agent = TrueOnlineSarsaLambda(env.observation_space,
env.action_space,
alpha=args.alpha,
gamma=args.gamma,
epsilon=args.epsilon,
lamb=args.lamb)
for run in range(1, args.runs + 1):
obs = env.reset()
agent.reset()
done = False
if args.fixed:
while not done:
_, _, done, _ = env.step({})
else:
while not done:
action = agent.act(agent.get_features(obs))
next_obs, r, done, _ = env.step(action=action)
agent.learn(state=obs,
action=action,
reward=r,
next_state=next_obs)
obs = next_obs
env.save_csv(out_csv, run)
env.close()
initial_states = env.reset()
ql_agents = {ts: QLAgent(starting_state=env.encode(initial_states[ts]),
state_space=env.observation_space,
action_space=env.action_space,
alpha=args.alpha,
gamma=args.gamma,
exploration_strategy=EpsilonGreedy(initial_epsilon=args.epsilon, min_epsilon=args.min_epsilon, decay=args.decay)) for ts in env.ts_ids}
done = {'__all__': False}
infos = []
if args.fixed:
while not done['__all__']:
_, _, done, _ = env.step({})
else:
while not done['__all__']:
actions = {ts: ql_agents[ts].act() for ts in ql_agents.keys()}
s, r, done, _ = env.step(actions=actions)
if args.v:
print('s=', env.radix_decode(ql_agents['t'].state), 'a=', actions['t'], 's\'=', env.radix_encode(s['t']), 'r=', r['t'])
for agent_id in ql_agents.keys():
ql_agents[agent_id].learn(new_state=env.encode(s[agent_id]), reward=r[agent_id])
env.save_csv()
env.close()