def render(self, mode='human', close=False):
"""
Gym function called at the end of a process.
"""
print("Rendering")
if not self.validation_process:
np.save(self.test_folder + '/memory_' + str(self.test_starts_index) + '.npy', self.memory)
title_act = '/test_actions_' + str(self.test_starts_index)
plt.plot_actions(self.memory, self.long_actions, self.short_actions, self.test_folder, title_act)
def plot_actions(self):
plt.plot_actions(self.folder, self.memory, self.long_actions, self.short_actions)
def main():
flags = {
'players': 2,
'num_episodes': 1,
'num_test_episodes': 1,
'agent_class': 'SimpleAgent',
'alpha': 0.1,
'gamma': 0.9,
'seed': 12345
}
options, arguments = getopt.getopt(sys.argv[1:], '', [
'players=', 'num_episodes=', 'num_test_episodes=', 'agent_class=',
'seed=', 'alpha=', 'gamma='
])
if arguments:
sys.exit('usage: rl_env_example.py [options]\n'
'--players number of players in the game.\n'
'--num_episodes number of game episodes to run.\n'
'--num_test_episodes number of test game episodes to run.\n'
'--alpha step size for Q-learning. \n'
'--gamma discount rate for Q-learning. \n'
'--seed random generator seed. \n'
'--agent_class {}'.format(' or '.join(
AGENT_CLASSES.keys())))
for flag, value in options:
flag = flag[2:] # Strip leading --.
flags[flag] = type(flags[flag])(value)
random.seed(flags['seed'])
form_str = 'players_{}-episodes_{}-test_episodes_{}-agent_{}-alpha_{}-gamma_{}-seed_{}'
suffix = form_str.format(flags['players'], flags['num_episodes'],
flags['num_test_episodes'], flags['agent_class'],
flags['alpha'], flags['gamma'], flags['seed'])
train_suffix = suffix + '-train'
test_suffix = suffix + '-test'
runner = Runner(flags)
print('*' * 100)
print('TRAINING')
agents, rewards, agents_rewards, actions = runner.run(
flags['num_episodes'])
print('*' * 100)
player_num = len(agents)
for ar in agents_rewards:
plot_learning_rewards(agents_rewards[ar], ar, train_suffix)
plot_actions(actions[ar], ar, train_suffix)
plot_hist(rewards, -1, train_suffix, flags['players'])
if flags['agent_class'] == 'QAgent':
greedy_agents = [
GreedyAgent(runner.agent_config, agent.Q) for agent in agents
]
print('*' * 100)
print('TESTING')
_, rewards, agents_rewards, actions = runner.run(
flags['num_test_episodes'], greedy_agents)
print('*' * 100)
plot_hist(rewards, -1, test_suffix, flags['players'])
for ar in agents_rewards:
plot_actions(actions[ar], ar, test_suffix)
print('rewards:\n{}'.format(rewards))
print('agent 0 rewards:\n{}'.format(agents_rewards[0]))
print('agent 1 rewards:\n{}'.format(agents_rewards[1]))