while not terminated_1 and not terminated_2:
# the first agent
# print("agent 1")
action_id = agent_1.action_based_on_policy(state_1, env)
one_hot_action = one_hot(action_id, nr_actions)
new_state, reward, terminated_1 = env.step(action_id, state_1)
scaled_state_1 = scale_state(state_1, env)
#histories_1.appending(reward, scaled_state_1, one_hot_action)
plt.scatter(state_1[0, 0], state_1[0, 1], s=100, c='#C1C7C9', marker='s')
plt.scatter(new_state[0, 0], new_state[0, 1], s=50, c='red')
plt.show()
plt.pause(0.1)
state_1, steps_1 = update_state_step(new_state, steps_1)
# the second agent
# print("agent 2")
action_id = agent_2.action_based_on_policy(state_2, env)
one_hot_action = one_hot(action_id, nr_actions)
new_state, reward, terminated_2 = env.step(action_id, state_2)
scaled_state_2 = scale_state(state_2, env)
#histories_2.appending(reward, scaled_state_2, one_hot_action)
plt.scatter(state_2[0, 0], state_2[0, 1], s=100, c='#C1C7C9', marker='s')
plt.scatter(new_state[0, 0], new_state[0, 1], s=50, c='blue')
plt.show()
plt.pause(0.1)
state_2, steps_2 = update_state_step(new_state, steps_2)
while not terminated:
action_id = agent_ler.action_based_on_Q_target(state,
env,
epsilon=epsilon)
new_state, reward, terminated, info = env.step(action_id)
new_state = single_shape_adaptor(new_state, nr_features)
this_event = event(state, action_id, reward, new_state, terminated,
env)
replay_buffer.consider_this_event(this_event)
state, steps = update_state_step(new_state, steps)
print("... " + str(steps) +
" new event are added to the replay_buffer")
print("... updating the Q started")
for k in range(K):
current_batch = replay_buffer.return_a_batch(batchsize=64)
agent_ler.learn(current_batch, env)
print("... updating the Q finished")
print("... the Q-target update is started.")