def setUp(self):
self.gesture_db = GestureDatabase()
self.gesture = self.gesture_db.get_gesture_entry("come_here")
self.D = self.gesture.dimensionality
self.gesture.process_signal_accommodation()
self.coordinates = self.gesture.get_merged_training_data()
for i in xrange(len(self.coordinates)):
for j in xrange(len(self.coordinates[i])):
self.coordinates[i][j] = self.coordinates[i][j] * 100.0
self.__learning_system = None
class TestGMMGestureModel(unittest.TestCase):
def setUp(self):
self.gesture_db = GestureDatabase()
self.gesture = self.gesture_db.get_gesture_entry("come_here")
self.D = self.gesture.dimensionality
self.gesture.process_signal_accommodation()
self.coordinates = self.gesture.get_merged_training_data()
for i in xrange(len(self.coordinates)):
for j in xrange(len(self.coordinates[i])):
self.coordinates[i][j] = self.coordinates[i][j] * 100.0
self.__learning_system = None
@property
def learning_system(self):
if self.__learning_system == None:
self.__learning_system = GMMGestureModel.create_and_learn(
self.D, "test_gesture", self.coordinates, STOP_TH,
MAX_CLUSTERS)
return self.__learning_system
def test_basic_comparation(self):
a = self.learning_system
return
def test_check_auto_likelihood(self):
other_gesture = copy(self.coordinates)
other_gesture[0][0] = other_gesture[0][0] * 2.0
self.assertGreater(
self.learning_system.get_training_set_likelihood(),
self.learning_system.compute_gesture_likelihood(other_gesture))
def test_save_load_model(self):
filename = INPUT_DIR + "/models/temp_gmm_model.yaml"
self.learning_system.save_to_file(filename)
loaded = GMMGestureModel.load_from_file(filename)
loaded_training = loaded.get_training_set_data()
self.assertEquals(len(loaded_training), len(self.coordinates))
self.assertEquals(len(loaded_training[0]), len(self.coordinates[0]))
help=
"models to evaluate. if no specified, all that are evaluated. Otherwise provide comma separated models names ie: -m HMMGestureModel,GMMGestureModel,Time-GMMGestureModel,DTWGestureModel"
)
(options, args) = parser.parse_args()
if options.gesture_entries != None:
gesture_entries_names = options.gesture_entries.split(",")
else:
#default gestures to evaluate
rospy.loginfo(
"gestures must be specified using the parameter -g or (--gestures-entries). See help."
)
exit()
rospy.loginfo("gestures to be analyzed: %s", str(gesture_entries_names))
gesture_db = GestureDatabase()
#gesture_entries_names=["come_here"]
print "models to load: %s" % str(options.models_types)
if options.models_types == None:
models_names = gesture_db.get_models_names()
else:
models_types_name = options.models_types.split(",")
models_names = []
for mt in models_types_name:
models_names = models_names + gesture_db.get_models_names(
model_type_name=mt)
print "models found: %s" % str(models_names)
#cache for avoiding load models
def generate_images(gestures_to_paint, time_wrapping=False):
gesture_db = GestureDatabase()
images = []
max_number_demos = 24
functions = get_plot_functions()
image_index = 0
for gesture_entry_name in gestures_to_paint:
#load gesture
gesture = gesture_db.get_gesture_entry(gesture_entry_name)
print "loading gesture: %s" % gesture_entry_name
try:
#data preprocessing
gesture.process_signal_accommodation(
offset_accomodate=True,
demo_longitude_accomodate=True,
regular_sampling_acomodation=True,
time_wrapping=False)
for f in functions:
#create plotting infrastructure
fig1 = matplotlib.pyplot.figure(image_index)
fig1.clear()
leg = []
#plot
number_of_demos = min(max_number_demos,
gesture.demonstration_count)
m = None
(plot_key_name, leg) = f(gesture, m, fig1, number_of_demos)
if leg != None:
legend(tuple(leg))
images.append((gesture.name, fig1))
if m != None:
imageurl = REPORT_DIR + "/" + f.__name__ + "/" + gesture_entry_name + "-" + plot_key_name + ".png"
else:
imageurl = REPORT_DIR + "/" + f.__name__ + "/" + gesture_entry_name + ".png"
fig1.savefig(imageurl)
image_index += 1
except:
continue
'''models=gesture_db.get_models_by_gesture_name(gesture.name+".time.*k2time_wrapped\.gmm")
print "loading models: %s"+str(models)
models= [None]+models
for m in models:
for f in functions:
#create plotting infrastructure
fig1 = matplotlib.pyplot.figure(image_index)
fig1.clear()
leg=[]
#plot
number_of_demos=min(max_number_demos,gesture.demonstration_count)
(plot_key_name,leg)=f(gesture,m,fig1,number_of_demos)
if leg!=None:
legend(tuple(leg))
images.append((gesture.name,fig1))
imageurl=REPORT_DIR+"/"+f.__name__+"/"+gesture_entry_name+"-"+plot_key_name+".png"
fig1.savefig(imageurl)
image_index+=1
'''
return images
default=None,
dest="negative_gestures_entries",
help=
"comma separated negative gestures names for the threshold calculation. Example: --negative-gestures-entries come_here_for_training,clapping"
)
parser.add_option(
"-m",
"--models-types",
default=None,
dest="models_types",
help=
"if no specified, all models are generated. Otherwise provide comma separated models names ie: -m HMMGestureModel,GMMGestureModel,Time-GMMGestureModel,DTWGestureModel"
)
(options, args) = parser.parse_args()
gesture_db = GestureDatabase()
if options.directory != None:
if os.path.exists(
options.directory) and not os.path.isfile(options.directory):
gesture_db.GESTURE_DATABASE_DIRECTORY = options.directory
else:
rospy.logfatal("Incorrect directory. exiting.")
exit(-1)
models_types = "all"
if options.models_types != None:
models_types = options.models_types.split(",")
print "[%s] models to generate: %s" % (options.models_types,
str(models_types))
"--loglevel",
default="INFO",
dest="loglevel",
help="loglevel \"INFO\" (default) or \"DEBUG\"")
(options, args) = parser.parse_args()
if options.loglevel == "DEBUG":
loglevel = rospy.DEBUG
rospy.loginfo("DEBUG logging mode")
else:
loglevel = rospy.INFO
rospy.init_node('batch_gesture_entries_from_bags')
bags_db = BagDatabase()
gesture_db = GestureDatabase()
if options.directory != "":
if os.path.exists(
options.directory) and not os.path.isfile(options.directory):
bags_db.BAGS_DIRECTORY = options.directory
else:
rospy.logfatal("Incorrect directory. exiting.")
exit(-1)
rospy.set_param("sim_time", True)
#files on the directory root are modeled independently
rospy.loginfo("input bag dir: %s", bags_db.BAGS_DIRECTORY)
#selecting config files
root_config_files = bags_db.get_root_config_files()
# MAIN
# -------------------------------------------------------
if __name__ == '__main__':
from optparse import OptionParser
parser = OptionParser("%prog [ opts ]\nThis examples take a gesture, creates the TMM model in the task space and show results in rviz")
parser.add_option("-g","--gesture-entry", default="task_space_wave", dest="gesture_entry", help="gesture to be get. Default: task_space_wave")
(options, args) = parser.parse_args()
rospy.init_node ('GMM_learning', anonymous = True)
if PLOT:
clusters_pub = rospy.Publisher ("visualization_marker", Marker)
gauss_pub = rospy.Publisher ("gaussian", Marker)
gesture_db=GestureDatabase()
rospy.loginfo("loading gesture entry: %s",options.gesture_entry)
gesture=gesture_db.get_gesture_entry(options.gesture_entry)
#print gesture.raw_data[0]
gesture.process_signal_accommodation(offset_accomodate=True,demo_longitude_accomodate=True,regular_sampling_acomodation=True,time_wrapping=False)
D=gesture.dimensionality
coordinates=gesture.get_training_data(demo_index=0,scale=1.0)
#explain the problem with singular matrixes ie: elbow -> shoulder... 2d
rospy.loginfo("%s points loaded from database",str(len(coordinates[0])))
coordinates= outlier_removal(coordinates)
rospy.loginfo("outliers removed, points len: %s",str(len(coordinates[0])))
if K == None:
gmm_gesture_model = GMMGestureModel.create_and_learn(options.gesture_entry,D,coordinates,STOP_TH,MAX_CLUSTERS)