Using "resume" with the async_sac_humanoid example

[cnatale]

I'm able to run the async_sac_humanoid example featured in the docs(https://slm-lab.gitbook.io/slm-lab/using-slm-lab/parallelizing-training-async-sac-on-humanoid), but attempting to continue training from a stopped trial results in the following error with stack trace included:

[2021-04-01 09:07:34,502 PID:6308 INFO run_lab.py get_spec_and_run] Running lab spec_file:slm_lab/spec/benchmark/async_sac/async_sac_roboschool.json spec_name:async_sac_humanoid in mode:train@latest
[2021-04-01 09:07:34,503 PID:6308 INFO run_lab.py get_spec] Resolved to train@data/async_sac_humanoid_2021_03_31_212345/
[2021-04-01 09:07:34,764 PID:6308 INFO logger.py info] Running distributed sessions
[2021-04-01 09:07:34,818 PID:6308 INFO openai.py __init__] OpenAIEnv:
- env_spec = {'max_frame': 50000000,
 'max_t': None,
 'name': 'RoboschoolHumanoid-v1',
 'num_envs': 8}
- eval_frequency = 1000
- log_frequency = 1000
- frame_op = None
- frame_op_len = None
- image_downsize = (84, 84)
- normalize_state = False
- reward_scale = None
- num_envs = 8
- name = RoboschoolHumanoid-v1
- max_t = 1000
- max_frame = 3125000
- to_render = False
- is_venv = True
- clock_speed = 8
- clock = <slm_lab.env.base.Clock object at 0x10de9bc18>
- done = False
- total_reward = nan
- u_env = <slm_lab.env.vec_env.ShmemVecEnv object at 0x13e3ba780>
- observation_space = Box(44,)
- action_space = Box(17,)
- observable_dim = {'state': 44}
- action_dim = 17
- is_discrete = False
[2021-04-01 09:07:35,057 PID:6308 INFO net_util.py load_algorithm] Loading algorithm SoftActorCritic nets ['net', 'q1_net', 'target_q1_net', 'q2_net', 'target_q2_net'] from data/async_sac_humanoid_2021_03_31_212345/model/async_sac_humanoid_t0_*.pt
Traceback (most recent call last):
  File "run_lab.py", line 99, in <module>
    main()
  File "run_lab.py", line 91, in main
    get_spec_and_run(*args)
  File "run_lab.py", line 75, in get_spec_and_run
    run_spec(spec, lab_mode)
  File "run_lab.py", line 58, in run_spec
    Trial(spec).run()
  File "/Users/natale/Projects/SLM-Lab/slm_lab/experiment/control.py", line 181, in run
    session_metrics_list = self.run_distributed_sessions()
  File "/Users/natale/Projects/SLM-Lab/slm_lab/experiment/control.py", line 170, in run_distributed_sessions
    global_nets = self.init_global_nets()
  File "/Users/natale/Projects/SLM-Lab/slm_lab/experiment/control.py", line 163, in init_global_nets
    session = Session(deepcopy(self.spec))
  File "/Users/natale/Projects/SLM-Lab/slm_lab/experiment/control.py", line 45, in __init__
    self.agent, self.env = make_agent_env(self.spec, global_nets)
  File "/Users/natale/Projects/SLM-Lab/slm_lab/experiment/control.py", line 19, in make_agent_env
    agent = Agent(spec, body=body, global_nets=global_nets)
  File "/Users/natale/Projects/SLM-Lab/slm_lab/agent/__init__.py", line 35, in __init__
    self.algorithm = AlgorithmClass(self, global_nets)
  File "/Users/natale/Projects/SLM-Lab/slm_lab/agent/algorithm/base.py", line 24, in __init__
    self.init_nets(global_nets)
  File "/Users/natale/Projects/SLM-Lab/slm_lab/agent/algorithm/sac.py", line 93, in init_nets
    self.end_init_nets()
  File "/Users/natale/Projects/SLM-Lab/slm_lab/agent/algorithm/base.py", line 50, in end_init_nets
    self.load()
  File "/Users/natale/Projects/SLM-Lab/slm_lab/agent/algorithm/base.py", line 102, in load
    net_util.load_algorithm(self)
  File "/Users/natale/Projects/SLM-Lab/slm_lab/agent/net/net_util.py", line 217, in load_algorithm
    load(net, model_path)
  File "/Users/natale/Projects/SLM-Lab/slm_lab/agent/net/net_util.py", line 203, in load
    net.load_state_dict(torch.load(util.smart_path(model_path), map_location=device))
  File "/Users/natale/miniconda3/envs/lab/lib/python3.7/site-packages/torch/serialization.py", line 419, in load
    f = open(f, 'rb')
FileNotFoundError: [Errno 2] No such file or directory: '/Users/natale/Projects/SLM-Lab/data/async_sac_humanoid_2021_03_31_212345/model/async_sac_humanoid_t0_net_model.pt'
(lab) Chris-Ns-MacBook-Pro:SLM-lab natale$ 

How would I continue training with this example?

[kengz]

Hi, seems like it did not find the checkpoint model weights �file (*.pt). This is saved during the checkpoint from this line, controlled by log_frequency in the spec file. Has your trial run for long enough to hit a checkpoint?