def train(self):
policies = {0: self.policy.clone_policy()}
try:
for episode_number in range(1, self.episodes + 1):
print(f"training on episode {episode_number}/{self.episodes}")
episode = Episode(self.policy, self.epsilon)
episode_states, episode_distributions, rewards, winner = episode.play()
self.states += episode_states
self.distributions += episode_distributions
self.rewards += rewards
self.train_policy()
self.epsilon -= self.epsilon_decay_rate
if episode_number % (self.episodes // self.amount_of_players) == 0:
policy_to_save = self.policy.clone_policy()
policies[episode_number] = policy_to_save
except (KeyboardInterrupt):
policy_to_save = self.policy.clone_policy()
policies[episode_number] = policy_to_save
return policies
return policies
