def start():
init(autoreset=True)
c = Character()
with open('art/dungeon.txt') as f:
print(f.read())
print('')
print('#' * 80 + '\n#' + ' ' * 78 + '#' + '\n#\tWelcome to Command-Line Dungeon Diver! '\
'Begin by selecting a class: #\n' + '#' + ' ' * 78 + '#' + '\n' + '#' * 80)
announce('Mage, Warrior, Fighter, Cleric, Ranger or Paladin?')
classtoplay = input('>>> ')
character = dict()
if classtoplay.lower() == 'mage':
character['stats'] = c.mage()
elif classtoplay.lower() == 'warrior':
character['stats'] = c.warrior()
elif classtoplay.lower() == 'fighter':
character['stats'] = c.fighter()
elif classtoplay.lower() == 'cleric':
character['stats'] = c.cleric()
elif classtoplay.lower() == 'ranger':
character['stats'] = c.ranger()
elif classtoplay.lower() == 'paladin':
character['stats'] = c.paladin()
else:
start()
announce('Ah, a {classtoplay}. What shall we call you?'.format(classtoplay=classtoplay))
character['name'] = input('>>> ')
character['stats']['health'] = math.floor(character['stats']['vitality'] -
.15 * character['stats']['vitality'])
character['experience'] = 0
character['level'] = 1
character['type'] = classtoplay.lower()
announce('Here are your stats, {name}..\n'.format(name=character['name']))
for k, v in character['stats'].items():
announce('\t{stat}: {value}'.format(stat=k, value=v))
env = Environments(level=character['level'])
environment = env.dungeon()
announce("You've entered a {env}! Clear out all of the enemies!".format(env=environment['name']))
shuffle(environment['mobs'])
for mob in environment['mobs']:
announce('Uh-oh! {mobname} attacks!'.format(mobname=mob['name']))
if mob['name'] == 'Bat':
with open('art/bat.txt', mode='r') as f:
print(f.read())
elif mob['name'] == 'Skeleton':
with open('art/skeleton.txt', mode='r') as f:
print(f.read())
fight(character, mob)
announce("\nYou've cleared out all of the enemies, now its time for the boss!")
if environment['boss']['name'] == 'Ogre':
with open('art/ogre.txt', mode='r') as f:
print(f.read())
fight(character, environment['boss'])
def evaluate(name, environment, n=0, uncertainty=0):
disablePrint()
name = name + '-' + str(n)
agent = pickle.load(
open(f"outputs/{'-'.join(name.split('-')[:-1])}/Agents/{name}.agent",
"rb"))
agent.explore = agent.exploration.greedy
agent.uncertainty = True
agent.state_avoidance = False
agent.uncertainty_weight = uncertainty
env = Environments(render=False,
envs=[environment for _ in range(20)],
agent=agent)
rews, dones = [], []
for i in range(20000):
obs, hn, cn = env.start()
act, obs_old, h0, c0, hn, cn, _, _ = agent.chooseMulti(obs, hn, cn)
obs, rew, done, info = env.step(act, hn, cn)
rews.append(sum(rew))
dones.append(sum(done))
enablePrint()
score = sum(rews) / sum(dones)
print(
name.ljust(20, ' '),
int(100 * (score - rMin[environment]) /
(rMax[environment] - rMin[environment])) / 100, " ", score)
def main():
name, environment, hours, total_agents, done = params['name'], params['environment'], params['hours'], params['total_agents'], None
agent = Agent(**params)
env = Environments(render=False, envs=[environment for _ in range(total_agents)], agent=agent)
collector = Collector(**params)
tid, f = time() + 3600 * hours - 300, 0
while time() < tid:
f += 1
obs, hn, cn = env.start()
act, obs_old, h0, c0, hn, cn, before_trace, after_trace = agent.chooseMulti(obs, hn, cn, done=done)
obs, rew, done, info = env.step(act, hn, cn)
collector.collect(rew, done, act, agent.onpolicy)
if not agent.onpolicy and f > 10:
agent.rememberMulti(obs_old, act, obs, rew, h0, c0, hn, cn, done, before_trace, after_trace)
agent.learn()
saveAgent(agent, name)
saveCollector(collector, name)
x = self.decoder(x)
return x
total_agents, display_every = 20, 5000
agent = Agent(memory=50000,
discount=0.995,
uncertainty=False,
update_every=100,
double=True,
use_distribution=False,
reward_normalization=True)
env = Environments(render=True,
envs=[
'bigfish', 'bossfight', 'caveflyer', 'chaser',
'climber', 'coinrun', 'dodgeball', 'fruitbot', 'heist',
'jumper', 'leaper', 'maze', 'miner', 'ninja', 'plunder',
'starpilot'
],
agent=agent)
network = NetWork().to(device)
print("Number of parameters in network:", count_parameters(network))
print("Number of parameters in encoder:", count_parameters(network.encoder))
print("Number of parameters in decoder:", count_parameters(network.decoder))
criterion = MSELoss()
optimizer = Adam(network.parameters(), lr=1e-4, weight_decay=1e-5)
for f in range(0, 10000000):
obs, hn, cn = env.start()
act, obs_old, h0, c0, hn, cn = agent.chooseMulti(obs, hn, cn)
obs, rew, done, info = env.step(act, hn, cn)
agent.rememberMulti(obs_old, act, obs, rew, h0, c0, hn, cn, done)
obs, action, obs_next, reward, h0, c0, hn, sn, done = agent.memory.sample(
if keyboard.Key.f3 == key:
save = True
keyboard.Listener(on_press=on_press).start()
total_agents, display_every = 20, 5000
agent = Agent(memory=40000,
discount=0.995,
uncertainty=False,
update_every=100,
double=True,
use_distribution=False,
reward_normalization=False)
env = Environments(render=True,
envs=['maze' for _ in range(total_agents)],
agent=agent)
collector = Collector(calculate_every=500, total_agents=total_agents)
for f in range(10000000):
obs, hn, cn = env.start()
act, obs_old, h0, c0, hn, cn = agent.chooseMulti(obs, hn, cn)
obs, rew, done, info = env.step(act, hn, cn)
collector.collect(rew, done, act)
agent.rememberMulti(obs_old, act, obs, rew, h0, c0, hn, cn, done)
agent.learn()
if showPrint:
plt.close('all')
displayer(obs[0].cpu(), agent, collector)
fig = plt.figure()
move_figure(fig, 0, 0)
# env.close()
# obs = clean(env.reset())
# hn = torch.zeros(1, 1, hidden_size, device=device)
# cn = torch.zeros(1, 1, hidden_size, device=device)
# disablePrint()
# t0 = time.time()
# obs, hn, cn = [obs for _ in range(k)], [hn for _ in range(k)], [cn for _ in range(k)]
# for i in range(n // k):
# agent.chooseMulti(obs, hn, cn)
# t1 = time.time()
# enablePrint()
# print("Choose", t1 - t0) # 17.568519115447998
agent = Agent()
env = Environments(render=True, envs=['fruitbot' for _ in range(20)])
all_return, all_dones = [], []
update_every = 100
disablePrint()
frames = 1000000
dones, total_rew = 0, 0
for f in range(1, frames + 1):
obs, hn, cn = env.start()
act, obs_old, h0, c0, hn, cn = agent.chooseMulti(obs, hn, cn)
obs, rew, done, info = env.step(act, hn, cn)
total_rew += sum(rew) / len(rew)
dones += sum(done) / len(done)
agent.rememberMulti(obs_old, act, obs, rew, h0, c0, hn, cn, done)
if f > update_every:
for _ in range(2):
agent.learn(double=True)