def __init__(self, trajectoryQueue, sync, stop, dumpTrajectories):
self.trajectoryFinder = FindCorrespondingTrajectories(8, False)
self.dumpTrajectories = dumpTrajectories
self.filePath = 'trajectories.csv'
''' use adacious as example application
make artificial circle trajectories. Debug purpose only'''
if platform == 'linux' or platform == 'linux2':
self.audacious = Audacious()
self.keystroke = KeystrokeSimulator()
#self.__getDebugCircleData()
self.__setTrajectoryData()
LearningThread.__init__(self, trajectoryQueue, sync, stop)
def __init__(self, trajectoryQueue, sync, stop, dumpTrajectories):
self.trajectoryFinder = FindCorrespondingTrajectories(8,False)
self.dumpTrajectories = dumpTrajectories
self.filePath = 'trajectories.csv'
''' use adacious as example application
make artificial circle trajectories. Debug purpose only'''
if platform == 'linux' or platform == 'linux2':
self.audacious = Audacious()
self.keystroke = KeystrokeSimulator()
#self.__getDebugCircleData()
self.__setTrajectoryData()
LearningThread.__init__(self, trajectoryQueue, sync, stop)
class BinaryLearningThread(LearningThread):
'''
trajectory dict:
#1 Up/Down
#2 to the right
'''
def __init__(self, trajectoryQueue, sync, stop, dumpTrajectories):
self.trajectoryFinder = FindCorrespondingTrajectories(8,False)
self.dumpTrajectories = dumpTrajectories
self.filePath = 'trajectories.csv'
''' use adacious as example application
make artificial circle trajectories. Debug purpose only'''
if platform == 'linux' or platform == 'linux2':
self.audacious = Audacious()
self.keystroke = KeystrokeSimulator()
#self.__getDebugCircleData()
self.__setTrajectoryData()
LearningThread.__init__(self, trajectoryQueue, sync, stop)
def dumpData(self, npTraj):
f = open(self.filePath, 'a')
separater = ','
dataString= ''
for row in npTraj:
rowString = ''
for entry in row:
rowString += str(entry)+separater
dataString += rowString+'\n'
print dataString
f.write(dataString)
def run(self):
trajectory = []
while True:
self.event.wait()
if self.stop.isSet():
print 'learning will stop'
return
try:
trajectoryPoint = self.trajectoryQueue.popDataPoint()
ID =trajectoryPoint.getID()
if ID == -1:
ID = trajectoryPoint.getID()
trajectory = []
elif ID == -2:
npTraj = np.transpose(np.array(trajectory))
if self.dumpTrajectories:
self.dumpData(npTraj)
print 'trajectory will be sent'
foundId = self.trajectoryFinder.findCorrespondenceOverDirection(npTraj)
print "found traj id:\t"+str(foundId)
self.doAction(foundId)
trajectory = []
else:
loc = trajectoryPoint.getLocation()
trajectory.append([loc[0], loc[1], trajectoryPoint.getTime()])
except IndexError:
print 'read from empty queue'
self.event.clear()
# print 'learning thread goes to sleep'
def doAction(self, foundId):
if self.audacious != None:
if foundId == 1:
self.audacious.pause()
if foundId == 2:
self.audacious.play_next()
if self.keystroke != None:
if foundId == 2:
self.keystroke.sendKey()
print('action')
def __readFileData(self, filepath):
trajectories = []
f1 = open(filepath, 'r')
counter = 0
trajectory = []
for line in f1:
counter +=1
values = np.fromstring(line, dtype=np.float64, sep=',')
trajectory.append(values)
if counter%3==0:
trajectories.append(np.array(trajectory))
trajectory = []
return trajectories
def __setTrajectoryData(self):
trajectory1 = self.__readFileData('trajectories_1.csv')
self.trajectoryFinder.learnTrajectory(trajectory1, 1)
trajectory2 = self.__readFileData('trajectories_2.csv')
self.trajectoryFinder.learnTrajectory(trajectory2, 2)
#########################################################
#########################################################
#########################################################
def corruptAndCopy(self, noiseRate,copyNumber,trajectory):
corruptedTrajectory = [];
for i in range(copyNumber):
corrupted = np.copy(trajectory);
noise = noiseRate*np.vstack((np.random.randn(corrupted.shape[1]),np.random.randn(corrupted.shape[1])))
corrupted[0:2,:] = corrupted[0:2,:] + noise
corruptedTrajectory.append(corrupted)
return corruptedTrajectory;
def makeCircle(self, w,time):
x = np.cos(w*time);
y = np.sin(w*time);
return np.vstack((x,y,time))
def makeLine(self,theta,velocity,time):
x = np.cos(theta)*velocity*time;
y = np.sin(theta)*velocity*time;
return np.vstack((x,y,time))
def makeWave(self,v,w,time):
x = v*time;
y = np.sin(w*time);
return np.vstack((x,y,time))
def __getDebugCircleData(self):
noiseRate = 0.001;
T = 3;
resolution = 1./30;
steps = T/resolution;
time = np.linspace(0, T, steps);
####### Generate Training-data and train ##########################
k = 5
# line = self.makeLine(180./180*np.pi,1./T,time)
# corLines = self.corruptAndCopy(noiseRate,k,line)
# self.trajectoryFinder.learnTrajectory(corLines, 1)
##
# line = self.makeLine(0./180*np.pi,1./T,time)
# corLines = self.corruptAndCopy(noiseRate,k,line)
# self.trajectoryFinder.learnTrajectory(corLines, 2)
line = self.makeCircle(2*np.pi/T,time)
corLines = self.corruptAndCopy(noiseRate,k,line)
self.trajectoryFinder.learnTrajectory(corLines, 3)
line = self.makeWave(1./T,2*np.pi,time)
corLines = self.corruptAndCopy(noiseRate,k,line)
self.trajectoryFinder.learnTrajectory(corLines, 4)
class BinaryLearningThread(LearningThread):
'''
trajectory dict:
#1 Up/Down
#2 to the right
'''
def __init__(self, trajectoryQueue, sync, stop, dumpTrajectories):
self.trajectoryFinder = FindCorrespondingTrajectories(8, False)
self.dumpTrajectories = dumpTrajectories
self.filePath = 'trajectories.csv'
''' use adacious as example application
make artificial circle trajectories. Debug purpose only'''
if platform == 'linux' or platform == 'linux2':
self.audacious = Audacious()
self.keystroke = KeystrokeSimulator()
#self.__getDebugCircleData()
self.__setTrajectoryData()
LearningThread.__init__(self, trajectoryQueue, sync, stop)
def dumpData(self, npTraj):
f = open(self.filePath, 'a')
separater = ','
dataString = ''
for row in npTraj:
rowString = ''
for entry in row:
rowString += str(entry) + separater
dataString += rowString + '\n'
print dataString
f.write(dataString)
def run(self):
trajectory = []
while True:
self.event.wait()
if self.stop.isSet():
print 'learning will stop'
return
try:
trajectoryPoint = self.trajectoryQueue.popDataPoint()
ID = trajectoryPoint.getID()
if ID == -1:
ID = trajectoryPoint.getID()
trajectory = []
elif ID == -2:
npTraj = np.transpose(np.array(trajectory))
if self.dumpTrajectories:
self.dumpData(npTraj)
print 'trajectory will be sent'
foundId = self.trajectoryFinder.findCorrespondenceOverDirection(
npTraj)
print "found traj id:\t" + str(foundId)
self.doAction(foundId)
trajectory = []
else:
loc = trajectoryPoint.getLocation()
trajectory.append(
[loc[0], loc[1],
trajectoryPoint.getTime()])
except IndexError:
print 'read from empty queue'
self.event.clear()
# print 'learning thread goes to sleep'
def doAction(self, foundId):
if self.audacious != None:
if foundId == 1:
self.audacious.pause()
if foundId == 2:
self.audacious.play_next()
if self.keystroke != None:
if foundId == 2:
self.keystroke.sendKey()
print('action')
def __readFileData(self, filepath):
trajectories = []
f1 = open(filepath, 'r')
counter = 0
trajectory = []
for line in f1:
counter += 1
values = np.fromstring(line, dtype=np.float64, sep=',')
trajectory.append(values)
if counter % 3 == 0:
trajectories.append(np.array(trajectory))
trajectory = []
return trajectories
def __setTrajectoryData(self):
trajectory1 = self.__readFileData('trajectories_1.csv')
self.trajectoryFinder.learnTrajectory(trajectory1, 1)
trajectory2 = self.__readFileData('trajectories_2.csv')
self.trajectoryFinder.learnTrajectory(trajectory2, 2)
#########################################################
#########################################################
#########################################################
def corruptAndCopy(self, noiseRate, copyNumber, trajectory):
corruptedTrajectory = []
for i in range(copyNumber):
corrupted = np.copy(trajectory)
noise = noiseRate * np.vstack((np.random.randn(
corrupted.shape[1]), np.random.randn(corrupted.shape[1])))
corrupted[0:2, :] = corrupted[0:2, :] + noise
corruptedTrajectory.append(corrupted)
return corruptedTrajectory
def makeCircle(self, w, time):
x = np.cos(w * time)
y = np.sin(w * time)
return np.vstack((x, y, time))
def makeLine(self, theta, velocity, time):
x = np.cos(theta) * velocity * time
y = np.sin(theta) * velocity * time
return np.vstack((x, y, time))
def makeWave(self, v, w, time):
x = v * time
y = np.sin(w * time)
return np.vstack((x, y, time))
def __getDebugCircleData(self):
noiseRate = 0.001
T = 3
resolution = 1. / 30
steps = T / resolution
time = np.linspace(0, T, steps)
####### Generate Training-data and train ##########################
k = 5
# line = self.makeLine(180./180*np.pi,1./T,time)
# corLines = self.corruptAndCopy(noiseRate,k,line)
# self.trajectoryFinder.learnTrajectory(corLines, 1)
##
# line = self.makeLine(0./180*np.pi,1./T,time)
# corLines = self.corruptAndCopy(noiseRate,k,line)
# self.trajectoryFinder.learnTrajectory(corLines, 2)
line = self.makeCircle(2 * np.pi / T, time)
corLines = self.corruptAndCopy(noiseRate, k, line)
self.trajectoryFinder.learnTrajectory(corLines, 3)
line = self.makeWave(1. / T, 2 * np.pi, time)
corLines = self.corruptAndCopy(noiseRate, k, line)
self.trajectoryFinder.learnTrajectory(corLines, 4)
