def test_action_encode():
env = SoccerEnv()
action1, action2 = 1, 2
x = env.encode_action(1,2)
assert (action1, action2) == env.decode_action(x)
# Q errors for plotting
Foe_Q_P1_Q_errors = []
for i_episode in range(n_episodes_MAX):
state = env.reset()
P1_Q_ref = Foe_Q_agent.Q[ref_state, ref_P1_action, ref_P2_action]
for t in range(steps_MAX):
joint_action = np.random.randint(num_actions)
# Take action A, observe R, S'
state_new, reward, done, info = env.step(joint_action)
# Update Q
P1_action, P2_action = env.decode_action(joint_action)
P1_reward, P2_reward = env.decode_reward(state, reward)
Foe_Q_agent.learn(P1_reward, state, state_new, P1_action, P2_action)
state = state_new
if done:
# if verbose:
# print("Episode finished after {} timesteps".format(t + 1))
break
# calc error at end of episode update
Foe_Q_P1_Q_errors.append(
np.abs(Foe_Q_agent.Q[ref_state, ref_P1_action, ref_P2_action] -
P1_Q_ref))