Off Policy Trainer self._max_episode_len does not update
bscoventry opened this issue · comments
Hello,
I'm running into an interesting issue where self._max_episode_len is always [] in /environments/time_limit.py
System:
Windows 10 (most recent update)
GPU: GTX1080
CPU: Core i7
Relevant Code:
`if name == 'main':
"""Setup simulation variables"""
envRender = True
envLogMode = ['stdout','tensorboard','csv']
save_intervals = 100
envLogDir = "./logs"
saveLog = "./models"
maxTimeSteps = 1000
max_ep_len = 1000
"""Setup simulation: Setup as a TD3 agent"""
env = VectorEnv("spiker-v0")
agent = TD3("mlp",env)
trainer = OffPolicyTrainer(agent,env,max_timesteps = maxTimeSteps,render = envRender,log_mode = envLogMode, log_interval = save_intervals, logdir = envLogDir, save_model = saveLog, max_ep_len = max_ep_len)
"""Run the simulation"""
trainer.train()
trainer.evaluate()`
spiker-v0 is a custom gym environment, but this also occurs on more standard datasets. Current work around is to set the initialization of max_episode_len on the initializer, but this is not satisfying.
Relevant error:
Traceback (most recent call last): File "c:/CodeRepos/gym-spiker/SpikerNet_Main.py", line 33, in <module> trainer.train() File "C:\Users\bcovent\AppData\Local\Programs\Python\Python36\lib\site-packages\genrl\trainers\offpolicy.py", line 92, in train next_state, reward, done, _ = self.env.step(action) File "C:\Users\bcovent\AppData\Local\Programs\Python\Python36\lib\site-packages\genrl\environments\vec_env\vector_envs.py", line 159, in step obs, reward, done, info = env.step(actions[i]) File "C:\Users\bcovent\AppData\Local\Programs\Python\Python36\lib\site-packages\genrl\environments\gym_wrapper.py", line 92, in step self.state, self.reward, self.done, self.info = self.env.step(action) File "C:\Users\bcovent\AppData\Local\Programs\Python\Python36\lib\site-packages\genrl\environments\time_limit.py", line 20, in step if self._steps_taken >= self._max_episode_len: TypeError: '>=' not supported between instances of 'int' and 'NoneType'
No matter how it is set in trainer init the same error occurs.
Thank you for your time