def probability_experiment():
m = model.train_model_cg()
mdp = grid_world.MDP()
score_d, score_r = test_learned_model(m, mdp)
print 'unnormalized data probability: ', score_d
print 'unnormalized random probability: ', score_r
def probability_experiment():
m = model.train_model_cg()
mdp = grid_world.MDP()
score_d, score_r = test_learned_model(m, mdp)
print "unnormalized data probability: ", score_d
print "unnormalized random probability: ", score_r
def main(model_size=(18, 2, 18), dist='uniform', n_iters=1):
file_str = 'model.%i.%i.%s.pickle.gz' % (model_size[0], model_size[1],
dist)
try:
with util.openz(file_str) as f:
print 'found previous file, using: ', f
m = pickle.load(f)
except IOError:
print 'no serialized model found, training a new one'
m = model.train_model_cg(model_size, dist=dist)
with util.openz(file_str, 'wb') as f:
pickle.dump(m, f)
mdp = grid_world.MDP()
map_policy = MapPolicy(m, mdp.env)
mdp.policy = map_policy
#optimal_policy = grid_world.OptimalPolicy(mdp.env)
for i in xrange(n_iters):
state = None
while not mdp.env.is_goal_state():
state = mdp.env.state
print 'current state: ', state
#actions = mdp.env.get_actions(state)
#act = map_policy.choose_action()
# convert state to binary vector
phi_s = np.zeros(model_size[0])
phi_s[state[0]] = 1
phi_s[mdp.env.n_rows + state[1]] = 1
map_config = get_map(m, phi_s)
pos, act, pos_p, r = parse_config(map_config, m, mdp.env)
print 'map mu vector: ', map_config
print 'map configuration: ', pos, act, pos_p, r
mdp.env.take_action(act)
print 'final state: ', mdp.env.state
def main(model_size = (18,2,18), dist = 'uniform', n_iters = 1):
file_str = 'model.%i.%i.%s.pickle.gz'%(model_size[0],model_size[1],dist)
try:
with util.openz(file_str) as f:
print 'found previous file, using: ', f
m = pickle.load(f)
except IOError:
print 'no serialized model found, training a new one'
m = model.train_model_cg(model_size, dist = dist)
with util.openz(file_str, 'wb') as f:
pickle.dump(m, f)
mdp = grid_world.MDP()
map_policy = MapPolicy(m, mdp.env)
mdp.policy = map_policy
#optimal_policy = grid_world.OptimalPolicy(mdp.env)
for i in xrange(n_iters):
state = None
while not mdp.env.is_goal_state():
state = mdp.env.state
print 'current state: ', state
#actions = mdp.env.get_actions(state)
#act = map_policy.choose_action()
# convert state to binary vector
phi_s = np.zeros(model_size[0])
phi_s[state[0]] = 1
phi_s[mdp.env.n_rows + state[1]] = 1
map_config = get_map(m, phi_s)
pos, act, pos_p, r = parse_config(map_config, m, mdp.env)
print 'map mu vector: ', map_config
print 'map configuration: ', pos, act, pos_p, r
mdp.env.take_action(act)
print 'final state: ', mdp.env.state
