def create_hmm(qsr_seq, filename):
r = ROSClient()
q, d = r.call_service(
HMMRepRequestCreate(
qsr_seq=qsr_seq,
qsr_type="qtch"
)
)
with open(filename+".hmm", 'w') as f: f.write(d)
class TestHMM(unittest.TestCase):
QTCB_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcb_sample_test.hmm')[0]
QTCC_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcc_sample_test.hmm')[0]
QTCBC_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcbc_sample_test.hmm')[0]
QTCB_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcb_passby_left.hmm')[0]
QTCC_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcc_passby_left.hmm')[0]
QTCBC_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcbc_passby_left.hmm')[0]
RCC3_TEST_HMM = find_resource(PKG, 'rcc3_test.hmm')[0]
QTCB_QSR = find_resource(PKG, 'qtcb.qsr')[0]
QTCC_QSR = find_resource(PKG, 'qtcc.qsr')[0]
QTCBC_QSR = find_resource(PKG, 'qtcbc.qsr')[0]
RCC3_QSR = find_resource(PKG, 'rcc3.qsr')[0]
correct_samples = {
"qtcb": [[u'--', u'0-', u'+-', u'+0', u'++']],
"qtcc": [[u'--+-', u'--0-', u'----', u'0---', u'+---', u'+0--', u'++--', u'---']],
"qtcbc": [[u'--', u'----', u'0---', u'+---', u'+0--', u'++--', u'++']]
}
correct_hashsum = {
"qtcb": "3fb65b50d0f7631a300132e8bca9ca13",
"qtcc": "dbf1529cb0b0c90aaebbe7eafe0e9b05",
"qtcbc": "0a3acf7b48c4c1155931442c86317ce4",
"rcc3": "6d5bcc6c44d9b1120c738efa1994a40a"
}
correct_loglikelihoods ={
"qtcb": -2.16887,
"qtcc": -6.07188,
"qtcbc": -2.23475,
"rcc3": -4.15914
}
def __init__(self, *args):
super(TestHMM, self).__init__(*args)
rospy.init_node(NAME)
self.r = ROSClient()
def _create_hmm(self, qsr_file, qsr_type):
with open(qsr_file, 'r') as f: qsr_seq = json.load(f)
_, d = self.r.call_service(
HMMRepRequestCreate(
qsr_seq=qsr_seq,
qsr_type=qsr_type
)
)
return d
def _create_sample(self, hmm_file, qsr_type):
with open(hmm_file, 'r') as f: hmm = f.read()
_, s = self.r.call_service(
HMMRepRequestSample(
qsr_type=qsr_type,
xml=hmm,
max_length=10,
num_samples=1
)
)
return s
def _calculate_loglikelihood(self, hmm_file, qsr_file, qsr_type):
with open(qsr_file, 'r') as f: qsr_seq = json.load(f)
with open(hmm_file, 'r') as f: hmm = f.read()
q, l = self.r.call_service(
HMMRepRequestLogLikelihood(
qsr_type=qsr_type,
xml=hmm,
qsr_seq=qsr_seq
)
)
return round(l, 5)
def _to_strings(self, array):
return [x.values()[0] for x in array]
def test_qtcb_create(self):
res = self._create_hmm(self.QTCB_QSR, 'qtcb')
self.assertEqual(hashlib.md5(res).hexdigest(), self.correct_hashsum["qtcb"])
def test_qtcb_sample(self):
res = self._create_sample(self.QTCB_SAMPLE_TEST_HMM, 'qtcb')
self.assertEqual(res, self.correct_samples["qtcb"])
def test_qtcb_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCB_PASSBY_LEFT_HMM, self.QTCB_QSR, 'qtcb')
self.assertEqual(res, self.correct_loglikelihoods["qtcb"])
def test_qtcc_create(self):
res = self._create_hmm(self.QTCC_QSR, 'qtcc')
self.assertEqual(hashlib.md5(res).hexdigest(), self.correct_hashsum["qtcc"])
def test_qtcc_sample(self):
res = self._create_sample(self.QTCC_SAMPLE_TEST_HMM, 'qtcc')
self.assertEqual(res, self.correct_samples["qtcc"])
def test_qtcc_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCC_PASSBY_LEFT_HMM, self.QTCC_QSR, 'qtcc')
self.assertEqual(res, self.correct_loglikelihoods["qtcc"])
def test_qtcbc_create(self):
res = self._create_hmm(self.QTCBC_QSR, 'qtcbc')
self.assertEqual(hashlib.md5(res).hexdigest(), self.correct_hashsum["qtcbc"])
def test_qtcbc_sample(self):
res = self._create_sample(self.QTCBC_SAMPLE_TEST_HMM, 'qtcbc')
self.assertEqual(res, self.correct_samples["qtcbc"])
def test_qtcbc_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCBC_PASSBY_LEFT_HMM, self.QTCBC_QSR, 'qtcbc')
self.assertEqual(res, self.correct_loglikelihoods["qtcbc"])
def test_rcc3_create(self):
res = self._create_hmm(self.RCC3_QSR, 'rcc3')
self.assertEqual(hashlib.md5(res).hexdigest(), self.correct_hashsum["rcc3"])
def test_rcc3_sample(self):
res = self._create_sample(self.RCC3_TEST_HMM, 'rcc3')
self.assertTrue(type(res) == list and len(res) > 0)
def test_rcc3_loglikelihood(self):
res = self._calculate_loglikelihood(self.RCC3_TEST_HMM, self.RCC3_QSR, 'rcc3')
self.assertEqual(res, self.correct_loglikelihoods["rcc3"])
subparsers.add_parser('loglikelihood', parents=[general, log_parse, qtc_parse])
# Parse arguments
args = parser.parse_args()
rospy.init_node("ros_client")
r = ROSClient()
if args.action == "create":
qsr_seq = load_files(args.input)
d = r.call_service(
HMMRepRequestCreate(
qsr_seq=qsr_seq,
qsr_type=args.qsr_type,
pseudo_transitions=args.pseudo_transitions,
lookup_table=load_json_file(args.lookup) if args.lookup != "" else None,
transition_matrix=load_json_file(args.trans) if args.trans != "" else None,
emission_matrix=load_json_file(args.emi) if args.emi != "" else None,
start_at_zero=args.start_at_zero
)
)
with open(args.output, 'w') as f: json.dump(d, f)
elif args.action == "sample":
with open(args.input, 'r') as f: hmm = json.load(f)
s = r.call_service(
HMMRepRequestSample(
qsr_type=args.qsr_type,
dictionary=hmm,
max_length=args.max_length,
num_samples=args.num_samples,
class StatePredictor(object):
__interaction_types = ["passby", "pathcrossing"]
__filters = {}
__classification_results = {}
model = None
rules = {}
def __init__(self, name):
self.isDebug = False
rospy.loginfo("Starting %s ..." % name)
self.client = ROSClient()
qmc = QTCModelCreation()
obs = qmc.create_observation_model(qtc_type=qmc.qtc_types.qtch, start_end=True)
self.lookup = qmc.create_states(qtc_type=qmc.qtc_types.qtch, start_end=True)
hmc = HMMModelCreation()
m = PfModel()
for f in os.listdir(rospy.get_param("~model_dir")):
filename = rospy.get_param("~model_dir") + '/' + f
if f.endswith(".hmm"):
rospy.loginfo("Creating prediction model from: %s", filename)
pred = hmc.create_prediction_model(input=filename)
m.add_model(f.split('.')[0], pred, obs)
elif f.endswith(".rules"):
with open(filename) as fd:
rospy.loginfo("Reading rules from: %s", filename)
# Necessary due to old rule format:
self.rules[f.split('.')[0]] = self._create_proper_qtc_keys(json.load(fd))
self.model = m.get()
visualisation = rospy.get_param("~visualisation")
self.visualisation_colors = visualisation['models']
self.default_color = ColorRGBA(
a=1.0,
r=visualisation['default']['r']/255.,
g=visualisation['default']['g']/255.,
b=visualisation['default']['b']/255.
)
self.pub = rospy.Publisher("~prediction_array", QTCPredictionArray, queue_size=10)
self.markpub = rospy.Publisher("~marker_array", MarkerArray, queue_size=10)
rospy.Subscriber(rospy.get_param("~qtc_topic", "/online_qtc_creator/qtc_array"), QTCArray, self.callback, queue_size=1)
rospy.Subscriber(rospy.get_param("~ppl_topic", "/people_tracker/positions"), PeopleTracker, self.people_callback, queue_size=1)
rospy.loginfo("... all done")
def _create_proper_qtc_keys(self, dictionary):
ret = {}
for k,v in dictionary.items():
if isinstance(v,dict):
v = self._create_proper_qtc_keys(v)
ret[','.join(k.replace('-1','-').replace('1','+').replace('9','').replace(',',''))] = v
return ret
def callback(self, msg):
out = QTCPredictionArray()
out.header = msg.header
for q in msg.qtc:
m = QTCPrediction()
m.uuid = q.uuid
if q.uuid not in self.__filters:
self.client.call_service(
PfRepRequestCreate(
num_particles=1000,
models=self.model,
state_lookup_table=self.lookup,
uuid=q.uuid,
ensure_particle_per_state=True,
debug=self.isDebug,
starvation_factor=0.1
)
)
self.__filters[q.uuid] = msg.header.stamp.to_sec()
qtc_robot = json.loads(q.qtc_robot_human)[-1].split(',')
qtc_goal = json.loads(q.qtc_goal_human)[-1].split(',')
qtc = [qtc_goal[1], qtc_goal[3], qtc_robot[1]]
if len(qtc_robot) == 4:
qtc.append(qtc_robot[3])
qtc = ','.join(qtc)
# start = time.time()
qtc_state = self.client.call_service(
PfRepRequestUpdate(
uuid=q.uuid,
observation=qtc,
debug=self.isDebug
)
)
# print "+++ elapsed", time.time()-start
if qtc_state == None:
rospy.logwarn("[" + rospy.get_name() + "]: " + str(qtc_state) + " state reported, aborting" )
return
self.__classification_results[q.uuid] = qtc_state[2]
try:
states = self.rules[qtc_state[2]][qtc_state[0]].keys()
probs = self.rules[qtc_state[2]][qtc_state[0]].values() # Both lists are always in a corresponding order
except KeyError as e:
rospy.logwarn("%s not in rules" % e)
return
pred = qtc_state[0].split(',')
print pred
prediction = states[probs.index(max(probs))].split(',')
print prediction
prediction = [prediction[0], pred[2]] \
if len(pred) < 4 else [prediction[0], pred[2], prediction[1], pred[3]]
prediction = ','.join(prediction)
rospy.logdebug("Prediction: %s" %prediction)
if prediction == None:
return
m.qtc_serialised = json.dumps(prediction)
out.qtc.append(m)
self.pub.publish(out)
self.__decay(self.__filters, decay_time=10.)
def _create_qtc_string(self, qtc):
qtc = np.array(qtc)
qtc = qtc[qtc!=9.]
return ','.join(map(str, qtc.astype(int))).replace('-1','-').replace('1','+')
def __decay(self, filters, decay_time=60.):
for k in filters.keys():
if filters[k] + decay_time < rospy.Time.now().to_sec():
rospy.logdebug("[" + rospy.get_name() + "]: " + "Deleting particle filter: " + str(k) + " last seen " + str(filters[k]) )
self.client.call_service(PfRepRequestRemove(uuid=k))
del filters[k]
def people_callback(self, msg):
people = [[],[]]
for uuid, pose in zip(msg.uuids, msg.poses):
people[0].append(pose)
try:
people[1].append(
ColorRGBA(
a=1.0,
r=self.visualisation_colors[self.__classification_results[uuid]]['color']['r']/255.,
g=self.visualisation_colors[self.__classification_results[uuid]]['color']['g']/255.,
b=self.visualisation_colors[self.__classification_results[uuid]]['color']['b']/255.
)
)
except KeyError:
people[1].append(self.default_color)
self.markpub.publish(
mc.marker_array_from_people_tracker_msg(
poses=people[0],
target_frame=msg.header.frame_id,
color=people[1]
)
)
class TestHMM(unittest.TestCase):
QTCB_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcb_sample_test.hmm')[0]
QTCC_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcc_sample_test.hmm')[0]
QTCBC_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcbc_sample_test.hmm')[0]
QTCB_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcb_passby_left.hmm')[0]
QTCC_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcc_passby_left.hmm')[0]
QTCBC_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcbc_passby_left.hmm')[0]
RCC3_TEST_HMM = find_resource(PKG, 'rcc3_test.hmm')[0]
QTCB_QSR = find_resource(PKG, 'qtcb.qsr')[0]
QTCC_QSR = find_resource(PKG, 'qtcc.qsr')[0]
QTCBC_QSR = find_resource(PKG, 'qtcbc.qsr')[0]
RCC3_QSR = find_resource(PKG, 'rcc3.qsr')[0]
correct_samples = {
"qtcb": [[u'--', u'0-', u'+-', u'+0', u'++']],
"qtcc": [[
u'--+-', u'--0-', u'----', u'0---', u'+---', u'+0--', u'++--',
u'---'
]],
"qtcbc": [[u'--', u'----', u'0---', u'+---', u'+0--', u'++--', u'++']]
}
correct_hashsum = {
"qtcb": "de1ea75d1b0d6c9ff8249a24583fedb9",
"qtcc": "5d74fb27e53ba00f84014d3be70e2740",
"qtcbc": "f73e2c85f1447a5109d6ab3d5201fb76",
"rcc3": "402c53a1cc004a5f518d0b607eb7ac38"
}
correct_loglikelihoods = {
"qtcb": -2.16887,
"qtcc": -6.07188,
"qtcbc": -2.23475,
"rcc3": -4.15914
}
def __init__(self, *args):
super(TestHMM, self).__init__(*args)
rospy.init_node(NAME)
self.r = ROSClient()
def _create_hmm(self, qsr_file, qsr_type, start_at_zero=True):
with open(qsr_file, 'r') as f:
qsr_seq = json.load(f)
d = self.r.call_service(
HMMRepRequestCreate(qsr_seq=qsr_seq,
qsr_type=qsr_type,
start_at_zero=start_at_zero))
return d
def _create_sample(self, hmm_file, qsr_type):
with open(hmm_file, 'r') as f:
hmm = json.load(f)
s = self.r.call_service(
HMMRepRequestSample(qsr_type=qsr_type,
dictionary=hmm,
max_length=10,
num_samples=1))
return s
def _calculate_loglikelihood(self, hmm_file, qsr_file, qsr_type):
with open(qsr_file, 'r') as f:
qsr_seq = json.load(f)
with open(hmm_file, 'r') as f:
hmm = json.load(f)
l = self.r.call_service(
HMMRepRequestLogLikelihood(qsr_type=qsr_type,
dictionary=hmm,
qsr_seq=qsr_seq))
return round(l, 5)
def _to_strings(self, array):
return [x.values()[0] for x in array]
def test_qtcb_create(self):
res = self._create_hmm(self.QTCB_QSR, 'qtcb')
self.assertEqual(
hashlib.md5(json.dumps(res)).hexdigest(),
self.correct_hashsum["qtcb"])
def test_qtcb_sample(self):
res = self._create_sample(self.QTCB_SAMPLE_TEST_HMM, 'qtcb')
self.assertEqual(res, self.correct_samples["qtcb"])
def test_qtcb_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCB_PASSBY_LEFT_HMM,
self.QTCB_QSR, 'qtcb')
self.assertEqual(res, self.correct_loglikelihoods["qtcb"])
def test_qtcc_create(self):
res = self._create_hmm(self.QTCC_QSR, 'qtcc')
self.assertEqual(
hashlib.md5(json.dumps(res)).hexdigest(),
self.correct_hashsum["qtcc"])
def test_qtcc_sample(self):
res = self._create_sample(self.QTCC_SAMPLE_TEST_HMM, 'qtcc')
self.assertEqual(res, self.correct_samples["qtcc"])
def test_qtcc_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCC_PASSBY_LEFT_HMM,
self.QTCC_QSR, 'qtcc')
self.assertEqual(res, self.correct_loglikelihoods["qtcc"])
def test_qtcbc_create(self):
res = self._create_hmm(self.QTCBC_QSR, 'qtcbc')
self.assertEqual(
hashlib.md5(json.dumps(res)).hexdigest(),
self.correct_hashsum["qtcbc"])
def test_qtcbc_sample(self):
res = self._create_sample(self.QTCBC_SAMPLE_TEST_HMM, 'qtcbc')
self.assertEqual(res, self.correct_samples["qtcbc"])
def test_qtcbc_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCBC_PASSBY_LEFT_HMM,
self.QTCBC_QSR, 'qtcbc')
self.assertEqual(res, self.correct_loglikelihoods["qtcbc"])
def test_rcc3_create(self):
res = self._create_hmm(self.RCC3_QSR, 'rcc3')
self.assertEqual(
hashlib.md5(json.dumps(res)).hexdigest(),
self.correct_hashsum["rcc3"])
def test_rcc3_sample(self):
res = self._create_sample(self.RCC3_TEST_HMM, 'rcc3')
self.assertTrue(type(res) == list and len(res) > 0)
def test_rcc3_loglikelihood(self):
res = self._calculate_loglikelihood(self.RCC3_TEST_HMM, self.RCC3_QSR,
'rcc3')
self.assertEqual(res, self.correct_loglikelihoods["rcc3"])
log_parse = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter,add_help=False)
log_parse.add_argument('-i', '--input', help="the xml file containing the HMM", type=str, required=True)
log_parse.add_argument('-q', '--qsr_seq', help="reads a file containing state chains", type=str, required=True)
subparsers.add_parser('loglikelihood', parents=[general, log_parse, qtc_parse])
# Parse arguments
args = parser.parse_args()
rospy.init_node("ros_client")
r = ROSClient()
if args.action == "create":
qsr_seq = load_files(args.input)
q, d = r.call_service(
HMMRepRequestCreate(
qsr_seq=qsr_seq,
qsr_type=args.qsr_type
)
)
with open(args.output, 'w') as f: f.write(d)
elif args.action == "sample":
with open(args.input, 'r') as f: hmm = f.read()
q, s = r.call_service(
HMMRepRequestSample(
qsr_type=args.qsr_type,
xml=hmm,
max_length=args.max_length,
num_samples=args.num_samples
)
)
try:
# Parse arguments
args = parser.parse_args()
rospy.init_node("ros_client")
r = ROSClient()
if args.action == "create":
models = load_files(args.model)
states = load_lookup(args.states)
d = r.call_service(
PfRepRequestCreate(
num_particles=args.num_particles,
models=models,
state_lookup_table=states,
starvation_factor=args.starvation_factor,
ensure_particle_per_state=args.ensure_particle_per_state,
debug=args.debug,
uuid=args.uuid if args.uuid != "" else None
)
)
print d
elif args.action == "predict":
p = r.call_service(
PfRepRequestPredict(
uuid=args.uuid,
num_steps=args.num_steps,
debug=args.debug
)
)
class TestHMM(unittest.TestCase):
QTCB_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcb_sample_test.hmm')[0]
QTCC_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcc_sample_test.hmm')[0]
QTCBC_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcbc_sample_test.hmm')[0]
QTCB_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcb_passby_left.hmm')[0]
QTCC_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcc_passby_left.hmm')[0]
QTCBC_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcbc_passby_left.hmm')[0]
RCC3_TEST_HMM = find_resource(PKG, 'rcc3_test.hmm')[0]
QTCB_QSR = find_resource(PKG, 'qtcb.qsr')[0]
QTCC_QSR = find_resource(PKG, 'qtcc.qsr')[0]
QTCBC_QSR = find_resource(PKG, 'qtcbc.qsr')[0]
RCC3_QSR = find_resource(PKG, 'rcc3.qsr')[0]
correct_samples = {
"qtcb": [[u'--', u'0-', u'+-', u'+0', u'++']],
"qtcc": [[
u'--+-', u'--0-', u'----', u'0---', u'+---', u'+0--', u'++--',
u'---'
]],
"qtcbc": [[u'--', u'----', u'0---', u'+---', u'+0--', u'++--', u'++']]
}
correct_hashsum = {
"qtcb": "3fb65b50d0f7631a300132e8bca9ca13",
"qtcc": "dbf1529cb0b0c90aaebbe7eafe0e9b05",
"qtcbc": "0a3acf7b48c4c1155931442c86317ce4",
"rcc3": "6d5bcc6c44d9b1120c738efa1994a40a"
}
correct_loglikelihoods = {
"qtcb": -2.16887,
"qtcc": -6.07188,
"qtcbc": -2.23475,
"rcc3": -4.15914
}
def __init__(self, *args):
super(TestHMM, self).__init__(*args)
rospy.init_node(NAME)
self.r = ROSClient()
def _create_hmm(self, qsr_file, qsr_type):
with open(qsr_file, 'r') as f:
qsr_seq = json.load(f)
_, d = self.r.call_service(
HMMRepRequestCreate(qsr_seq=qsr_seq, qsr_type=qsr_type))
return d
def _create_sample(self, hmm_file, qsr_type):
with open(hmm_file, 'r') as f:
hmm = f.read()
_, s = self.r.call_service(
HMMRepRequestSample(qsr_type=qsr_type,
xml=hmm,
max_length=10,
num_samples=1))
return s
def _calculate_loglikelihood(self, hmm_file, qsr_file, qsr_type):
with open(qsr_file, 'r') as f:
qsr_seq = json.load(f)
with open(hmm_file, 'r') as f:
hmm = f.read()
q, l = self.r.call_service(
HMMRepRequestLogLikelihood(qsr_type=qsr_type,
xml=hmm,
qsr_seq=qsr_seq))
return round(l, 5)
def _to_strings(self, array):
return [x.values()[0] for x in array]
def test_qtcb_create(self):
res = self._create_hmm(self.QTCB_QSR, 'qtcb')
self.assertEqual(
hashlib.md5(res).hexdigest(), self.correct_hashsum["qtcb"])
def test_qtcb_sample(self):
res = self._create_sample(self.QTCB_SAMPLE_TEST_HMM, 'qtcb')
self.assertEqual(res, self.correct_samples["qtcb"])
def test_qtcb_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCB_PASSBY_LEFT_HMM,
self.QTCB_QSR, 'qtcb')
self.assertEqual(res, self.correct_loglikelihoods["qtcb"])
def test_qtcc_create(self):
res = self._create_hmm(self.QTCC_QSR, 'qtcc')
self.assertEqual(
hashlib.md5(res).hexdigest(), self.correct_hashsum["qtcc"])
def test_qtcc_sample(self):
res = self._create_sample(self.QTCC_SAMPLE_TEST_HMM, 'qtcc')
self.assertEqual(res, self.correct_samples["qtcc"])
def test_qtcc_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCC_PASSBY_LEFT_HMM,
self.QTCC_QSR, 'qtcc')
self.assertEqual(res, self.correct_loglikelihoods["qtcc"])
def test_qtcbc_create(self):
res = self._create_hmm(self.QTCBC_QSR, 'qtcbc')
self.assertEqual(
hashlib.md5(res).hexdigest(), self.correct_hashsum["qtcbc"])
def test_qtcbc_sample(self):
res = self._create_sample(self.QTCBC_SAMPLE_TEST_HMM, 'qtcbc')
self.assertEqual(res, self.correct_samples["qtcbc"])
def test_qtcbc_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCBC_PASSBY_LEFT_HMM,
self.QTCBC_QSR, 'qtcbc')
self.assertEqual(res, self.correct_loglikelihoods["qtcbc"])
def test_rcc3_create(self):
res = self._create_hmm(self.RCC3_QSR, 'rcc3')
self.assertEqual(
hashlib.md5(res).hexdigest(), self.correct_hashsum["rcc3"])
def test_rcc3_sample(self):
res = self._create_sample(self.RCC3_TEST_HMM, 'rcc3')
self.assertTrue(type(res) == list and len(res) > 0)
def test_rcc3_loglikelihood(self):
res = self._calculate_loglikelihood(self.RCC3_TEST_HMM, self.RCC3_QSR,
'rcc3')
self.assertEqual(res, self.correct_loglikelihoods["rcc3"])
class TestPf(unittest.TestCase):
CROSSING_OBS = find_resource(PKG, 'crossing.obs')[0]
CROSSING_PRED = find_resource(PKG, 'crossing.pred')[0]
PASSBY_OBS = find_resource(PKG, 'passby.obs')[0]
PASSBY_PRED = find_resource(PKG, 'passby.pred')[0]
STATES = find_resource(PKG, 'qtch.states')[0]
filters = map(str, range(5))
def __init__(self, *args):
super(self.__class__, self).__init__(*args)
rospy.init_node(NAME)
self.r = ROSClient()
def _create_model(self, obs, pred):
m = PfModel()
for f in pred:
name = f.split('/')[-1].split('.')[0]
with open(f, 'r') as a:
m.add_prediction_matrix(name, np.loadtxt(a))
for f in obs:
name = f.split('/')[-1].split('.')[0]
with open(f, 'r') as a:
m.add_observation_matrix(name, np.loadtxt(a))
return m.get()
def _load_lookup(self, filename):
with open(filename, 'r') as f:
return json.load(f)
def _create_pf(self, pred, obs, states, num_particles=1000, uuid="", starvation_factor=0.1, ensure_particle_per_state=True, debug=True):
models = self._create_model(pred, obs)
states = self._load_lookup(states)
d = self.r.call_service(
PfRepRequestCreate(
num_particles=num_particles,
models=models,
state_lookup_table=states,
starvation_factor=starvation_factor,
ensure_particle_per_state=ensure_particle_per_state,
debug=debug,
uuid=uuid if uuid != "" else None
)
)
return d
def _predict(self, uuid, num_steps=5, debug=True):
p = self.r.call_service(
PfRepRequestPredict(
uuid=uuid,
num_steps=num_steps,
debug=debug
)
)
return p
def _update(self, uuid, obs, debug=True):
p = self.r.call_service(
PfRepRequestUpdate(
uuid=uuid,
observation=obs,
debug=debug
)
)
return p
def _list(self):
p = self.r.call_service(
PfRepRequestList()
)
return p
def _remove(self, uuid):
p = self.r.call_service(
PfRepRequestRemove(
uuid=uuid
)
)
return p
def test1_pf_create(self):
res = []
for uuid in self.filters:
res.append(self._create_pf(
pred=[self.CROSSING_PRED, self.PASSBY_PRED],
obs=[self.CROSSING_OBS, self.PASSBY_OBS],
states=self.STATES,
uuid=uuid
))
self.assertEqual(map(str, res), self.filters)
def test2_pf_predict(self):
res = []
for uuid in self.filters:
res.append(self._predict(
uuid=uuid,
num_steps=1
)[0])
self.assertEqual(len(res), len(self.filters))
def test3_pf_update(self):
res = []
for uuid in self.filters:
res.append(self._update(
uuid=uuid,
obs='-,-,-'
))
self.assertEqual(len(res), len(self.filters))
def test4_pf_list(self):
res = self._list()
self.assertEqual(len(res), len(self.filters))
def test5_pf_remove(self):
for uuid in self.filters:
self._remove(uuid)
self.assertFalse(len(self._list()))
class TestHMM(unittest.TestCase):
QTCB_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcb_sample_test.hmm')[0]
QTCC_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcc_sample_test.hmm')[0]
QTCBC_SAMPLE_TEST_HMM = find_resource(PKG, 'qtcbc_sample_test.hmm')[0]
QTCB_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcb_passby_left.hmm')[0]
QTCC_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcc_passby_left.hmm')[0]
QTCBC_PASSBY_LEFT_HMM = find_resource(PKG, 'qtcbc_passby_left.hmm')[0]
RCC3_TEST_HMM = find_resource(PKG, 'rcc3_test.hmm')[0]
QTCB_QSR = find_resource(PKG, 'qtcb.qsr')[0]
QTCC_QSR = find_resource(PKG, 'qtcc.qsr')[0]
QTCBC_QSR = find_resource(PKG, 'qtcbc.qsr')[0]
RCC3_QSR = find_resource(PKG, 'rcc3.qsr')[0]
correct_samples = {
"qtcb": [[u'--', u'0-', u'+-', u'+0', u'++']],
"qtcc": [[u'--+-', u'--0-', u'----', u'0---', u'+---', u'+0--', u'++--', u'---']],
"qtcbc": [[u'--', u'----', u'0---', u'+---', u'+0--', u'++--', u'++']]
}
correct_hashsum = {
"qtcb": "de1ea75d1b0d6c9ff8249a24583fedb9",
"qtcc": "5d74fb27e53ba00f84014d3be70e2740",
"qtcbc": "f73e2c85f1447a5109d6ab3d5201fb76",
"rcc3": "402c53a1cc004a5f518d0b607eb7ac38"
}
correct_loglikelihoods ={
"qtcb": -2.16887,
"qtcc": -6.07188,
"qtcbc": -2.23475,
"rcc3": -4.15914
}
def __init__(self, *args):
super(TestHMM, self).__init__(*args)
rospy.init_node(NAME)
self.r = ROSClient()
def _create_hmm(self, qsr_file, qsr_type, start_at_zero=True):
with open(qsr_file, 'r') as f: qsr_seq = json.load(f)
d = self.r.call_service(
HMMRepRequestCreate(
qsr_seq=qsr_seq,
qsr_type=qsr_type,
start_at_zero=start_at_zero
)
)
return d
def _create_sample(self, hmm_file, qsr_type):
with open(hmm_file, 'r') as f: hmm = json.load(f)
s = self.r.call_service(
HMMRepRequestSample(
qsr_type=qsr_type,
dictionary=hmm,
max_length=10,
num_samples=1
)
)
return s
def _calculate_loglikelihood(self, hmm_file, qsr_file, qsr_type):
with open(qsr_file, 'r') as f: qsr_seq = json.load(f)
with open(hmm_file, 'r') as f: hmm = json.load(f)
l = self.r.call_service(
HMMRepRequestLogLikelihood(
qsr_type=qsr_type,
dictionary=hmm,
qsr_seq=qsr_seq
)
)
return round(l, 5)
def _to_strings(self, array):
return [x.values()[0] for x in array]
def test_qtcb_create(self):
res = self._create_hmm(self.QTCB_QSR, 'qtcb')
self.assertEqual(hashlib.md5(json.dumps(res)).hexdigest(), self.correct_hashsum["qtcb"])
def test_qtcb_sample(self):
res = self._create_sample(self.QTCB_SAMPLE_TEST_HMM, 'qtcb')
self.assertEqual(res, self.correct_samples["qtcb"])
def test_qtcb_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCB_PASSBY_LEFT_HMM, self.QTCB_QSR, 'qtcb')
self.assertEqual(res, self.correct_loglikelihoods["qtcb"])
def test_qtcc_create(self):
res = self._create_hmm(self.QTCC_QSR, 'qtcc')
self.assertEqual(hashlib.md5(json.dumps(res)).hexdigest(), self.correct_hashsum["qtcc"])
def test_qtcc_sample(self):
res = self._create_sample(self.QTCC_SAMPLE_TEST_HMM, 'qtcc')
self.assertEqual(res, self.correct_samples["qtcc"])
def test_qtcc_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCC_PASSBY_LEFT_HMM, self.QTCC_QSR, 'qtcc')
self.assertEqual(res, self.correct_loglikelihoods["qtcc"])
def test_qtcbc_create(self):
res = self._create_hmm(self.QTCBC_QSR, 'qtcbc')
self.assertEqual(hashlib.md5(json.dumps(res)).hexdigest(), self.correct_hashsum["qtcbc"])
def test_qtcbc_sample(self):
res = self._create_sample(self.QTCBC_SAMPLE_TEST_HMM, 'qtcbc')
self.assertEqual(res, self.correct_samples["qtcbc"])
def test_qtcbc_loglikelihood(self):
res = self._calculate_loglikelihood(self.QTCBC_PASSBY_LEFT_HMM, self.QTCBC_QSR, 'qtcbc')
self.assertEqual(res, self.correct_loglikelihoods["qtcbc"])
def test_rcc3_create(self):
res = self._create_hmm(self.RCC3_QSR, 'rcc3')
self.assertEqual(hashlib.md5(json.dumps(res)).hexdigest(), self.correct_hashsum["rcc3"])
def test_rcc3_sample(self):
res = self._create_sample(self.RCC3_TEST_HMM, 'rcc3')
self.assertTrue(type(res) == list and len(res) > 0)
def test_rcc3_loglikelihood(self):
res = self._calculate_loglikelihood(self.RCC3_TEST_HMM, self.RCC3_QSR, 'rcc3')
self.assertEqual(res, self.correct_loglikelihoods["rcc3"])
class TestPf(unittest.TestCase):
CROSSING_OBS = find_resource(PKG, 'crossing.obs')[0]
CROSSING_PRED = find_resource(PKG, 'crossing.pred')[0]
PASSBY_OBS = find_resource(PKG, 'passby.obs')[0]
PASSBY_PRED = find_resource(PKG, 'passby.pred')[0]
STATES = find_resource(PKG, 'qtch.states')[0]
filters = map(str, range(5))
def __init__(self, *args):
super(self.__class__, self).__init__(*args)
rospy.init_node(NAME)
self.r = ROSClient()
def _create_model(self, obs, pred):
m = PfModel()
for f in pred:
name = f.split('/')[-1].split('.')[0]
with open(f, 'r') as a:
m.add_prediction_matrix(name, np.loadtxt(a))
for f in obs:
name = f.split('/')[-1].split('.')[0]
with open(f, 'r') as a:
m.add_observation_matrix(name, np.loadtxt(a))
return m.get()
def _load_lookup(self, filename):
with open(filename, 'r') as f:
return json.load(f)
def _create_pf(self,
pred,
obs,
states,
num_particles=1000,
uuid="",
starvation_factor=0.1,
ensure_particle_per_state=True,
debug=True):
models = self._create_model(pred, obs)
states = self._load_lookup(states)
d = self.r.call_service(
PfRepRequestCreate(
num_particles=num_particles,
models=models,
state_lookup_table=states,
starvation_factor=starvation_factor,
ensure_particle_per_state=ensure_particle_per_state,
debug=debug,
uuid=uuid if uuid != "" else None))
return d
def _predict(self, uuid, num_steps=5, debug=True):
p = self.r.call_service(
PfRepRequestPredict(uuid=uuid, num_steps=num_steps, debug=debug))
return p
def _update(self, uuid, obs, debug=True):
p = self.r.call_service(
PfRepRequestUpdate(uuid=uuid, observation=obs, debug=debug))
return p
def _list(self):
p = self.r.call_service(PfRepRequestList())
return p
def _remove(self, uuid):
p = self.r.call_service(PfRepRequestRemove(uuid=uuid))
return p
def test1_pf_create(self):
res = []
for uuid in self.filters:
res.append(
self._create_pf(pred=[self.CROSSING_PRED, self.PASSBY_PRED],
obs=[self.CROSSING_OBS, self.PASSBY_OBS],
states=self.STATES,
uuid=uuid))
self.assertEqual(map(str, res), self.filters)
def test2_pf_predict(self):
res = []
for uuid in self.filters:
res.append(self._predict(uuid=uuid, num_steps=1)[0])
self.assertEqual(len(res), len(self.filters))
def test3_pf_update(self):
res = []
for uuid in self.filters:
res.append(self._update(uuid=uuid, obs='-,-,-'))
self.assertEqual(len(res), len(self.filters))
def test4_pf_list(self):
res = self._list()
self.assertEqual(len(res), len(self.filters))
def test5_pf_remove(self):
for uuid in self.filters:
self._remove(uuid)
self.assertFalse(len(self._list()))
'--qsr_seq',
help="reads a file containing state chains",
type=str,
required=True)
subparsers.add_parser('loglikelihood',
parents=[general, log_parse, qtc_parse])
# Parse arguments
args = parser.parse_args()
rospy.init_node("ros_client")
r = ROSClient()
if args.action == "create":
qsr_seq = load_files(args.input)
q, d = r.call_service(
HMMRepRequestCreate(qsr_seq=qsr_seq, qsr_type=args.qsr_type))
with open(args.output, 'w') as f:
f.write(d)
elif args.action == "sample":
with open(args.input, 'r') as f:
hmm = f.read()
q, s = r.call_service(
HMMRepRequestSample(qsr_type=args.qsr_type,
xml=hmm,
max_length=args.max_length,
num_samples=args.num_samples))
try:
with open(args.output, 'w') as f:
json.dump(s, f)
except TypeError:
subparsers.add_parser('remove', parents=[general, remove_parse])
# Parse arguments
args = parser.parse_args()
rospy.init_node("ros_client")
r = ROSClient()
if args.action == "create":
models = load_files(args.model)
states = load_lookup(args.states)
d = r.call_service(
PfRepRequestCreate(
num_particles=args.num_particles,
models=models,
state_lookup_table=states,
starvation_factor=args.starvation_factor,
ensure_particle_per_state=args.ensure_particle_per_state,
debug=args.debug,
uuid=args.uuid if args.uuid != "" else None))
print d
elif args.action == "predict":
p = r.call_service(
PfRepRequestPredict(uuid=args.uuid,
num_steps=args.num_steps,
debug=args.debug))
print p
elif args.action == "update":
p = r.call_service(
PfRepRequestUpdate(uuid=args.uuid,