def main():
svg_parser = SvgParser.SvgParser(
"ground-truth/locations/", "images/", "task/train.txt",
"task/valid.txt", "task/test.txt", "ground-truth/transcription.txt",
False)
svg_parser.get_cropped_images()
print_timer("Cropping")
training_samples, validation_samples, test_samples, transcriptions = svg_parser.get_binarized_word_images(50)
output_lines = []
print("Keyword spotting...")
# Iterate trough test_keywords and for each keyword, find matching training sample, search for it in test samples.
with open("test_keywords.txt", 'r') as f:
for line in f.readlines():
line = line.rstrip()
keyword_and_id = line.split(',')
keyword = svg_parser.binarize(
"cropped/train/%s.png" % keyword_and_id[1], 50)
dtw = DTW(keyword)
output_line = [keyword_and_id[0]]
print("searching %s" % keyword_and_id[0])
for sample in test_samples:
print("comparing to %s" % sample[1])
output_line.append(sample[1])
output_line.append(
dtw.calculate_cost_and_matrix(sample[0])[0])
output_lines.append(output_line)
print(output_line)
with open("output.csv", "w") as f:
writer = csv.writer(f)
writer.writerows(output_lines)
def apply_dtw(template):
dtw = DTW(template)
results = [
dtw.calculate_cost_and_matrix(enr) for enr in enrollment
if enr.get_user() == template.get_user()
]
template.cost = min(results)
def classify(self, seq_test):
''' Classify a sequence based in its 1-nearest neighbor '''
id_class = -1
min_distance = float('inf')
for seq in self.training_data:
dtw = DTW(seq[1], seq_test)
tmp_distance = dtw.iterative_dtw() #recursive_dtw(len(seq[1])-1, len(seq_test)-1)
if tmp_distance < min_distance:
id_class = seq[0]
min_distance = tmp_distance
return id_class
def KNN(test, trainSet):
k = 5
content = []
ids = []
counter=0
results = []
for item in trainSet:
item = item[1]
x = DTW(test, item)
#print x
content.append(x)
ids.append(counter)
counter += 1
max1 = max(set(content))
min = max1
for j in range(0, k, 1):
for i in range(0, len(content), 1):
if i in results:
continue
if content[i] < min:
min = content[i]
minPos = i
results.append(minPos)
print minPos, " - ", content[minPos]
min = max1
return results
'''
spin1_2 = fingers_spin[0]
spin2_2 = fingers_spin[1]
spin3_2 = fingers_spin[2]
spin4_2 = fingers_spin[3]
spin5_2 = fingers_spin[4]
roll1_2 = fingers_roll[0]
roll2_2 = fingers_roll[1]
roll3_2 = fingers_roll[2]
roll4_2 = fingers_roll[3]
roll5_2 = fingers_roll[4]
f.close()
print(DTW(finger1_1, finger1_2, 0))
print(DTW(finger2_1, finger2_2, 0))
print(DTW(finger3_1, finger3_2, 0))
print(DTW(finger4_1, finger4_2, 0))
print(DTW(finger5_1, finger5_2, 0))
print('/n')
print(DTW(spin1_1, spin1_2, 1))
print(DTW(spin2_1, spin2_2, 1))
print(DTW(spin3_1, spin3_2, 1))
print(DTW(spin4_1, spin4_2, 1))
print(DTW(spin5_1, spin5_2, 1))
print('/n')
print(DTW(roll1_1, roll1_2, 2))
print(DTW(roll2_1, roll2_2, 2))
print(DTW(roll3_1, roll3_2, 2))
print(DTW(roll4_1, roll4_2, 2))
def print_timer(purpose_message=""):
global start_time
print("%s timer: %ss" % (purpose_message, int(time.clock()-start_time)))
start_time = time.clock()
""" Reading Data """
svg_parser = SvgParser.SvgParser("ground-truth/locations/", "images/", "task/train.txt", "task/valid.txt")
svg_parser.get_cropped_images()
print_timer("Cropping")
training_samples, validation_samples = svg_parser.get_binarized_word_images(50)
print("binarized training sample size: %s" % len(training_samples))
print_timer("Binarizing")
""" Calculating Features """
features = features.calculate_features(training_samples)
print("Features of trainingset[0]: %s" % features[0])
print_timer("Feature computation")
""" Searching for Keyword """
keyword = svg_parser.binarize("cropped/train/270_32_out.png", 50)
dtw = DTW(keyword)
results = [dtw.calculate_cost_and_matrix(sample) for sample in training_samples]
print_timer("Computing String Distance")
print([costs[0] for costs in sorted(results)][:30])
print("Word found in image numbers: &s" % [i for i, x in enumerate(results[:, 0]) if x < 6000])
index_col='tripId')
testSet = pd.read_csv(
'test_dataset/test_set_a1.csv', # replace with the correct path
sep="\t",
converters={"Trajectory": literal_eval})
path = 'Test_Trip/'
if not os.path.exists(path):
os.makedirs(path)
#trainSet = trainSet[:200]
for i in range(len(testSet)):
start_time = time.time()
origin = testSet['Trajectory'].iloc[i]
knn = KNN(5, DTW(Harvesine), origin, False)
for j in range(len(trainSet)):
knn.calculate_neighbor(trainSet['Trajectory'].iloc[j], j)
results = knn.results()
elapsed_time = time.time() - start_time
path = 'Test_Trip/Test_Trip_' + str(i + 1)
if not os.path.exists(path):
os.makedirs(path)
file = open(path + '/results', 'w')
file.write("Test_Trip_" + str(i + 1) + "\n")
print_map(origin, 'Test_Trip_' + str(i + 1), path)
count = 1
for k in results:
print_map(trainSet['Trajectory'].iloc[k[0]], 'Neighbor_' + str(count),
path)
file.write('Neighbor_' + str(count) + '\nJP_ID: ' +
def apply_dtw(template):
dtw = DTW(template)
results = [dtw.calculate_cost_and_matrix(enr) for enr in enrollment if enr.get_user() == template.get_user()]
template.cost = min(results)
def trainTask(self):
def stopper(self):
raw_input('Press Enter to stop demonstration.')
self.stopperBool = False
def recordData(self):
dataLeft = np.zeros((6, 0))
dataRight = np.zeros((6, 0))
timeArray = np.array([[0.1]])
thread.start_new_thread(stopper, (self, ))
while self.stopperBool:
startTime = time.time()
resultLeft = self.motionProxy.getPosition('LArm', 0, True)
resultLeft = np.reshape(resultLeft, (6, 1))
resultRight = self.motionProxy.getPosition('RArm', 0, True)
resultRight = np.reshape(resultRight, (6, 1))
dataLeft = np.hstack((dataLeft, resultLeft))
dataRight = np.hstack((dataRight, resultRight))
elapsed = time.time() - startTime + 0.005
timeArray = np.hstack(
(timeArray, np.array([[timeArray[0, -1] + elapsed]])))
timeArray = timeArray[:, :(timeArray.size - 1)]
self.stopperBool = True
return np.vstack((timeArray, dataLeft)), np.vstack(
(timeArray, dataRight))
self.motionProxy.setStiffnesses([
'LShoulderPitch', 'LShoulderRoll', 'LElbowYaw', 'LElbowRoll',
'LWristYaw', 'LHand', 'RShoulderPitch', 'RShoulderRoll',
'RElbowYaw', 'RElbowRoll', 'RWristYaw', 'RHand'
], 0.0)
DataLeft = []
DataRight = []
while True:
startDemo = raw_input(
'Do you want to start new demonstration? (Y/N): ')
if startDemo == 'Y':
left, right = recordData(self)
DataLeft.append(left)
DataRight.append(right)
elif startDemo == 'N':
print 'Data recording is finished.'
break
else:
print 'Incorrect input given.'
self.motionProxy.setStiffnesses([
'LShoulderPitch', 'LShoulderRoll', 'LElbowYaw', 'LElbowRoll',
'LWristYaw', 'LHand', 'RShoulderPitch', 'RShoulderRoll',
'RElbowYaw', 'RElbowRoll', 'RWristYaw', 'RHand'
], 1.0)
inputMat = DataLeft[0][0, :]
DataLeft = DTW(DataLeft)
DataRight = DTW(DataRight)
leftGMR = GMM_GMR(self.numberOfStates)
rightGMR = GMM_GMR(self.numberOfStates)
leftGMR.fit(DataLeft)
rightGMR.fit(DataRight)
leftGMR.predict(inputMat)
rightGMR.predict(inputMat)
fig = plt.figure()
ax1 = fig.add_subplot(231)
plt.title('Left Hand x-axis trajectory')
leftGMR.plot(ax=ax1, plotType="Data")
leftGMR.plot(ax=ax1, plotType="Regression")
ax2 = fig.add_subplot(232)
plt.title('Left Hand y-axis trajectory')
leftGMR.plot(yAxis=2, ax=ax2, plotType="Data")
leftGMR.plot(yAxis=2, ax=ax2, plotType="Regression")
ax3 = fig.add_subplot(233)
plt.title('Left Hand z-axis trajectory')
leftGMR.plot(yAxis=3, ax=ax3, plotType="Data")
leftGMR.plot(yAxis=3, ax=ax3, plotType="Regression")
ax4 = fig.add_subplot(234)
plt.title('Left Hand x-axis orientation')
leftGMR.plot(yAxis=4, ax=ax4, plotType="Data")
leftGMR.plot(yAxis=4, ax=ax4, plotType="Regression")
ax5 = fig.add_subplot(235)
plt.title('Left Hand y-axis orientation')
leftGMR.plot(yAxis=5, ax=ax5, plotType="Data")
leftGMR.plot(yAxis=5, ax=ax5, plotType="Regression")
ax6 = fig.add_subplot(236)
plt.title('Left Hand z-axis orientation')
leftGMR.plot(yAxis=6, ax=ax6, plotType="Data")
leftGMR.plot(yAxis=6, ax=ax6, plotType="Regression")
fig = plt.figure()
ax1 = fig.add_subplot(231)
plt.title('Right Hand x-axis trajectory')
rightGMR.plot(ax=ax1, plotType="Data")
rightGMR.plot(ax=ax1, plotType="Regression")
ax2 = fig.add_subplot(232)
plt.title('Right Hand y-axis trajectory')
rightGMR.plot(yAxis=2, ax=ax2, plotType="Data")
rightGMR.plot(yAxis=2, ax=ax2, plotType="Regression")
ax3 = fig.add_subplot(233)
plt.title('Right Hand z-axis trajectory')
rightGMR.plot(yAxis=3, ax=ax3, plotType="Data")
rightGMR.plot(yAxis=3, ax=ax3, plotType="Regression")
ax4 = fig.add_subplot(234)
plt.title('Right Hand x-axis orientation')
rightGMR.plot(yAxis=4, ax=ax4, plotType="Data")
rightGMR.plot(yAxis=4, ax=ax4, plotType="Regression")
ax5 = fig.add_subplot(235)
plt.title('Right Hand y-axis orientation')
rightGMR.plot(yAxis=5, ax=ax5, plotType="Data")
rightGMR.plot(yAxis=5, ax=ax5, plotType="Regression")
ax6 = fig.add_subplot(236)
plt.title('Right Hand z-axis orientation')
rightGMR.plot(yAxis=6, ax=ax6, plotType="Data")
rightGMR.plot(yAxis=6, ax=ax6, plotType="Regression")
plt.show()
self.trainedTasks.append(
(leftGMR.getPredictedMatrix(), rightGMR.getPredictedMatrix()))
import csv
from KNN import KNN
from sklearn import preprocessing
from DTW import DTW
from Harvesine import Harvesine
from sklearn.cross_validation import train_test_split
trainSet = pd.read_csv(
'../train_set.csv', # replace with the correct path
converters={"Trajectory": literal_eval},
index_col='tripId')
train,_ = train_test_split(trainSet,test_size=0.9)
#Initialize Encoder
le = preprocessing.LabelEncoder()
le.fit(train["journeyPatternId"])
y = le.transform(train["journeyPatternId"])
X = train['Trajectory']
knn = KNN(5,DTW(Harvesine))
scoring = ['accuracy']
classifier_pipeline = make_pipeline(knn)
scores = cross_validate(classifier_pipeline, X, y, cv=10,scoring=scoring)
print mean(scores['test_accuracy'])
def findDistance(self, x, y):
return DTW(x, y)