def configure_gym_marlo(self, env_name):
""" Configure environment for Marlo platform
Require previously launched minecraft environment
- to launch: $MALMO_MINECRAFT_ROOT/launchClient.sh -p 10000
Available params defined in:
- https://github.com/crowdAI/marLo/blob/8652f8daef2caf9202881d002a2d3c28c882d941/marlo/base_env_builder.py
"""
import marlo
marlo.logger.setLevel('ERROR')
# client_pool = [('127.0.0.1', 10000), ('127.0.0.1', 10001)]
client_pool = [('127.0.0.1', 10000), ('127.0.0.1', 10001)]
join_tokens = marlo.make(
env_name,
params={
"client_pool": client_pool,
"videoResolution": [84, 84],
"tick_length": 1,
# "prioritise_offscreen_rendering": False,
# "comp_all_commands": ['move', "turn"],
})
join_token = join_tokens[0]
env = marlo.init(join_token)
return env
def __init__(self):
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make('MarLo-CliffWalking-v0',
params={
"client_pool": client_pool,
"comp_all_commands": ["movenorth", "movesouth", "moveeast", "movewest"],
"prioritise_offscreen_rendering":False
})
join_token = join_tokens[0]
self.env = marlo.init(join_token)
self.env.params.suppress_info = False
self.env.mission_spec.setViewpoint(1)
for z in range(0,13,1):
x = 2
self.env.mission_spec.drawBlock( x,45,z,"lava")
# command_parser = commands.CommandParser()
# self.commands = command_parser.get_commands(self.env.params.mission_xml.encode('ascii'), self.env.params.role)
# self.commands_size = len(self.commands)
for _space, _commands in zip(self.env.action_spaces, self.env.action_names):
self.commands = _commands
self.commands_size = len(_commands) - 1
print(self.commands)
self.previousState = None
self.alpha = 0.1
self.gamma = 1.0
self.epsilon = 0.1
self.q_table = {}
self.test = 0
def make_env(self, env_id):
params = merge(params_default, {
'client_pool': self.client_pool
})
print("{} Minecraft clients available".format(len(self.client_pool)))
join_token = marlo.make(env_id,
params)
return marlo.init(join_token[0])
def make_env(env_seed=0):
join_tokens = marlo.make("MarLo-FindTheGoal-v0",
params=dict(
allowContinuousMovement=["move", "turn"],
videoResolution=[84, 84],
kill_clients_after_num_rounds=500))
env = marlo.init(join_tokens[0])
obs = env.reset()
# env.render(mode="rgb_array")
action = env.action_space.sample()
obs, r, done, info = env.step(action)
env.seed(int(env_seed))
return env
def make_env(env_seed=0):
join_tokens = marlo.make("MarLo-FindTheGoal-v0",
params=dict(
comp_all_commands=["move", "turn"],
allowContinuousMovement=True,
videoResolution=[336, 336],
))
env = marlo.init(join_tokens[0])
obs = env.reset()
action = env.action_space.sample()
obs, r, done, info = env.step(action)
env.seed(int(env_seed))
return env
def create_marlo(env_config: dict):
config = deepcopy(MALMO_DEFAULTS)
config.update(env_config)
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make('MarLo-DefaultFlatWorld-v0',
params=dict(
allowContinuousMovement=["move", "turn"],
client_pool = client_pool,
videoResolution=[84, 84]
))
env = marlo.init(join_tokens[0])
return env
def get_join_tokens():
if marlo.is_grading():
"""
In the crowdAI Evaluation environment obtain the join_tokens
from the evaluator
the `params` parameter passed to the `evaluator_join_token` only allows
the following keys :
"seed",
"tick_length",
"max_retries",
"retry_sleep",
"step_sleep",
"skip_steps",
"videoResolution",
"continuous_to_discrete",
"allowContinuousMovement",
"allowDiscreteMovement",
"allowAbsoluteMovement",
"add_noop_command",
"comp_all_commands"
# TODO: Add this to the official documentation ?
# Help Wanted :D Pull Requests welcome :D
"""
join_tokens = marlo.evaluator_join_token(params={})
else:
"""
When debugging locally,
Please ensure that you have a Minecraft client running on port 10000 and 10001
by doing :
$MALMO_MINECRAFT_ROOT/launchClient.sh -port 10000
$MALMO_MINECRAFT_ROOT/launchClient.sh -port 10001
"""
print("Generating join tokens locally...")
client_pool = [('127.0.0.1', 10000), ('127.0.0.1', 10001)]
join_tokens = marlo.make('MarLo-BuildbattleTrain1-v0',
params={
"client_pool": client_pool,
"agent_names" : [
"MarLo-Agent-0",
"MarLo-Agent-1"
]
})
return join_tokens
def get_join_tokens():
if marlo.is_grading():
"""
In the crowdAI Evaluation environment obtain the join_tokens
from the evaluator
the `params` parameter passed to the `evaluator_join_token` only allows
the following keys :
"seed",
"tick_length",
"max_retries",
"retry_sleep",
"step_sleep",
"skip_steps",
"videoResolution",
"continuous_to_discrete",
"allowContinuousMovement",
"allowDiscreteMovement",
"allowAbsoluteMovement",
"add_noop_command",
"comp_all_commands"
# TODO: Add this to the official documentation ?
# Help Wanted :D Pull Requests welcome :D
"""
join_tokens = marlo.evaluator_join_token(params={})
else:
"""
When debugging locally,
Please ensure that you have a Minecraft client running on port 10000 and 10001
by doing :
$MALMO_MINECRAFT_ROOT/launchClient.sh -port 10000
$MALMO_MINECRAFT_ROOT/launchClient.sh -port 10001
"""
print("Generating join tokens locally...")
client_pool = [('127.0.0.1', 10000), ('127.0.0.1', 10001)]
join_tokens = marlo.make('MarLo-BuildbattleTrain1-v0',
params={
"client_pool":
client_pool,
"agent_names":
["MarLo-Agent-0", "MarLo-Agent-1"]
})
return join_tokens
def init_environment(env_name):
client_pool = [('127.0.0.1', 10000), ('127.0.0.1', 10001)]
join_tokens = marlo.make(env_name,
params={
"client_pool": client_pool
})
# As this is a single agent scenario,there will just be a single token
# assert len(join_tokens) == 1
join_token = join_tokens[0]
# initialize environment
env = marlo.init(join_token)
# env.configure(videoResolution=[84,84])
# I need to figure out where in marlo I can fix the videoresolutions. probably in the params of marlo.make
# env.configure(stack_frames=4)
# there is the frame stacking logic which is in the environment
# I have to either transpose it here
# or let go of the DQN
# I can just transpose
# issue is I can't observe 4 times
# it is turn based so every observation I do is one step I take
# create env and add specific conifigurations to Malmo
# I can probably configure all that in the wrapper
# I have to understand what in the wrapper I will still use in the new environment
# but the logic of getting 4 states in a row will need to go to the wrapper
# only thing wrapper is currently doing is adding the float
# the black and white transformation is also not happening
# hence I will still get and extra dimension 3 channels, instead of 4,
# and since 4 is not fixed anywhere it should work fine
env = wrap_env_marlo(env)
if RANDOM_SEED:
env.seed(RANDOM_SEED)
print("Observation Space: ", env.observation_space)
print("Action Space: ", env.action_space)
return env
def get_join_tokens():
if marlo.is_grading():
"""
In the crowdAI Evaluation environment obtain the join_tokens
from the evaluator
"""
join_tokens = marlo.evaluator_join_token()
else:
"""
When debugging locally,
Please ensure that you have a Minecraft client running on port 10000
by doing :
$MALMO_MINECRAFT_ROOT/launchClient.sh -port 10000
"""
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make('MarLo-FindTheGoal-v0',
params={"client_pool": client_pool})
return join_tokens
def env_creator(env_config):
time.sleep(0.5)
tokens = marlo.make('MarLo-SimpleRoom-v0',
params=dict(
role=getRole(),
client_pool= getClientPool(),
observeGrid= [-1, -1, -1, 1, 0, 1], # (x1, y1, z1, x2, y2, z2, name)
gridSize= 18,
episode_timelimit= 40e3,
seed= 10,
comp_all_commands=['move', 'turn'],
add_noop_command=False,
tick_length=50,
forceWorldReset=False,
# "recordDestination": "data.tgz",
# "recordMP4": [20, 400000],
))
join_token = tokens[0]
return marlo.init(join_token)
def make_env(env_name, env_seed=0, demo=False):
join_tokens = marlo.make(env_name,
params=dict(
comp_all_commands=["move", "turn"],
allowContinuousMovement=True,
videoResolution=[336, 336],
kill_clients_retry=10,
step_sleep=0.01,
kill_clients_after_num_rounds=100,
prioritise_offscreen_rendering=not demo,
))
env = marlo.init(join_tokens[0])
env = Monitor(env)
obs = env.reset()
action = env.action_space.sample()
obs, r, done, info = env.step(action)
env.seed(int(env_seed))
return env
def setupEnv(mission='MarLo-FindTheGoal-v0', videoResolution = [800, 600], port=10000):
# Sets up marlo environment
client_pool = [('127.0.0.1', port)]
# Step sleep at to 0.2 to handle lag between marlo and Malmo
join_tokens = marlo.make(mission, params={
"client_pool": client_pool,
'suppress_info': False,
'videoResolution': videoResolution,
'tick_length': 50,
'step_sleep': 0.2})
# As this is a single agent scenario,
# there will just be a single token
assert len(join_tokens) == 1
join_token = join_tokens[0]
env = marlo.init(join_token)
# Change the spec of the mission by loading xml from file
# This is necessary to use our custom XML files
missionXML= loadMissionFile(mission+'.xml')
env.mission_spec = MalmoPython.MissionSpec(missionXML, True)
return env
def env_creator(env_config):
print(
'---------------------------------------------\n\n\nENV_CREATE FUNCTION CALLED\n\n\n---------------------------------------------'
)
join_tokens = marlo.make(
'MarLo-SimpleRoom-v0',
params=dict(
client_pool=None,
observeGrid=[-1, -1, -1, 1, 0,
1], # (x1, y1, z1, x2, y2, z2, name)
gridSize=18,
episode_timelimit=40e3,
seed=10,
comp_all_commands=['move', 'turn'],
add_noop_command=False,
tick_length=10,
forceWorldReset=False,
))
join_token = join_tokens[0]
return marlo.init(join_token)
import marlo
import time
import json
import base64
client_pool = [('127.0.0.1', 10000), ('127.0.0.1', 10001)]
join_tokens = marlo.make('MarLo-MazeRunner-v0',
params={
"agent_names": ["MarLo-Agent0"],
"videoResolution": [800, 600],
"forceWorldReset": True,
"client_pool": client_pool,
"allowContinuousMovement": ["move", "turn"],
"recordMP4": "video"
})
@marlo.threaded
def run_agent(join_token):
env = marlo.init(join_token)
frame = env.reset()
done = False
while not done:
_action = env.action_space.sample()
obs, reward, done, info = env.step(_action)
time.sleep(0.5)
print("reward:", reward)
print("done:", done)
print("info", info)
if not os.path.exists(out_dir):
os.makedirs(out_dir)
out_dir_logs = out_dir + '/logging'
if not os.path.exists(out_dir_logs):
os.makedirs(out_dir_logs)
if save_dir and not os.path.exists(save_dir):
os.makedirs(save_dir)
experiments.set_log_base_dir(out_dir)
# Ensure that you have a minecraft-client running with : marlo-server --port 10000
# "MarLo-FindTheGoal-v0"
# 'MarLo-CatchTheMob-v0'
join_tokens = marlo.make("MarLo-FindTheGoal-v0",
params=dict(allowContinuousMovement=["move", "turn"],
videoResolution=[84, 84],
kill_clients_after_num_rounds=500))
env = marlo.init(join_tokens[0])
obs = env.reset()
env.render(mode="rgb_array")
print('initial observation:', obs)
action = env.action_space.sample()
obs, r, done, info = env.step(action)
print('next observation:', obs)
print('reward:', r)
print('done:', done)
print('info:', info)
print('actions:', str(env.action_space))
parser.add_argument('--turn_based', action='store_true')
args = parser.parse_args()
turn_based = args.turn_based
number_of_rollouts = args.rollouts
# Register the multi-agent environment.
env_name = 'malmo-multi-agent-v0'
register(
id=env_name,
entry_point='marlo.envs:MinecraftEnv',
# Make sure mission xml is in the marlo/assets directory.
kwargs={'mission_file': args.mission_file})
env = marlo.make(env_name)
resolution = [84, 84] # [800, 600]
config = {
'allowDiscreteMovement': ["move", "turn"],
'videoResolution': resolution,
"turn_based": turn_based
}
join_agents = start_agents(env,
env_name,
None,
config,
number_of_rollouts + 1,
daemon=True)
import marlo
import time
import json
import base64
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make('MarLo-MazeRunner-v0',
params={
"videoResolution": [800, 600],
"client_pool": client_pool,
"agent_names": ["MarLo-Agent0"],
"allowContinuousMovement": ["move", "turn"],
})
def run_agent(join_token):
env = marlo.init(join_token)
frame = env.reset()
done = False
count = 0
while not done:
_action = env.action_space.sample()
# obs, reward, done, info = env.step(_action)
time.sleep(0.5)
# print("reward:", reward)
# print("done:", done)
# print("info", info)
env.close()
for _join_token in join_tokens:
import pickle
from enum import Enum
import matplotlib.pyplot as plt
from lxml import etree
import logging
frame_size = [60, 60]
num_input = frame_size[0] * frame_size[1]
client_pool = [('127.0.0.1', 10000), ('127.0.0.1', 10001)]
join_tokens = marlo.make("MarLo-MobchaseTrain1-v0",
params=dict(
client_pool=client_pool,
agent_names=["MarLo-Agent-0", "MarLo-Agent-1"],
videoResolution=frame_size,
kill_clients_after_num_rounds=500,
forceWorldReset=False,
max_retries=500,
retry_sleep=0.1,
step_sleep=0.1,
prioritise_offscreen_rendering=False,
suppress_info=False))
assert len(join_tokens) == 2
class Convolutional():
def __init__(self, num_classes, learningRate):
self.x = tf.placeholder("float", [None, num_input])
#Reshape the flatten data
self.input_layer = tf.reshape(self.x,
[-1, frame_size[1], frame_size[0], 1])
def train(arglist):
############################################
@marlo.threaded
def funcion(env, action, agent_num):
contador = 0
while True: # Ejecutamos la accion evitando errores
_, r, done, info, new_obs = env.step(np.argmax(action) + 1)
new_obs = new_obs['observation']
if new_obs == None:
new_obs = last_obs[agent_num]
else:
new_obs = [
new_obs.get('XPos'),
new_obs.get('ZPos'),
new_obs.get('Yaw')
]
contador += 1
if r != 0:
break
elif info != None:
if "caught_the_Chicken" in info:
r += 1
print("SE HA HARCODEADO LA PUNTUACION ", done, " ", info)
break
if "Agent0_defaulted" in info:
r = -0.02
break
if "Agent1_defaulted" in info:
r = -0.02
break
elif contador >= 100:
print("SE HA TARDADO MUCHO EN REALIZAR LA ACCION")
break
return new_obs, r, done, info
#######################################################
with U.single_threaded_session():
# Create environment
client_pool = [('127.0.0.1', 10000), ('127.0.0.1', 10001)]
join_tokens = marlo.make(
"MarLo-MobchaseTrain1-v0",
params=dict(client_pool=client_pool,
agent_names=["MarLo-Agent-0", "MarLo-Agent-1"],
videoResolution=[64, 64],
kill_clients_after_num_rounds=500,
forceWorldReset=False,
max_retries=500,
retry_sleep=0.1,
step_sleep=0.1,
prioritise_offscreen_rendering=False,
suppress_info=False))
assert len(join_tokens) == 2
# Create agent trainers
#obs_shape_n = [(64,64,3,),(64,64,3,)]
observation_space = [
gym.spaces.Box(low=-np.inf,
high=+np.inf,
shape=(6, ),
dtype=np.float32),
gym.spaces.Box(low=-np.inf,
high=+np.inf,
shape=(6, ),
dtype=np.float32)
]
obs_shape_n = [observation_space[i].shape for i in range(2)]
action_space = [gym.spaces.Discrete(4), gym.spaces.Discrete(4)]
num_adversaries = 0
trainers = get_trainers(num_adversaries, obs_shape_n, action_space,
arglist)
# Initialize
U.initialize()
epis_trans = 0
epsilon = 0.0
# Load previous results, if necessary
if arglist.load_dir == "":
arglist.load_dir = arglist.save_dir
if arglist.restore:
print('Loading previous state...')
resbuf = pickle.load(open("./saves/losbuffers.p", "rb"))
epis_trans = resbuf[2]
epsilon = resbuf[3]
U.load_state(arglist.load_dir + str(epis_trans))
trainers[0].replay_buffer = resbuf[0]
trainers[1].replay_buffer = resbuf[1]
episode_rewards = []
agent_rewards = [
[] for _ in range(2)
] # lista de sumas de las recompensas de cada episodio
final_ep_rewards = [] # sum of rewards for training curve
final_ep_ag_rewards = [] # agent rewards for training curve
saver = tf.train.Saver()
t_start = time.time()
#inicial0 = [1.5, 2.5, 270, 5.5, 6.5, 180]
#inicial1 = [5.5, 6.5, 180, 1.5, 2.5, 270]
inicial0 = [1.5, 2.5, 270, 3.5, 4.5, 180]
inicial1 = [3.5, 4.5, 180, 1.5, 2.5, 270]
while True:
#NEW
last_obs = []
agent_rewards[0].append(0)
agent_rewards[1].append(0)
env0 = marlo.init(join_tokens[0])
env1 = marlo.init(join_tokens[1])
# Run agent-0
agent_thread_0, res0 = reiniciar(env0)
# Run agent-1
agent_thread_1, res1 = reiniciar(env1)
obs0 = res0.get()
obs1 = res1.get()
obs0 = inicial0
obs1 = inicial1
done0 = False
done1 = False
num_eps = 0
#Ejecutar 10 episodios
while True:
if (random() > epsilon):
action0 = trainers[0].action(np.array(
obs0)) # se obtine la accion que ejecuta la politica
else:
action0 = np.random.dirichlet(np.ones(4), size=1)[0]
if (random() > epsilon):
action1 = trainers[1].action(np.array(
obs0)) # se obtine la accion que ejecuta la politica
else:
action1 = np.random.dirichlet(np.ones(4), size=1)[0]
#print("Estan dentro")
# Run agent-0
agent_thread_0, resul0 = funcion(env0, action0, 0)
# Run agent-1
agent_thread_1, resul1 = funcion(env1, action1, 1)
# Wait for both the threads to complete execution
agent_thread_0.join()
#print("Esta fuera 1")
agent_thread_1.join()
#print("Estan fuera")
nob0, r0, done0, i0 = resul0.get()
nob1, r1, done1, i1 = resul1.get()
last_obs = [copy.deepcopy(nob0), copy.deepcopy(nob1)]
# Las nuevas observciones
varhelp = copy.deepcopy(nob0)
nob0.extend(nob1)
nob1.extend(varhelp)
#print("ESTAS SON LAS OBSERVACIONES")
#print(nob0)
#print(nob1)
trainers[0].experience(np.array(obs0), action0, r0,
np.array(nob0), done0, False)
trainers[1].experience(np.array(obs1), action1, r1,
np.array(nob1), done1, False)
agent_rewards[0][-1] += r0
agent_rewards[1][-1] += r1
obs0 = nob0
obs1 = nob1
if done0 or done1:
print("EPISODIO NUMERO:", num_eps)
# Run agent-0
agent_thread_0, res0 = reiniciar(env0)
# Run agent-1
agent_thread_1, res1 = reiniciar(env1)
obs0 = res0.get()
obs1 = res1.get()
obs0 = inicial0
obs1 = inicial1
done0 = False
done1 = False
num_eps += 1
loss = None
for agent in trainers:
agent.preupdate()
for agent in trainers:
loss = agent.update(trainers)
print("LA LOSS", loss)
if num_eps % epi_per_iter == 0:
break
agent_rewards[0].append(0)
agent_rewards[1].append(0)
#Fin de ejecutar 10 episodios
print("FIN DEL SAMPLE")
# Se obtiene una lista de tuplas que contienen las rewards de los agentes emparejadas por episodios utilizadno los ultimos episodios generados en la iteracion
# A estas tuplas se transforman a listas y se aplica sum()
# El resultado de esto se coloca al final de episode_rewards
#
# En resumen: se suman las ultimas rewards de los agentes por episodios y se añaden a la lista
episode_rewards.extend(
list(
map(
sumtuple,
list(
zip(agent_rewards[0][epis_trans:],
agent_rewards[1][epis_trans:])))))
epis_trans += 10
if epsilon > 0.1:
epsilon -= 0.002
print("TOTAL DE EPISODIOS TRANSCURRIDOS: ", epis_trans,
" Epsilon: ", epsilon)
# update all trainers, if not in display or benchmark mode
# save model, display training output
if (epis_trans % arglist.save_rate == 0):
U.save_state(arglist.save_dir + str(epis_trans), saver=saver)
losbuffers = [
trainers[0].replay_buffer, trainers[1].replay_buffer,
epis_trans, epsilon
]
pickle.dump(
losbuffers,
open("./saves/losbuffers" + str(epis_trans) + ".p", "wb"))
pickle.dump(losbuffers, open("./saves/losbuffers.p", "wb"))
if (epis_trans % 1000 == 0):
break
#!/usr/bin/env python
# Please ensure that you have two Minecraft clients running on port 10000 and
# port 10001 by doing :
# $MALMO_MINECRAFT_ROOT/launchClient.sh -port 10000
# $MALMO_MINECRAFT_ROOT/launchClient.sh -port 10001
import marlo
client_pool = [('127.0.0.1', 10000), ('127.0.0.1', 10001)]
join_tokens = marlo.make('MarLo-MazeRunner-v0',
params={
"client_pool": client_pool,
"agent_names": ["MarLo-Agent-0", "MarLo-Agent-1"]
})
# As this is a two-agent scenario,
# there will just two join tokens
assert len(join_tokens) == 2
@marlo.threaded
def run_agent(join_token):
env = marlo.init(join_token)
observation = env.reset()
done = False
count = 0
while not done:
_action = env.action_space.sample()
obs, reward, done, info = env.step(_action)
print("reward:", reward)
print("done:", done)
print("info", info)
env.close()
# !/usr/bin/env python
# $MALMO_MINECRAFT_ROOT/launchClient.sh -port 10000
import marlo
import time
import random
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make('MarLo-MineField-v0',
params=dict(
client_pool= client_pool,
observeGrid= [-1, 0, -1, 1, 0, 1], # (x1, y1, z1, x2, y2, z2, name)
gridSize= 9,
seed=1, #seed=99 !!!
episode_timelimit=60e3,
comp_all_commands=['move', 'turn'],
add_noop_command=False,
videoResolution=None,
))
# As this is a single agent scenario,
# there will just be a single tokenw
assert len(join_tokens) == 1
join_token = join_tokens[0]
env = marlo.init(join_token)
observation = env.reset()
done = False
total_reward = 0
i = 0
n_hidden_channels = args.n_hidden_channels
n_hidden_layers = args.n_hidden_layers
if DEBUG_ON:
print("n_hidden_channels " + str(n_hidden_channels) + " n_hidden_layers " +
str(n_hidden_layers))
# GAME SELECTION AND CONNECTION TO THE CLIENT
# Ensure that you have a minecraft-client running with : marlo-server --port 10000
client_pool = [('127.0.0.1', 10020)]
if DEBUG_ON:
print("Game:", GAME)
join_tokens = marlo.make(GAME,
params=dict(videoResolution=[VIDEO_RES, VIDEO_RES],
kill_clients_after_num_rounds=500))
env = marlo.init(join_tokens[0])
# ------------------------------------------
obs = env.reset()
env.render(mode="rgb_array")
if DEBUG_ON:
print('initial observation:', obs)
action = env.action_space.sample()
obs, r, done, info = env.step(action)
if DEBUG_ON:
print('next observation:', obs)
#!/usr/bin/env python
# Please ensure that you have a Minecraft client running on port 10000
# by doing :
# $MALMO_MINECRAFT_ROOT/launchClient.sh -port 10000
import marlo
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make('MarLo-MazeRunner-v0',
params={"client_pool": client_pool})
# As this is a single agent scenario,
# there will just be a single token
assert len(join_tokens) == 1
join_token = join_tokens[0]
env = marlo.init(join_token)
observation = env.reset()
done = False
while not done:
_action = env.action_space.sample()
obs, reward, done, info = env.step(_action)
print("reward:", reward)
print("done:", done)
print("info", info)
env.close()
import marlo
# import os
#
#
client_pool = [('127.0.0.1', 10000)]
#
# join_tokens = marlo.make('MarLo-FindTheGoal-v0',params={"client_pool": client_pool})
join_tokens = marlo.make('MarLo-CatchTheMob-v0',
params={"client_pool": client_pool})
assert len(join_tokens) == 1
join_token = join_tokens[0]
env = marlo.init(join_token)
observation = env.reset()
done = False
while not done:
_action = env.action_space.sample()
obs, reward, done, info = env.step(_action)
print("reward:", reward)
print("done:", done)
print("info", info)
env.close()
#!/usr/bin/env python
# Please ensure that you have a Minecraft client running on port 10000
# by doing :
# $MALMO_MINECRAFT_ROOT/launchClient.sh -port 10000
import marlo
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make('MarLo-FindTheGoal-v0',
params={"client_pool": client_pool})
# As this is a single agent scenario,
# there will just be a single token
assert len(join_tokens) == 1
join_token = join_tokens[0]
env = marlo.init(join_token)
observation = env.reset()
done = False
while not done:
_action = env.action_space.sample()
obs, reward, done, info = env.step(_action)
print("reward:", reward)
print("done:", done)
print("info", info)
env.close()
import random
import ray
from ray import tune
from ray.tune.registry import register_env
from ray.rllib.agents.pg import PGTrainer
#client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make(
'MarLo-SimpleRoom-v0',
params=dict(
client_pool=None,
observeGrid=[-1, -1, -1, 1, 0, 1], # (x1, y1, z1, x2, y2, z2, name)
gridSize=18,
episode_timelimit=40e3,
seed=10,
comp_all_commands=['move', 'turn'],
add_noop_command=False,
tick_length=10, #in millisec
forceWorldReset=False,
# "recordDestination": "data.tgz",
# "recordMP4": [20, 400000],
))
assert len(join_tokens) == 1
join_token = join_tokens[0]
def env_creator(env_config):
return marlo.init(join_token)
target_update_interval = 10 ** 2
update_interval = 1
target_update_method = 'hard'
soft_update_tau = 1e-2
rbuf_capacity = 5 * 10 ** 5
steps = 10 ** 5
eval_n_runs = 100
eval_interval = 10 ** 4
def phi(obs):
return obs.astype(np.float32)
# Ensure that you have a minecraft-client running with : marlo-server --port 10000
join_tokens = marlo.make('MinecraftCliffWalking1-v0',
params=dict(
allowContinuousMovement=["move", "turn"],
videoResolution=[800, 600]
))
env = marlo.init(join_tokens[0])
obs = env.reset()
env.render(mode="rgb_array")
print('initial observation:', obs)
action = env.action_space.sample()
obs, r, done, info = env.step(action)
print('next observation:', obs)
print('reward:', r)
print('done:', done)
print('info:', info)
# import malmo.minecraftbootstrap; malmo.minecraftbootstrap.set_malmo_xsd_path()
#!/usr/bin/env python
# Please ensure that you have a Minecraft client running on port 10000
# by doing :
# $MALMO_MINECRAFT_ROOT/launchClient.sh -port 10000
import marlo
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make('MarLo-BuildbattleTrainX-v0',
params={
"client_pool": client_pool
})
# As this is a single agent scenario,
# there will just be a single token
assert len(join_tokens) == 1
join_token = join_tokens[0]
env = marlo.init(join_token)
observation = env.reset()
done = False
while not done:
_action = env.action_space.sample()
obs, reward, done, info = env.step(_action)
print("reward:", reward)
print("done:", done)
print("info", info)
env.close()
from collections import deque
import marlo
import keras
from keras.models import Sequential
from keras.layers import Dense, Dropout, Flatten, Activation
from keras.optimizers import RMSprop, Adam
import numpy as np
from enum import Enum
import QLearn
from QAgent import QAgent
print("testo")
episodes = 1000
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make('MarLo-CliffWalking-v0',
params={
"client_pool": client_pool
})
# As this is a single agent scenario,
# there will just be a single token
assert len(join_tokens) == 1
join_token = join_tokens[0]
env = marlo.init(join_token)
if __name__ == "__main__":
env = marlo.init(join_token)
trials = 1000
trial_len = 200
updateTargetNework = 100
DQN_agent = QLearn.DQN(env=env)
steps = []
print('nombre d\'essais: ', trials)
import marlo
from Agent import Agent
from GameState import GameState
import numpy as np
from bs4 import BeautifulSoup
mission_name = 'MarLo-FindTheGoal-v0'
from marlo.base_env_builder import dotdict
from MonteCarlo import MonteCarlo
from Q import Q
# config
client_pool = [('127.0.0.1', 10000)]
join_tokens = marlo.make(mission_name,
params={
"client_pool": client_pool,
"suppress_info": False,
"skip_steps": 2,
"step_sleep": 0.5
})
# As this is a single agent scenario,
# there will just be a single token
assert len(join_tokens) == 1
join_token = join_tokens[0]
# define the environment variable and reset observation
env = marlo.init(join_token)
env.default_base_params.allowDiscreteMovement = True
env.default_base_params.allowContinuousMovement = False
env.default_base_params.allowAbsoluteMovement = False