y_t = rewards_vector + GAMMA * target_q_values
loss = critic.model.train_on_batch([states, actions], y_t)
losses.append(loss)
loss_writer.writerow({
'episode': episode,
'avg_reward': r_t,
'critic_loss': loss
}) # record losses
# Update actor
a_for_grad = actor.model.predict(states)
grads = critic.gradients(states, a_for_grad)
actor.train(states, grads)
# Update target networks
actor.update_actor_target()
critic.update_critic_target()
# Gradually decrease exploration
epsilon *= EPSILON_DECAY
# Print to terminal
print("Episode: ", episode)
print("Epsilon: ", epsilon)
# print("S_t", s_t)
print("Defender mu_sigma (a_t): ", a_t)
# print("Defender locations list: ", def_coords_list)
print("Defender average coords (row, col): ", def_avg_coords)
print("Average reward (r_t): ", r_t)
print("Critic Loss: ", loss)
print()
