net=net,
nb_actions=nb_actions,
gamma=params.gamma,
device=device,
)
if args.evaluate:
agent.net.load_state_dict(torch.load(args.evaluate))
env = make_env(params.env_name, episodic=False)
evaluate(agent, env, render=args.render)
exit()
if args.resume:
agent.load(args.resume[0])
memory = ExperienceBuffer(params.memory_capacity, obs_shape,
params.frame_stack)
opt = torch.optim.Adam(net.parameters(), lr=params.learning_rate)
eps_schedule = EpisilonAnnealer(params.epsilon_start, params.epsilon_end,
params.epsilon_frames)
save_id = str(datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
path_str = lambda p: p.absolute().as_posix(
) # converts a pathlib Path to string
# Create the run directories
runs_dir = Path(f"runs")
root_dir = runs_dir / f'{args.task}-{save_id}'
chk_dir = root_dir / 'checkpoints'
runs_dir.mkdir(exist_ok=True)
unroll_steps=params.unroll_steps,
device=device)
if args.evaluate:
agent.net.load_state_dict(torch.load(args.evaluate))
env = make_env(params.env_name, episodic=False)
evaluate(agent, env, render=args.render)
exit()
if args.resume:
agent.load(args.resume[0])
memory = ExperienceBuffer(params.memory_capacity,
obs_shape,
params.frame_stack,
prioritized=True,
unroll_steps=params.unroll_steps,
beta_start=params.beta_start,
beta_end=params.beta_end,
beta_steps=params.beta_frames)
opt = torch.optim.Adam(net.parameters(), lr=params.learning_rate)
save_id = str(datetime.datetime.now().strftime("%Y-%m-%d_%H-%M-%S"))
path_str = lambda p: p.absolute().as_posix(
) # converts a pathlib Path to string
# Create the run directories
runs_dir = Path(f"runs")
root_dir = runs_dir / f'{args.task}-{save_id}'
chk_dir = root_dir / 'checkpoints'