer_class(args.optimizer)
######################## 测试阶段 ################################
if args.eval:
main_eval(args, create_shared_model, init_agent)
return
####################### 训练阶段 #################################
start_time = time.time()
local_start_time_str = time.strftime("%Y-%m-%d_%H:%M:%S",
time.localtime(start_time))
np.random.seed(args.seed)
torch.manual_seed(args.seed)
random.seed(args.seed)
# 设置日志参数
if args.log_dir is not None:
tb_log_dir = args.log_dir + "/" + args.title + "-" + local_start_time_str
log_writer = SummaryWriter(log_dir=tb_log_dir)
else:
log_writer = SummaryWriter(comment=args.title)
if args.gpu_ids == -1:
args.gpu_ids = [-1]
else:
torch.cuda.manual_seed(args.seed)
mp.set_start_method("spawn")
# 创建一个 torch.nn.Module的子类对象
shared_model = create_shared_model(args)
optimizer = optimizer_type(
filter(lambda p: p.requires_grad, shared_model.parameters()), args)
# 加载预先保存的模型
train_total_ep, n_frames = load_checkpoint(args, shared_model, optimizer)
# TODO: delete this after debug
# train_total_ep = 1000001
if shared_model is not None:
# 模型在多进程间共享参数 这个参数是torch.mutiprocessing 调用fork之前必须调用的方法
shared_model.share_memory()
# 创建一个 torch.optim.Optimizer的子类对象
# filter 函数把model中所有需要梯度更新的变量 作为参数送到optimizer的constructor中
optimizer.share_memory()
print(shared_model)
else:
assert (args.agent_type == "RandomNavigationAgent"
), "The model is None but agent is not random agent"
optimizer = None
processes = []
end_flag = mp.Value(ctypes.c_bool, False)
global_ep = mp.Value(ctypes.c_int)
global_ep.value = train_total_ep
# 多进程共享资源队列
train_res_queue = mp.Queue()
# 创建多进程
# target 进程执行目标函数
#
for rank in range(0, args.workers):
p = mp.Process(
target=target,
args=(rank, args, create_shared_model, shared_model, init_agent,
optimizer, train_res_queue, end_flag, global_ep),
)
p.start()
processes.append(p)
time.sleep(0.1)
print("Train agents created.")
train_thin = args.train_thin
train_scalars = ScalarMeanTracker()
# 主线程
try:
while train_total_ep < args.max_ep:
train_result = train_res_queue.get()
train_scalars.add_scalars(train_result)
train_total_ep += 1
global_ep.value = train_total_ep
n_frames += train_result["ep_length"]
if (train_total_ep % train_thin) == 0:
log_writer.add_scalar("n_frames", n_frames, train_total_ep)
tracked_means = train_scalars.pop_and_reset()
for k in tracked_means:
log_writer.add_scalar(k + "/train", tracked_means[k],
train_total_ep)
if (train_total_ep % args.ep_save_freq) == 0:
print(n_frames)
if not os.path.exists(args.save_model_dir):
os.makedirs(args.save_model_dir)
state_to_save = shared_model.state_dict()
save_path = os.path.join(
args.save_model_dir,
"{0}_{1}_{2}_{3}.dat".format(args.title, n_frames,
train_total_ep,
local_start_time_str),
)
torch.save(state_to_save, save_path)
if (train_total_ep % args.ep_save_ckpt) == 0:
print("save check point at episode {}".format(train_total_ep))
checkpoint = {
'train_total_ep': train_total_ep,
'n_frames': n_frames,
'shared_model': shared_model.state_dict(),
'optimizer': optimizer.state_dict()
}
checkpoint_path = os.path.join(args.save_model_dir,
"checkpoint.dat")
torch.save(checkpoint, checkpoint_path)
finally:
log_writer.close()
end_flag.value = True
for p in processes:
time.sleep(0.1)
p.join()
def main():
torch.backends.cudnn.deterministic = True
torch.backends.cudnn.benchmark = False
args = flag_parser.parse_arguments()
create_shared_model = model_class(args.model)
init_agent = agent_class(args.agent_type)
args.episode_type = "TestValEpisode"
args.test_or_val = "val"
tb_log_dir = args.log_dir + "/" + '{}_{}_{}'.format(
args.title, args.test_or_val,
time.strftime("%Y-%m-%d_%H:%M:%S", time.localtime(time.time())))
log_writer = SummaryWriter(log_dir=tb_log_dir)
print('Start Loading!')
optimal_action_path = './data/AI2thor_Combine_Dataset/Optimal_Path_Combine.json'
with open(optimal_action_path, 'r') as read_file:
optimal_action_dict = json.load(read_file)
manager = Manager()
optimal_action = manager.dict()
optimal_action.update(optimal_action_dict)
glove_file_path = './data/AI2thor_Combine_Dataset/det_feature_512_eval.hdf5'
glove_file = hdf5_to_dict(glove_file_path)
print('Loading Success!')
# Get all valid saved_models for the given title and sort by train_ep.
checkpoints = [(f, f.split("_")) for f in os.listdir(args.save_model_dir)]
checkpoints = [(f, int(s[-3])) for (f, s) in checkpoints
if len(s) >= 4 and f.startswith(args.title)]
checkpoints.sort(key=lambda x: x[1])
best_model_on_val = None
best_performance_on_val = 0.0
for (f, train_ep) in tqdm(checkpoints, desc="Checkpoints."):
model = os.path.join(args.save_model_dir, f)
args.load_model = model
# run eval on model
args.test_or_val = "val"
main_eval(args, create_shared_model, init_agent, glove_file,
optimal_action)
# check if best on val.
with open(args.results_json, "r") as f:
results = json.load(f)
if results["success"] > best_performance_on_val:
best_model_on_val = model
best_performance_on_val = results["success"]
log_writer.add_scalar("val/success", results["success"], train_ep)
log_writer.add_scalar("val/spl", results["spl"], train_ep)
# run on test.
args.test_or_val = "test"
main_eval(args, create_shared_model, init_agent, glove_file,
optimal_action)
with open(args.results_json, "r") as f:
results = json.load(f)
log_writer.add_scalar("test/success", results["success"], train_ep)
log_writer.add_scalar("test/spl", results["spl"], train_ep)
args.record_route = True
args.test_or_val = "test"
args.load_model = best_model_on_val
main_eval(args, create_shared_model, init_agent, glove_file,
optimal_action)
with open(args.results_json, "r") as f:
results = json.load(f)
print(
tabulate(
[
["SPL >= 1:", results["GreaterThan/1/spl"]],
["Success >= 1:", results["GreaterThan/1/success"]],
["SPL >= 5:", results["GreaterThan/5/spl"]],
["Success >= 5:", results["GreaterThan/5/success"]],
],
headers=["Metric", "Result"],
tablefmt="orgtbl",
))
print("Best model:", args.load_model)
def main():
setproctitle.setproctitle("Train/Test Manager")
args = flag_parser.parse_arguments()
if args.model == "BaseModel" or args.model == "GCN":
args.learned_loss = False
args.num_steps = 50
target = nonadaptivea3c_val if args.eval else nonadaptivea3c_train
else:
args.learned_loss = True
args.num_steps = 6
target = savn_val if args.eval else savn_train
create_shared_model = model_class(args.model)
init_agent = agent_class(args.agent_type)
optimizer_type = optimizer_class(args.optimizer)
if args.eval:
main_eval(args, create_shared_model, init_agent)
return
start_time = time.time()
local_start_time_str = time.strftime("%Y-%m-%d_%H:%M:%S",
time.localtime(start_time))
np.random.seed(args.seed)
torch.manual_seed(args.seed)
random.seed(args.seed)
if args.log_dir is not None:
tb_log_dir = args.log_dir + "/" + args.title + "-" + local_start_time_str
log_writer = SummaryWriter(log_dir=tb_log_dir)
else:
log_writer = SummaryWriter(comment=args.title)
if args.gpu_ids == -1:
args.gpu_ids = [-1]
else:
torch.cuda.manual_seed(args.seed)
mp.set_start_method("spawn")
shared_model = create_shared_model(args)
train_total_ep = 0
n_frames = 0
if shared_model is not None:
shared_model.share_memory()
optimizer = optimizer_type(
filter(lambda p: p.requires_grad, shared_model.parameters()), args)
optimizer.share_memory()
print(shared_model)
else:
assert (args.agent_type == "RandomNavigationAgent"
), "The model is None but agent is not random agent"
optimizer = None
processes = []
print('Start Loading!')
optimal_action_path = './data/AI2thor_Combine_Dataset/Optimal_Path_Combine.json'
with open(optimal_action_path, 'r') as read_file:
optimal_action_dict = json.load(read_file)
manager = Manager()
optimal_action = manager.dict()
optimal_action.update(optimal_action_dict)
glove_file_path = './data/AI2thor_Combine_Dataset/det_feature_512_train.hdf5'
glove_file = hdf5_to_dict(glove_file_path)
# det_gt_path = './data/AI2thor_Combine_Dataset/Instance_Detection_Combine.pkl'
# with open(det_gt_path, 'rb') as read_file:
# det_gt = pickle.load(read_file)
print('Loading Success!')
end_flag = mp.Value(ctypes.c_bool, False)
train_res_queue = mp.Queue()
for rank in range(0, args.workers):
p = mp.Process(
target=target,
args=(
rank,
args,
create_shared_model,
shared_model,
init_agent,
optimizer,
train_res_queue,
end_flag,
glove_file,
optimal_action,
# det_gt,
),
)
p.start()
processes.append(p)
time.sleep(0.1)
print("Train agents created.")
train_thin = args.train_thin
train_scalars = ScalarMeanTracker()
# start_ep_time = time.time()
try:
while train_total_ep < args.max_ep:
train_result = train_res_queue.get()
train_scalars.add_scalars(train_result)
train_total_ep += 1
n_frames += train_result["ep_length"]
# if train_total_ep % 10 == 0:
# print(n_frames / train_total_ep)
# print((time.time() - start_ep_time) / train_total_ep)
if (train_total_ep % train_thin) == 0:
log_writer.add_scalar("n_frames", n_frames, train_total_ep)
tracked_means = train_scalars.pop_and_reset()
for k in tracked_means:
log_writer.add_scalar(k + "/train", tracked_means[k],
train_total_ep)
if (train_total_ep % args.ep_save_freq) == 0:
print(n_frames)
if not os.path.exists(args.save_model_dir):
os.makedirs(args.save_model_dir)
state_to_save = shared_model.state_dict()
save_path = os.path.join(
args.save_model_dir,
"{0}_{1}_{2}_{3}.dat".format(args.title, n_frames,
train_total_ep,
local_start_time_str),
)
torch.save(state_to_save, save_path)
finally:
log_writer.close()
end_flag.value = True
for p in processes:
time.sleep(0.1)
p.join()
def full_eval(args=None):
if args is None:
args = command_parser.parse_arguments()
create_shared_model = model_class(args.model)
init_agent = agent_class(args.agent_type)
args.phase = 'eval'
args.episode_type = 'TestValEpisode'
args.test_or_val = 'val'
# if args.num_category != 60:
# args.detection_feature_file_name = 'det_feature_{}_categories.hdf5'.format(args.num_category)
start_time = time.time()
local_start_time_str = time.strftime('%Y_%m_%d_%H_%M_%S',
time.localtime(start_time))
tb_log_dir = args.log_dir + "/" + args.title + '_' + args.phase + '_' + local_start_time_str
log_writer = SummaryWriter(log_dir=tb_log_dir)
# Get all valid saved_models for the given title and sort by train_ep.
checkpoints = [(f, f.split("_")) for f in os.listdir(args.save_model_dir)]
checkpoints = [(f, int(s[-7])) for (f, s) in checkpoints
if len(s) >= 4 and f.startswith(args.title)
and int(s[-7]) >= args.test_start_from]
checkpoints.sort(key=lambda x: x[1])
best_model_on_val = None
best_performance_on_val = 0.0
for (f, train_ep) in tqdm(checkpoints, desc="Checkpoints."):
model = os.path.join(args.save_model_dir, f)
args.load_model = model
# run eval on model
# args.test_or_val = "val"
args.test_or_val = "test"
main_eval(args, create_shared_model, init_agent)
# check if best on val.
with open(args.results_json, "r") as f:
results = json.load(f)
if results["success"] > best_performance_on_val:
best_model_on_val = model
best_performance_on_val = results["success"]
log_writer.add_scalar("val/success", results["success"], train_ep)
log_writer.add_scalar("val/spl", results["spl"], train_ep)
if args.include_test:
args.test_or_val = "test"
main_eval(args, create_shared_model, init_agent)
with open(args.results_json, "r") as f:
results = json.load(f)
log_writer.add_scalar("test/success", results["success"], train_ep)
log_writer.add_scalar("test/spl", results["spl"], train_ep)
args.test_or_val = "test"
args.load_model = best_model_on_val
main_eval(args, create_shared_model, init_agent)
with open(args.results_json, "r") as f:
results = json.load(f)
print(
tabulate(
[
["SPL >= 1:", results["GreaterThan/1/spl"]],
["Success >= 1:", results["GreaterThan/1/success"]],
["SPL >= 5:", results["GreaterThan/5/spl"]],
["Success >= 5:", results["GreaterThan/5/success"]],
],
headers=["Metric", "Result"],
tablefmt="orgtbl",
))
print("Best model:", args.load_model)
def main():
setproctitle.setproctitle("Train/Test Manager")
args = command_parser.parse_arguments()
print('Training started from: {}'.format(
time.strftime('%Y-%m-%d %H:%M:%S', time.localtime(time.time())))
)
args.learned_loss = False
args.num_steps = 50
target = a3c_val if args.eval else a3c_train
if args.csiro:
args.data_dir = './data/'
else:
check_data(args)
scenes = loading_scene_list(args)
create_shared_model = model_class(args.model)
init_agent = agent_class(args.agent_type)
optimizer_type = optimizer_class(args.optimizer)
np.random.seed(args.seed)
torch.manual_seed(args.seed)
random.seed(args.seed)
if args.eval:
main_eval(args, create_shared_model, init_agent)
return
start_time = time.time()
local_start_time_str = time.strftime(
'%Y_%m_%d_%H_%M_%S', time.localtime(start_time)
)
tb_log_dir = args.log_dir + '/' + args.title + '_' + args.phase + '_' + local_start_time_str
log_writer = SummaryWriter(log_dir=tb_log_dir)
if args.gpu_ids == -1:
args.gpu_ids = [-1]
else:
torch.cuda.manual_seed(args.seed)
mp.set_start_method("spawn")
shared_model = create_shared_model(args)
train_total_ep = 0
n_frames = 0
if args.continue_training is not None:
saved_state = torch.load(
args.continue_training, map_location=lambda storage, loc: storage
)
shared_model.load_state_dict(saved_state)
train_total_ep = int(args.continue_training.split('_')[-7])
n_frames = int(args.continue_training.split('_')[-8])
if args.fine_tuning is not None:
saved_state = torch.load(
args.fine_tuning, map_location=lambda storage, loc: storage
)
model_dict = shared_model.state_dict()
pretrained_dict = {k: v for k, v in saved_state.items() if (k in model_dict and v.shape == model_dict[k].shape)}
model_dict.update(pretrained_dict)
shared_model.load_state_dict(model_dict)
if args.update_meta_network:
for layer, parameters in shared_model.named_parameters():
if not layer.startswith('meta'):
parameters.requires_grad = False
shared_model.share_memory()
if args.fine_tune_graph:
optimizer = optimizer_type(
[
{'params': [v for k, v in shared_model.named_parameters() if v.requires_grad and not k.startswith('graph')],
'lr': 0.00001},
{'params': [v for k, v in shared_model.named_parameters() if v.requires_grad and k.startswith('graph')],
'lr': args.lr},
]
)
else:
optimizer = optimizer_type(
[v for k, v in shared_model.named_parameters() if v.requires_grad], lr=args.lr
)
optimizer.share_memory()
print(shared_model)
processes = []
end_flag = mp.Value(ctypes.c_bool, False)
train_res_queue = mp.Queue()
for rank in range(0, args.workers):
p = mp.Process(
target=target,
args=(
rank,
args,
create_shared_model,
shared_model,
init_agent,
optimizer,
train_res_queue,
end_flag,
scenes,
),
)
p.start()
processes.append(p)
time.sleep(0.1)
print("Train agents created.")
train_thin = args.train_thin
train_scalars = ScalarMeanTracker()
try:
while train_total_ep < args.max_ep:
train_result = train_res_queue.get()
train_scalars.add_scalars(train_result)
train_total_ep += 1
n_frames += train_result['ep_length']
if (train_total_ep % train_thin) == 0:
log_writer.add_scalar('n_frames', n_frames, train_total_ep)
tracked_means = train_scalars.pop_and_reset()
for k in tracked_means:
log_writer.add_scalar(
k + '/train', tracked_means[k], train_total_ep
)
if (train_total_ep % args.ep_save_freq) == 0:
print('{}: {}'.format(train_total_ep, n_frames))
if not os.path.exists(args.save_model_dir):
os.makedirs(args.save_model_dir)
state_to_save = shared_model.state_dict()
save_path = os.path.join(
args.save_model_dir,
'{0}_{1}_{2}_{3}.dat'.format(
args.title, n_frames, train_total_ep, local_start_time_str
),
)
torch.save(state_to_save, save_path)
finally:
log_writer.close()
end_flag.value = True
for p in processes:
time.sleep(0.1)
p.join()
if args.test_after_train:
full_eval()
