def run_bbox(verbose=False):
n_features = n_actions = max_time = -1
if bbox.is_level_loaded():
bbox.reset_level()
else:
bbox.load_level("../levels/train_level.data", verbose=1)
n_features = bbox.get_num_of_features()
n_actions = bbox.get_num_of_actions()
max_time = bbox.get_max_time()
av_table = ActionValueTable(n_features, n_actions)
av_table.initialize(0.2)
print av_table._params
learner = Q(0.5, 0.1)
learner._setExplorer(EpsilonGreedyExplorer(0.4))
agent = LearningAgent(av_table, learner)
environment = GameEnvironment()
task = GameTask(environment)
experiment = Experiment(task, agent)
while environment.finish_flag:
experiment.doInteractions(1)
agent.learn()
bbox.finish(verbose=1)
def prepare_bbox():
global n_features, n_actions
if bbox.is_level_loaded():
bbox.reset_level()
else:
bbox.load_level("../levels/test_level.data", verbose=1)
n_features = bbox.get_num_of_features()
n_actions = bbox.get_num_of_actions()
def prepare_bbox():
global n_features, n_actions
if bbox.is_level_loaded():
bbox.reset_level()
else:
bbox.load_level("../levels/test_level.data", verbose=1)
n_features = bbox.get_num_of_features()
n_actions = bbox.get_num_of_actions()
def prepare_bbox():
global n_f, n_a, max_time
if bbox.is_level_loaded():
bbox.reset_level()
else:
bbox.load_level("../levels/train_level.data", verbose=1)
n_f = bbox.get_num_of_features()
n_a = bbox.get_num_of_actions()
max_time = bbox.get_max_time()
def prepare_bbox():
global n_features, n_actions, max_time
# Reset environment to the initial state, just in case
if bbox.is_level_loaded():
bbox.reset_level()
else:
# Load the game level
bbox.load_level("../levels/train_level.data", verbose=1)
n_features = bbox.get_num_of_features()
n_actions = bbox.get_num_of_actions()
max_time = bbox.get_max_time()
def prepare_box():
global n_features, n_actions, max_time
# Reset the environment to the initial state, just in case
if bbox.is_level_loaded():
bbox.reset_level()
else:
# Load the game level
bbox.load_level('levels/train_level.data', verbose=1)
n_features = bbox.get_num_of_features()
n_actions = bbox.get_num_of_actions()
max_time = bbox.get_max_time()
def reset(self):
if bbox.is_level_loaded():
bbox.reset_level()
else:
bbox.load_level(self.level, verbose=1)
self.n_features = bbox.get_num_of_features()
self.n_actions = bbox.get_num_of_actions()
self.max_time = bbox.get_max_time()
self._steps = 0
self._state = np.zeros((1, self.n_features))
self._is_over = False
self._prev_score = -float('inf')
self._actions_log = []
def reset(self):
if bbox.is_level_loaded():
bbox.reset_level()
else:
bbox.load_level(self.level, verbose=1)
self.n_features = bbox.get_num_of_features()
self.n_actions = bbox.get_num_of_actions()
self.max_time = bbox.get_max_time()
self._steps = 0
self._state = np.zeros((1, self.n_features))
self._is_over = False
self._prev_score = -float('inf')
self._actions_log = []
def load_level(level, verbosity):
"""
Loads the named level and returns its name and description.
"""
path = join(dirname(__file__), 'levels', level)
if not path.endswith('_level.data'):
path = '{}_level.data'.format(path)
if verbosity > 3:
print("Loading level from {}".format(path))
bb_load_level(path, verbose=verbosity > 3)
return {'key': level,
'steps': get_max_time(),
'actions': get_num_of_actions(),
'features': get_num_of_features()}
def prepare_bbox():
global n_features, n_actions, max_time, vectors, pool, num_of_vectors
if bbox.is_level_loaded():
bbox.reset_level()
else:
bbox.load_level("../levels/train_level.data", verbose=1)
n_features = bbox.get_num_of_features()
n_actions = bbox.get_num_of_actions()
max_time = bbox.get_max_time()
vectors = np.zeros((num_of_vectors, n_features), np.float32)
print("preparing")
pool = multiprocessing.Pool(processes=processes)
def load_level(level, verbosity):
"""
Loads the named level and returns its name and description.
"""
path = join(dirname(__file__), 'levels', level)
if not path.endswith('_level.data'):
path = '{}_level.data'.format(path)
if verbosity > 3:
print("Loading level from {}".format(path))
bb_load_level(path, verbose=verbosity > 3)
return {
'key': level,
'steps': get_max_time(),
'actions': get_num_of_actions(),
'features': get_num_of_features()
}
def prepare_bbox():
global n_features, n_actions, max_time
## TODO: Save the interactions with the environment as an output data frame
global interaction_list
interaction_list = []
## Reset the environment to initial state, just in case
if bbox.is_level_loaded():
bbox.reset_level()
else:
## Load the game level
bbox.load_level("../levels/train_level.data", verbose=True)
n_features = bbox.get_num_of_features()
n_actions = bbox.get_num_of_actions()
max_time = bbox.get_max_time()
## The matrix that contains the output data frame
states = ['state_'] * n_features
state_list = [states[i] + str(i) for i in range(n_features)]
header_list = state_list + ['reward', 'action']
interaction_list.append(header_list)
def prepare_bbox():
'''
Prepares the environment (learning/test data).
'''
global n_features
global n_actions
global max_time
global q_function
global epsilon
global gamma
global alpha
global valid_actions
global init_value
if bbox.is_level_loaded():
## Reset the environment to initial state
bbox.reset_level()
else:
## Load the training/test data
bbox.load_level('../levels/train_level.data', verbose=True)
n_features = bbox.get_num_of_features()
n_actions = bbox.get_num_of_actions()
max_time = bbox.get_max_time()
#!/usr/bin/env python3
"""
A minimal bot player.
Loads the level and params and lets the bot act.
"""
from interface import (get_max_time, get_num_of_actions, get_num_of_features,
finish, load_level)
from numpy import get_include, load
from pyximport import install
install(setup_args={'include_dirs': get_include()}, reload_support=True)
from bot_wrapper import do_act
if __name__ == '__main__':
load_level('../levels/train_level.data', verbose=1)
level = {
'steps': get_max_time(),
'actions': get_num_of_actions(),
'features': get_num_of_features()
}
params = dict(load('params.npz'))
do_act(level, params)
finish(verbose=1)
#!/usr/bin/env python3
"""
A minimal bot player.
Loads the level and params and lets the bot act.
"""
from interface import (get_max_time, get_num_of_actions, get_num_of_features,
finish, load_level)
from numpy import get_include, load
from pyximport import install
install(setup_args={'include_dirs': get_include()}, reload_support=True)
from bot_wrapper import do_act
if __name__ == '__main__':
load_level('../levels/train_level.data', verbose=1)
level = {
'steps': get_max_time(),
'actions': get_num_of_actions(),
'features': get_num_of_features()
}
params = dict(load('params.npz'))
do_act(level, params)
finish(verbose=1)