def seek(self, target_pos, on_torus=True):
if on_torus:
dv = utility.dist_vec_on_torus(target_pos, self.agent.pos,
game_state.WIDTH, game_state.HEIGHT)
desired_vel = self.agent.max_speed * utility.normalize(dv)
else:
desired_vel = self.agent.max_speedmax_speed * utility.normalize(
target_pos - self.agent.pos)
return desired_vel - self.agent.vel
def update(self, scene):
if self.owner.move_component is None:
return
vel = self.owner.move_component.vel
friction = np.array([vel[0], vel[1]])
if friction[0] != 0 or friction[1] != 0:
utility.normalize(friction)
friction_coefficient = 0.2
friction[
0] *= -friction_coefficient * self.owner.move_component.mass
friction[
1] *= -friction_coefficient * self.owner.move_component.mass
self.owner.move_component.apply_force(friction)
def encrypt(str_input):
str_input = normalize(str_input)
bacon_dictionary = {
"A": "AAAAA",
"B": "AAAAB",
"C": "AAABA",
"D": "AAABB",
"E": "AABAA",
"F": "AABAB",
"G": "AABBA",
"H": "AABBB",
"I": "ABAAA",
"J": "ABAAB",
"K": "ABABA",
"L": "ABABB",
"M": "ABBAA",
"N": "ABBAB",
"O": "ABBBA",
"P": "ABBBB",
"Q": "BAAAA",
"R": "BAAAB",
"S": "BAABA",
"T": "BAABB",
"U": "BABAA",
"V": "BABAB",
"W": "BABBA",
"X": "BABBB",
"Y": "BBAAA",
"Z": "BBAAB"
}
bring_home_the_bacon = ""
for eachletter in str_input:
bring_home_the_bacon += bacon_dictionary[eachletter]
return bring_home_the_bacon
def decrypt(str_input):
str_input = normalize(str_input)
baconlist = chunks_of_bacon(str_input)
bacon_dictionary = {
"AAAAA": "A",
"AAAAB": "B",
"AAABA": "C",
"AAABB": "D",
"AABAA": "E",
"AABAB": "F",
"AABBA": "G",
"AABBB": "H",
"ABAAA": "I",
"ABAAB": "J",
"ABABA": "K",
"ABABB": "L",
"ABBAA": "M",
"ABBAB": "N",
"ABBBA": "O",
"ABBBB": "P",
"BAAAA": "Q",
"BAAAB": "R",
"BAABA": "S",
"BAABB": "T",
"BABAA": "U",
"BABAB": "V",
"BABBA": "W",
"BABBB": "X",
"BBAAA": "Y",
"BBAAB": "Z"
}
bring_home_the_bacon = ""
for bacon in baconlist:
bring_home_the_bacon += bacon_dictionary[bacon]
return bring_home_the_bacon
def turn_around_time(self, to_target):
to_target = utility.normalize(to_target)
dot = np.dot(self.agent.heading, to_target)
coefficient = 0.5
return (dot - 1.0) * -coefficient
def __init__(self, B):
""" Algo specific parameters here.
"""
super(low_regret, self).__init__()
self.B = B
self.experts = [fixed_spread(b) for b in B]
self.Expert_distribution = utility.normalize(np.ones(len(B)))
def arrive(self, target_pos, deceleration=3, on_torus=True):
to_target = utility.dist_vec_on_torus(target_pos, self.agent.pos,
game_state.WIDTH,
game_state.HEIGHT)
dist = utility.magnitude(to_target)
if dist <= 0:
return np.array([0, 0])
if deceleration > 5:
deceleration = 5
elif deceleration < 1:
deceleration = 1
deceleration_tweaker = 0.5
speed = dist / float(deceleration) * deceleration_tweaker
if speed > self.agent.max_speed:
speed = self.agent.max_speed
if on_torus:
desired_vel = speed * utility.normalize(to_target)
else:
desired_vel = (speed / dist) * (target_pos - self.agent.pos)
return desired_vel - self.agent.vel
def encrypt(plaintext, n):
plaintext = normalize(plaintext)
count = cycle(range(0, n) + range(1, n - 1)[::-1])
rails = [''] * n
for i in plaintext:
index = count.next()
rails[index] = rails[index] + i
return ''.join(rails)
def encrypt(plaintext, n):
plaintext = normalize(plaintext)
count = cycle(range(0, n) + range(1, n - 1)[::-1])
rails = [""] * n
for i in plaintext:
index = count.next()
rails[index] = rails[index] + i
return "".join(rails)
def encrypt(key, str_input):
alphabet = string.ascii_uppercase
str_input = normalize(str_input)
returnlist = []
for i, c in enumerate(str_input):
for j, d in enumerate(alphabet):
if c == d:
returnlist.append(key[j])
return "".join(returnlist).upper()
def encrypt(key, str_input):
alphabet = string.ascii_uppercase
str_input = normalize(str_input)
returnlist = []
for i, c in enumerate(str_input):
for j, d in enumerate(alphabet):
if (c == d):
returnlist.append(key[j])
return ''.join(returnlist).upper()
def forward(self, inputs, inputs_lengths):
input_emb = self.item_embeddings(inputs)
pos_emb_input = torch.cat(inputs.size(0)*[torch.arange(start=0,end=inputs.size(1)).unsqueeze(0)])
pos_emb_input = pos_emb_input.long()
pos_emb = self.pos_embeddings(pos_emb_input)
x = input_emb+pos_emb
x = self.dropout(x)
mask = torch.ne(inputs, self.item_num).float().unsqueeze(-1)
x *= mask
for i in range (self.num_blocks):
x = self.multihead_attention(queries=normalize(x),keys=x)
x = self.feedforward(normalize(x))
x *= mask
x = normalize(x)
out = extract_axis_1_torch(x,inputs_lengths-1)
out = self.fc1(out)
return out
def encrypt(str_input):
str_input = normalize(str_input)
bacon_dictionary = {
"A":"AAAAA","B":"AAAAB","C":"AAABA","D":"AAABB",
"E":"AABAA","F":"AABAB","G":"AABBA","H":"AABBB",
"I":"ABAAA","J":"ABAAB","K":"ABABA","L":"ABABB",
"M":"ABBAA","N":"ABBAB","O":"ABBBA","P":"ABBBB",
"Q":"BAAAA","R":"BAAAB","S":"BAABA","T":"BAABB",
"U":"BABAA","V":"BABAB","W":"BABBA","X":"BABBB",
"Y":"BBAAA","Z":"BBAAB"}
bring_home_the_bacon = ""
for eachletter in str_input:
bring_home_the_bacon += bacon_dictionary[eachletter]
return bring_home_the_bacon
def dataLoaded(self):
# Note - this function only runs once the data directory has been loaded
self.setMouseTracking(True)
color = QColor(255, 255, 255)
self.samplesSelected.clear()
self.samplePoints.clear()
self.sampleAreaVisible.clear()
self.samplePointsInFile.clear()
self.penSelected.clear()
for t, p in common.SamplingPattern:
self.samplePoints.append((0, 0)) # these will need to be recomputed as photo scales
self.samplePointsInFile.append((0, 0)) # these only need to be computed once per photo
self.sampleAreaVisible.append([])
color.setHsv(t, int(utility.normalize(p, 0, 90) * 127 + 128), 255)
self.penSelected.append(QPen(color, 3, Qt.SolidLine))
def decrypt(str_input):
str_input = normalize(str_input)
baconlist = chunks_of_bacon(str_input)
bacon_dictionary = {
"AAAAA":"A","AAAAB":"B","AAABA":"C","AAABB":"D",
"AABAA":"E","AABAB":"F","AABBA":"G","AABBB":"H",
"ABAAA":"I","ABAAB":"J","ABABA":"K","ABABB":"L",
"ABBAA":"M","ABBAB":"N","ABBBA":"O","ABBBB":"P",
"BAAAA":"Q","BAAAB":"R","BAABA":"S","BAABB":"T",
"BABAA":"U","BABAB":"V","BABBA":"W","BABBB":"X",
"BBAAA":"Y","BBAAB":"Z"}
bring_home_the_bacon = ""
for bacon in baconlist:
bring_home_the_bacon += bacon_dictionary[bacon]
return bring_home_the_bacon
def __init__(self, body_i, ai_bar, bi_bar, body_j, aj_bar, bj_bar, k,
theta_0, c, h):
"""
ai_bar: axis about which the torque is applied to body i (unit vector)
bi_bar: perpendicular to ai_bar and used with bj_bar to define rotation
angle theta_ij
aj_bar: axis about which the torque is applied to body j (unit vector)
bj_bar: perpendicular to aj_bar and used with bi_bar to define rotation
angle theta_ij
k: spring constant of actuator
theta_0: zero_tension angle
c: damping coefficient
h: function that describes the effects of an actuator (hydraulic, electric, etc.)
must be of the form h(theta, theta_dot, t)
t: time
"""
self.body_i = body_i
self.body_j = body_j
self.ai_bar = normalize(column(ai_bar))
self.bi_bar = normalize(column(bi_bar))
self.aj_bar = normalize(column(aj_bar))
self.bj_bar = normalize(column(bj_bar))
self.k = k
self.theta_0 = theta_0
self.c = c
self.h = h
self.theta_old = 0
self.n = 0 #number of full revolutions
def get_normals(points):
normals = []
length = len(points)
for i in range(0, length - 1):
segment = [
points[i + 1][0] - points[i][0], points[i + 1][1] - points[i][1]
]
normal = [-segment[1], segment[0]] # left normal
normal = utility.normalize(normal)
normals.append(normal)
segment = [
points[0][0] - points[length - 1][0],
points[0][1] - points[length - 1][1]
]
normal = [-segment[1], segment[0]] # left normal
normal = utility.normalize(normal)
normals.append(normal)
return normals
def flee(self, target_pos, on_torus=True):
panic_dis = 100.0 * 100.0
if on_torus:
dv = utility.dist_vec_on_torus(self.agent.pos, target_pos,
game_state.WIDTH, game_state.HEIGHT)
distance = utility.magnitude_square(dv)
else:
dv = self.agent.pos - target_pos
distance = utility.magnitude_square(self.agent.pos - target_pos)
if distance > panic_dis:
return np.array([0.0, 0.0])
desired_vel = self.agent.max_speed * utility.normalize(dv)
return desired_vel - self.agent.vel
def circle_circle_collision(a, b, scene):
a_pos = scene.transformed_entity_pts[a.id][0]
b_pos = scene.transformed_entity_pts[b.id][0]
r = a.render_component.bounding_radius(
) + b.render_component.bounding_radius()
dis = math.sqrt((a_pos[0] - b_pos[0]) * (a_pos[0] - b_pos[0]) +
(a_pos[1] - b_pos[1]) * (a_pos[1] - b_pos[1]))
if r < dis:
return False
dv = [b_pos[0] - a_pos[0], b_pos[1] - a_pos[1]]
dv = utility.normalize(dv) * (r - dis)
return dv
def translate_title(name: str) -> str:
name = utility.normalize(name)
name = utility.fullwidth_to_halfwidth(name)
name = utility.escape_markdown_symbol(name)
# remove 【...】
name = re.sub(r'【[^【】]*(電子|特典|OFF)[^【】]*】', '', name)
name = re.sub(r'【(期間限定|)([^【】]+セット)】', ' \g<2>', name)
name = name.strip()
# '(N)' -> ' N'
name = re.sub(r'\(([0-9]+)\)$', r' \g<1>', name)
# ': N' -> ' N'
name = re.sub(r': ([0-9]+)$', r' \g<1>', name)
# 'N巻' -> 'N'
name = re.sub(r'([0-9]+)巻$', r'\g<1>', name)
# replace continuous space
name = re.sub(r'\s+', ' ', name)
# coin
coin_match = re.match(r'BOOK☆WALKER 期間限定コイン (?P<coin>[0-9,]+)円分', name)
if coin_match:
name = '期間限定コイン {0}円分'.format(
coin_match.group('coin').replace(',', ''))
return name
def wander(self):
self.wander_target += np.array([
random.uniform(-1, 1) * self.wander_jitter,
random.uniform(-1, 1) * self.wander_jitter
])
self.wander_target = self.wander_radius * utility.normalize(
self.wander_target)
target_local = self.wander_target + np.array([self.wander_distance, 0])
utility.rotate_point(target_local, self.agent.side, self.agent.up)
target_world = self.agent.pos + target_local
"""
circle_pos = np.array([self.wander_distance, 0])
utility.rotate_point(circle_pos, self.agent.side, self.agent.heading)
circle_pos += self.agent.pos
pygame.draw.circle(engine.surface, color.RED, [int(circle_pos[0]), int(circle_pos[1])], self.wander_radius, 1)
pygame.draw.circle(engine.surface, color.RED, [int(target_world[0]), int(target_world[1])], 3)
"""
return target_world - self.agent.pos
def update(self, time_elapsed):
if self.owner.move_component is None:
return
speed = self.owner.move_component.max_speed
heading = self.owner.up
right = -self.owner.side
move_force = np.array([0.0, 0.0])
side = np.array([
self.owner.side[0] * self.max_turn_rate * self.turn,
self.owner.side[1] * self.max_turn_rate * self.turn
])
if self.move_left_right != 0.0:
move_force[0] -= right[0] * self.move_left_right * speed
move_force[1] -= right[1] * self.move_left_right * speed
self.move_left_right = 0.0
if self.move_backward_forward != 0.0:
move_force[0] += heading[0] * self.move_backward_forward * speed
move_force[1] += heading[1] * self.move_backward_forward * speed
self.move_backward_forward = 0.0
if self.turn != 0.0:
self.owner.up[0] += side[0]
self.owner.up[1] += side[1]
self.owner.up = utility.normalize(self.owner.up)
self.owner.side = utility.perpendicular(self.owner.up)
self.turn = 0.0
length = pyrr.vector3.length(move_force)
if length > 0.1:
self.owner.move_component.apply_force(move_force)
locations1, features1 = sift.siftFeature(img1)
locations2, features2 = sift.siftFeature(img2)
# use PCAac to reduce dimensions
numberOfFeaturesOne = locations1.shape[0]
numberOfFeaturesTwo = locations2.shape[0]
features = np.vstack((features1, features2))
V, S, mean = pca.pca(features)
pcaFeatures1 = pca.project(features1, V, 36)
pcaFeatures2 = pca.project(features2, V, 36)
# normalize features
util.normalize(pcaFeatures1)
util.normalize(pcaFeatures2)
np.savetxt("pcafeature1", pcaFeatures1, delimiter="\t")
np.savetxt("pcafeature2", pcaFeatures2, delimiter="\t")
# interface with Java program to do matching
rowX = pcaFeatures1.shape[0]
rowY = pcaFeatures2.shape[0]
column = pcaFeatures1.shape[1]
matchCommand = "java match " +str(rowX)+ " " +str(rowY)+ " " +str(column)
print matchCommand
os.system(matchCommand)
# plot according to match stored in "data" folder
def update_expert_distribution(self):
k = 0.0001
inventory = np.array([e.inventory for e in self.experts])
Loss = np.exp(-k * np.abs(inventory))
self.Expert_distribution = utility.normalize(Loss)
#
# cktboard_texture = "input\\cktboard.raw"
# img3 = MyImage("cktboard_texture", [365, 120], cktboard_texture)
# dimensions = [256, 256]
# img3.matrix = co_Occurrence(img3, dimensions)
# my_io.customised_matrix_write8(img3, dimensions)
flowers = "input\\flowers.raw"
img1 = MyImage("flowers", [400, 300], flowers)
flowers_template = "input\\flowers-template.raw"
img2 = MyImage("flowers_template", [42, 45], flowers_template)
img3 = MyImage("flowers_correlation_coefficient", [358, 255])
img3.matrix = CorrelationCoefficient.CC(img1, img2)
# my_io.write(img3, "int8")
img3.matrix = utility.normalize(img3.matrix, [358, 255])
# my_io.write(img3, "int8")
my_io.customised_matrix_write8(img3, [358, 255])
# lena = "input\\lena.raw"
# img3 = MyImage("lena_resampling", [512, 512], lena)
# img3.matrix = DU.down(img3)
# img3.matrix = DU.up(img3)
# my_io.write(img3, "int8")
# lena = "input\\lena_resampling2.raw"
# img3 = MyImage("lena_resampling", [256, 256], lena)
# # img3.matrix = DU.down(img3)
# img3.matrix = DU.up(img3)
# # my_io.write(img3, "int8")
# my_io.customised_matrix_write8(img3, [512, 512])
def DataGenBB(self, DataStrs, train_start, train_end):
generateFunc = [
"original", "scale", "rotate", "translate", "scaleAndTranslate",
"brightnessAndContrast"
]
# generateFunc = ["original", "scale", "rotate", "translate", "scaleAndTranslate"]
InputData = np.zeros(
[self.batch_size * len(generateFunc), self.imSize, self.imSize, 3],
dtype=np.float32)
# InputLabel = np.zeros([self.batch_size * len(generateFunc), 7], dtype = np.float32)
InputLabel = np.zeros([self.batch_size * len(generateFunc), 3],
dtype=np.float32)
InputNames = []
count = 0
for i in range(train_start, train_end):
strLine = DataStrs[i]
strCells = strLine.rstrip(' \n').split(' ')
imgName = strCells[0]
labels = np.array(strCells[1:]).astype(np.float)
if len(labels) == 78:
# print "switch to menpo39"
labelsPTS = labels[:136].reshape([39, 2])
self.ifMenpo39Data = True
else:
# print "not menpo39"
labelsPTS = labels[:136].reshape([68, 2])
self.ifMenpo39Data = False
# if self.debug:
# print "imgName: ", imgName
img = cv2.imread(imgName)
if img != None:
# print "find image: ", imgName
# print "img.shape: ", img.shape
img = cv2.resize(img, (self.imSize, self.imSize))
# print "img.shape: ", img.shape
if self.ifMenpo39Data:
x, y = self.unpackLandmarks(labelsPTS)
else:
x, y = ut.unpackLandmarks(labelsPTS, self.imSize)
# newImg, newX, newY = img, x, y
for index in range(len(generateFunc)):
method = generateFunc[index]
(w, h, _) = img.shape
# tag = random.choice(generateFunc)
if method == "resize":
newImg, newX, newY = ut.resize(img,
x,
y,
xMaxBound=w,
yMaxBound=h,
random=True)
elif method == "rotate":
newImg, newX, newY = ut.rotate(img, x, y, w=w, h=h)
elif method == "mirror":
newImg, newX, newY = ut.mirror(img, x, y, w=w, h=h)
elif method == "translate" or method == "scaleAndTranslate":
newImg, newX, newY = ut.translate(img, x, y, w=w, h=h)
elif method == "brightnessAndContrast":
newImg, newX, newY = ut.contrastBrightess(img, x, y)
elif method == "original":
newImg, newX, newY = img, x, y
elif method == "scale":
newImg, newX, newY = img, x, y
else:
raise "not existing function"
# if self.debug:
# plotOriginal = ut.plotLandmarks(img, x, y, self.imSize, ifReturn = True)
# plotNew = ut.plotLandmarks(newImg, newX, newY, self.imSize, ifReturn = True)
# cv2.imwrite(self.outputDir + 'testOriginal' + str(count) + '.jpg', img)
# cv2.imwrite(self.outputDir + 'testNew' + str(count) + '.jpg', newImg)
# cv2.imwrite(self.outputDir + 'plotOriginal' + str(count) + '.jpg', plotOriginal)
# cv2.imwrite(self.outputDir + 'plotNew' + str(count) + '.jpg', plotNew)
# print "before normalize: ", newX
# normX = ut.normalize(newX)
# normY = ut.normalize(newY)
# print "after normalize: ", newX
# print "after denormalize again: ", ut.deNormalize(newX)
# normXMin = min(normX)
# normYMin = min(normY)
# normXMax = max(normX)
# normYMax = max(normY)
# normXMean = (normXMax + normXMin)/2.0
# normYMean = (normYMax + normYMin)/2.0
# normEdge = max(normYMax - normYMin, normXMax - normXMin)
newXMin = min(newX)
newYMin = min(newY)
newXMax = max(newX)
newYMax = max(newY)
newXMean = (newXMax + newXMin) / 2.0
newYMean = (newYMax + newYMin) / 2.0
edge = max(newYMax - newYMin, newXMax - newXMin)
# if method == "scale":
# cv2.imshow("originalImg", newImg)
# cv2.waitKey(0)
if method == "scale" or method == "scaleAndTranslate":
newEdge = np.random.uniform(0.7, 0.9) * edge
newXMin = int(newXMean - newEdge / 2.0)
newXMax = int(newXMean + newEdge / 2.0)
newYMin = int(newYMean - newEdge / 2.0)
newYMax = int(newYMean + newEdge / 2.0)
newXMean = newXMean - newXMin
newYMean = newYMean - newYMin
# print "newXMin, newYMin, newXMax, newYMax: ", newXMin, newYMin, newXMax, newYMax
newImg = Image.fromarray(newImg.astype(np.uint8))
cropImg = newImg.crop(
(newXMin, newYMin, newXMax, newYMax))
newImg = np.array(cropImg)
# cv2.imshow("processing", newImg)
# cv2.waitKey(0)
w, h, _ = newImg.shape
edge = edge * self.imSize / w
newXMean = newXMean * self.imSize / w
newYMean = newYMean * self.imSize / h
newImg = cv2.resize(newImg, (self.imSize, self.imSize))
# print "newXMin: ", newXMin
# print "newYMin: ", newYMin
# print "newXMax: ", newXMax
# print "newYMax: ", newYMax
# print "newXMean: ", newXMean
# print "newYMean: ", newYMean
# print "newEdge: ", newEdge
# if method == "scale":
# newImg = ut.plotTarget(newImg, [newXMean, newYMean, edge], ifSquareOnly = True, ifGreen = True)
# cv2.imshow("newImg", newImg)
# cv2.waitKey(0)
# if self.ifMenpo39Data == False:
print "imgName: ", imgName.split("/")[-2] + imgName.split(
"/")[-1].split(".")[0]
print "inputCheck/" + imgName.split(
"/")[-2] + imgName.split("/")[-1].split(".")[0] + str(
method) + str(count) + '.jpg'
cv2.imwrite(
"inputCheck/" + imgName.split("/")[-2] +
imgName.split("/")[-1].split(".")[0] + str(method) +
str(count) + '.jpg', newImg)
normX = ut.normalize(newX, self.imSize)
normY = ut.normalize(newY, self.imSize)
# normPTS = np.asarray(ut.packLandmarks(normX, normY))
normXMean, normYMean, normEdge = ut.normalize(
newXMean, self.imSize), ut.normalize(
newYMean,
self.imSize), ut.normalize(edge, self.imSize)
# print "newPTS: ", newPTS.shape
# print "ut.deNormalize(normXMin): ", ut.deNormalize(normXMin)
# print "ut.deNormalize(normYMin): ", ut.deNormalize(normYMin)
# print "ut.deNormalize(normXMax): ", ut.deNormalize(normXMax)
# print "ut.deNormalize(normYMax): ", ut.deNormalize(normYMax)
# print "ut.deNormalize(normXMean): ",ut.deNormalize(normXMean)
# print "ut.deNormalize(normYMean): ",ut.deNormalize(normYMean)
# print "ut.deNormalize(normEdge): ", ut.deNormalize(normEdge)
# print "method: ", method
# print "newImg.shape: ", newImg.shape
# print "len(InputData): ", len(InputData)
InputData[count, ...] = newImg
# labels = np.array([normPTS[27][0], normPTS[27][1], normPTS[8][0],
# normPTS[8][1], normXMean, normYMean, normEdge])
labels = np.array([normXMean, normYMean, normEdge])
InputLabel[count, ...] = labels
InputNames.append(imgName)
# print "count: ", count
count += 1
else:
print "cannot : ", imgName
return InputData, InputLabel, np.asarray(InputNames)
def model_xor_snn():
hrs_set_voltage = -4
positive_bias = 1
fixing_pulses = 37 * 10 ** 4
current_application_time = 500
spike = 1
true = 20
false = 0
input_spikes_1 = [false, false, true, true]
input_spikes_2 = [false, true, false, true]
targets = [false, true, true, false]
# Create a network with 2 input neurons, 2 hidden layer neurons
# and 1 output neuron. The input neurons are simulated by their spikes.
hidden_neuron_1 = spiking_neuron.SpikingNeuron()
hidden_neuron_2 = spiking_neuron.SpikingNeuron()
output_neuron = spiking_neuron.SpikingNeuron()
# Create 6 synapses with input_synapse_ij representing the connection
# between the input neuron i and the hidden neuron j, and hidden_synapse_k
# representing the connection between the hidden neuron k and the output neuron.
# The inhibitory synapses are input_synapse_12 and input_synapse_21.
input_synapse_11 = memristor.Memristor()
input_synapse_12 = memristor.Memristor()
input_synapse_21 = memristor.Memristor()
input_synapse_22 = memristor.Memristor()
hidden_synapse_1 = memristor.Memristor()
hidden_synapse_2 = memristor.Memristor()
# Set all the synapses to a high resistance state.
input_synapse_11.set_resistance(hrs_set_voltage)
input_synapse_12.set_resistance(hrs_set_voltage)
input_synapse_21.set_resistance(hrs_set_voltage)
input_synapse_22.set_resistance(hrs_set_voltage)
hidden_synapse_1.set_resistance(hrs_set_voltage)
hidden_synapse_2.set_resistance(hrs_set_voltage)
# Fix the resistance values in the synapses by applying consecutive
# positive bias voltage pulses.
for _ in range(fixing_pulses):
input_synapse_11.apply_voltage(positive_bias)
input_synapse_12.apply_voltage(positive_bias)
input_synapse_21.apply_voltage(positive_bias)
input_synapse_22.apply_voltage(positive_bias)
hidden_synapse_1.apply_voltage(positive_bias)
hidden_synapse_2.apply_voltage(positive_bias)
c_11 = []
c_12 = []
c_21 = []
c_22 = []
hidden_spikes_1 = []
hidden_spikes_2 = []
z_1 = []
z_2 = []
output_spikes = []
for spikes_1, spikes_2 in zip(input_spikes_1, input_spikes_2):
# The normalized current c_ij passes through the input_synapse_ij.
# The current in the inhibitory synapses is inverted. The input voltage
# is 1 V if the input neuron fired as many pulses as the encoding of TRUE
# or more, else it is 0 V.
c_11.append(utility.normalize(utility.burst(spikes_1, true) / input_synapse_11.read_resistance()))
c_12.append(utility.invert(utility.normalize(utility.burst(spikes_1, true) / input_synapse_12.read_resistance())))
c_21.append(utility.invert(utility.normalize(utility.burst(spikes_2, true) / input_synapse_21.read_resistance())))
c_22.append(utility.normalize(utility.burst(spikes_2, true) / input_synapse_22.read_resistance()))
hidden_spikes_1.append(hidden_neuron_1.apply_current(c_11[-1] + c_21[-1],
current_application_time)[0].count(spike))
hidden_spikes_2.append(hidden_neuron_2.apply_current(c_12[-1] + c_22[-1],
current_application_time)[0].count(spike))
# The normalized curent z_i passes through the hidden_synapse_i. The
# input voltage is 1 V if the hidden neuron fired as many pulses as
# the encoding of TRUE or more, else it is 0 V.
z_1.append(utility.normalize(utility.burst(hidden_spikes_1[-1], true) / hidden_synapse_1.read_resistance()))
z_2.append(utility.normalize(utility.burst(hidden_spikes_2[-1], true) / hidden_synapse_2.read_resistance()))
output_spikes.append(output_neuron.apply_current(z_1[-1] + z_2[-1],
current_application_time)[0].count(spike))
# The neurons are resting until the next input.
hidden_neuron_1.rest()
hidden_neuron_2.rest()
output_neuron.rest()
# Store the data into a data frame.
data = pd.DataFrame()
data["input_spikes_1"] = input_spikes_1
data["input_spikes_2"] = input_spikes_2
data["c_11"] = c_11
data["c_12"] = c_12
data["c_21"] = c_21
data["c_22"] = c_22
data["hidden_spikes_1"] = hidden_spikes_1
data["hidden_spikes_2"] = hidden_spikes_2
data["z_1"] = z_1
data["z_2"] = z_2
data["output_spikes"] = output_spikes
data["target"] = targets
data["current_application_time"] = [current_application_time] * 4
print(data)
def solve_xor_complex_noisy_snn(learning_pulses, epochs, output_to_csv=False):
hrs_set_voltage = -4
current_application_time = 500
spike = 1
training_voltage = 1
true = 20
false = 0
input_spikes_1 = [false, false, true, true]
input_spikes_2 = [false, true, false, true]
targets = [false, true, true, false]
# Create a network with 2 input neurons, 2 hidden layer neurons
# and 1 output neuron. The input neurons are simulated by their spikes.
hidden_neuron_1 = spiking_neuron.SpikingNeuron()
hidden_neuron_2 = spiking_neuron.SpikingNeuron()
output_neuron = spiking_neuron.SpikingNeuron()
# Create 6 synapses with pos_input_synapse_ij representing the excitatory
# connection between the input neuron i and the hidden neuron j, and
# pos_hidden_synapse_k representing the excitatory connection between
# the hidden neuron k and the output neuron.
pos_input_synapse_11 = memristor.Memristor()
pos_input_synapse_12 = memristor.Memristor()
pos_input_synapse_21 = memristor.Memristor()
pos_input_synapse_22 = memristor.Memristor()
pos_hidden_synapse_1 = memristor.Memristor()
pos_hidden_synapse_2 = memristor.Memristor()
# Create 6 synapses with neg_input_synapse_ij representing the inhibitory
# connection between the input neuron i and the hidden neuron j, and
# neg_hidden_synapse_k representing the inhibitory connection between
# the hidden neuron k and the output neuron.
neg_input_synapse_11 = memristor.Memristor()
neg_input_synapse_12 = memristor.Memristor()
neg_input_synapse_21 = memristor.Memristor()
neg_input_synapse_22 = memristor.Memristor()
neg_hidden_synapse_1 = memristor.Memristor()
neg_hidden_synapse_2 = memristor.Memristor()
# Set all the synapses to a high resistance state.
pos_input_synapse_11.set_resistance(hrs_set_voltage)
pos_input_synapse_12.set_resistance(hrs_set_voltage)
pos_input_synapse_21.set_resistance(hrs_set_voltage)
pos_input_synapse_22.set_resistance(hrs_set_voltage)
pos_hidden_synapse_1.set_resistance(hrs_set_voltage)
pos_hidden_synapse_2.set_resistance(hrs_set_voltage)
neg_input_synapse_11.set_resistance(hrs_set_voltage)
neg_input_synapse_12.set_resistance(hrs_set_voltage)
neg_input_synapse_21.set_resistance(hrs_set_voltage)
neg_input_synapse_22.set_resistance(hrs_set_voltage)
neg_hidden_synapse_1.set_resistance(hrs_set_voltage)
neg_hidden_synapse_2.set_resistance(hrs_set_voltage)
c_11 = []
c_12 = []
c_21 = []
c_22 = []
hidden_spikes_1 = []
hidden_spikes_2 = []
z_1 = []
z_2 = []
output_spikes = []
errors = []
squared_errors = []
epoch_numbers = []
for epoch_number in range(1, epochs + 1):
for spikes_1, spikes_2, target in zip(input_spikes_1, input_spikes_2, targets):
epoch_numbers.append(epoch_number)
# The normalized noisy current c_ij passes through the input_synapse_ij. The input
# voltage is 1 V if the input neuron fired as many pulses as the encoding of
# TRUE or more, else it is 0 V.
pos_c_11 = utility.add_noise(utility.normalize(utility.burst(spikes_1, true)
/ pos_input_synapse_11.read_resistance()))
pos_c_12 = utility.add_noise(utility.normalize(utility.burst(spikes_1, true)
/ pos_input_synapse_12.read_resistance()))
pos_c_21 = utility.add_noise(utility.normalize(utility.burst(spikes_2, true)
/ pos_input_synapse_21.read_resistance()))
pos_c_22 = utility.add_noise(utility.normalize(utility.burst(spikes_2, true)
/ pos_input_synapse_22.read_resistance()))
# The normalized noisy current neg_c_ij passes through the neg_input_synapse_ij.
# The current is inverted to represent the inhibitory connection. The input
# voltage is 1 V if the input neuron fired as many pulses as the encoding
# of TRUE or more, else it is 0 V.
neg_c_11 = utility.invert(utility.add_noise(utility.normalize(utility.burst(spikes_1, true)
/ neg_input_synapse_11.read_resistance())))
neg_c_12 = utility.invert(utility.add_noise(utility.normalize(utility.burst(spikes_1, true)
/ neg_input_synapse_12.read_resistance())))
neg_c_21 = utility.invert(utility.add_noise(utility.normalize(utility.burst(spikes_2, true)
/ neg_input_synapse_21.read_resistance())))
neg_c_22 = utility.invert(utility.add_noise(utility.normalize(utility.burst(spikes_2, true)
/ neg_input_synapse_22.read_resistance())))
c_11.append(pos_c_11 + neg_c_11)
c_12.append(pos_c_12 + neg_c_12)
c_21.append(pos_c_21 + neg_c_21)
c_22.append(pos_c_22 + neg_c_22)
hidden_spikes_1.append(hidden_neuron_1.apply_current(c_11[-1] + c_21[-1],
current_application_time)[0].count(spike))
hidden_spikes_2.append(hidden_neuron_2.apply_current(c_12[-1] + c_22[-1],
current_application_time)[0].count(spike))
# The normalized noisy curent pos_z_i passes through the pos_hidden_synapse_i.
# The input voltage is 1 V if the hidden neuron fired as many pulses as
# the encoding of TRUE or more, else it is 0 V.
pos_z_1 = utility.add_noise(utility.normalize(utility.burst(hidden_spikes_1[-1], true)
/ pos_hidden_synapse_1.read_resistance()))
pos_z_2 = utility.add_noise(utility.normalize(utility.burst(hidden_spikes_2[-1], true)
/ pos_hidden_synapse_2.read_resistance()))
# The normalized noisy curent neg_z_i passes through the neg_hidden_synapse_i.
# The current is inverted to represent the inhibitory connection. The
# input voltage is 1 V if the hidden neuron fired as many pulses as the
# encoding of TRUE or more, else it is 0 V.
neg_z_1 = utility.invert(utility.add_noise(utility.normalize(utility.burst(hidden_spikes_1[-1], true)
/ neg_hidden_synapse_1.read_resistance())))
neg_z_2 = utility.invert(utility.add_noise(utility.normalize(utility.burst(hidden_spikes_2[-1], true)
/ neg_hidden_synapse_2.read_resistance())))
z_1.append(pos_z_1 + neg_z_1)
z_2.append(pos_z_2 + neg_z_2)
output_spikes.append(output_neuron.apply_current(z_1[-1] + z_2[-1],
current_application_time)[0].count(spike))
errors.append(target - output_spikes[-1])
squared_errors.append(errors[-1] ** 2)
# Update the synapses based on the error.
# If the output neuron should have fired more times and it did not,
# then strengthen the excitatory hidden synapses of the hidden neurons
# that fired as many pulses as the encoding of TRUE or more. Otherwise,
# do the equivalent update for the input synapses of the input neurons
# that fired as many pulses as the encoding of TRUE or more.
if utility.is_false_negative(errors[-1]):
if utility.burst(hidden_spikes_1[-1], true) or utility.burst(hidden_spikes_2[-1], true):
if utility.burst(hidden_spikes_1[-1], true):
for _ in range(learning_pulses):
pos_hidden_synapse_1.apply_voltage(training_voltage)
if utility.burst(hidden_spikes_2[-1], true):
for _ in range(learning_pulses):
pos_hidden_synapse_2.apply_voltage(training_voltage)
else:
if utility.burst(spikes_1, true):
for _ in range(learning_pulses):
pos_input_synapse_11.apply_voltage(training_voltage)
pos_input_synapse_12.apply_voltage(training_voltage)
if utility.burst(spikes_2, true):
for _ in range(learning_pulses):
pos_input_synapse_21.apply_voltage(training_voltage)
pos_input_synapse_22.apply_voltage(training_voltage)
# If the output neuron should not have fired and it did, then strengthen
# the inhibitory hidden synapses of the hidden neurons that fired as many
# pulses as the encoding of TRUE or more.
elif utility.is_false_positive(errors[-1]):
if utility.burst(hidden_spikes_1[-1], true):
for _ in range(learning_pulses):
neg_hidden_synapse_1.apply_voltage(training_voltage)
if utility.burst(hidden_spikes_2[-1], true):
for _ in range(learning_pulses):
neg_hidden_synapse_2.apply_voltage(training_voltage)
# The neurons are resting until the next input.
hidden_neuron_1.rest()
hidden_neuron_2.rest()
output_neuron.rest()
# Store the data into a data frame.
data = pd.DataFrame()
data["input_spikes_1"] = input_spikes_1 * epochs
data["input_spikes_2"] = input_spikes_2 * epochs
data["c_11"] = c_11
data["c_12"] = c_12
data["c_21"] = c_21
data["c_22"] = c_22
data["hidden_spikes_1"] = hidden_spikes_1
data["hidden_spikes_2"] = hidden_spikes_2
data["z_1"] = z_1
data["z_2"] = z_2
data["output_spikes"] = output_spikes
data["target"] = targets * epochs
data["error"] = errors
data["squared_error"] = squared_errors
data["epoch"] = epoch_numbers
data["learning_pulses"] = [learning_pulses] * epochs * 4
data["training_voltage"] = [training_voltage] * epochs * 4
data["current_application_time"] = [current_application_time] * epochs * 4
if output_to_csv:
utility.save_data(data, "./output", "solve-xor-complex-noisy-snn")
# Plot the MSE as a function of epoch.
utility.plot_mse(data)
# 构建N个带有Pi个图片的predictor的数组
i = 0
while i < N:
j = 0
path1 = sys.path[0] + "\\orl_faces\\s" + str(i + 1) + "\\" # 第一层路径循环文件夹
ims = [] # 初始化临时存储图片变量
# 读取一组图片到list中
while j < Pi:
# 读取一张图片
path = path1 + str(j + 1) + ".pgm" # 第二层路径循环图片文件
im = Image.open(path)
im = im.resize(downsampling_size, sampling_type) # 重采样
# 将该图片插入到该类数组中
ims.append(list(im.getdata()))
# 标准化
normalize(ims[j])
j += 1
# 将ims转换为predictor并插入到数组中
p = np.matrix(ims[0]) # 先插入头一个图片
j = 1 # 此时j已经有一个
# 将剩下的插入临时predictor
while j < Pi:
p = np.vstack((p, ims[j]))
j += 1
# 将predictor转置之后插入list
predictor.append(p.T)
i = i + 1
# 创建一个N*Pj大小的存储隶属类别的数组
inner = []
result = []
datafiles = ['dataspam']
k = [1,3,5,7,9]
metrics = [sim_cosine, sim_pearson, euclidean]
r_method = ['uniform', 'distance']
n = [True, False]
#percentage = [10, 20, 30]
#k = [5]
#percentage = [10]
prod = product(datafiles, k, metrics, r_method, n)
for filename, k, metric, r_m, n in prod:
if filename == 'datahouse':
X, y = getXy(filename)
else:
X, y = getXy(filename, delimiter=',')
if n:
normalize(X)
print 'data is normalized'
else:
print 'data is not normalized'
reg = KNeighborClassifier(X, y, k, metric, r_method=r_m)
e = Evaluater(reg, test_percentage = 30)
print '\t', e.evaluate()
print '\n'
def __init__(self, df, model=Models.PROPHET,
upsample_freq=None,
train_test_split_ratio=Constants.TRAIN_TEST_SPLIT_RATIO.value,
epochs=Constants.EPOCHS.value,
initial_epoch=Constants.INITIAL_EPOCH.value,
batch_size=Constants.BATCH_SIZE.value,
sliding_window_size_or_time_steps=Constants.SLIDING_WINDOW_SIZE_OR_TIME_STEPS.value,
do_shuffle=True):
logging.info("resample: {}. future_prediction: {}, epochs: {}, batch_size: {},"
" window_size: {}, eurons: {}"
.format(Constants.RESAMPLING_FREQ.value
, Constants.SHIFT_IN_TIME_STEP_TO_PREDICT.value
, epochs
, batch_size
, sliding_window_size_or_time_steps
, Constants.NEURONS.value
))
if logging.getLogger().isEnabledFor(logging.INFO):
explore_data(df)
# first step is to create a timestamp column as index to turn it to a TimeSeries data
df.index = pd.to_datetime(df[ColumnNames.DATE.value] + df[ColumnNames.TIME.value],
format='%Y-%m-%d%H:%M:%S', errors='raise')
if 'Unnamed: 0' in df.columns:
df.drop('Unnamed: 0', axis=1, inplace=True)
# keep a copy of original dataset for future comparison
self.df_original = df.copy()
# we interpolate temperature using prophet to use it in a multivariate forecast
temperature = ColumnNames.TEMPERATURE.value
interpolated_df = facebook_prophet_filter(df, temperature,
Constants.FORECASTED_TEMPERATURE_FILE.value)
interpolated_df.index = df.index
df[[temperature]] = interpolated_df[[ColumnNames.FORECAST.value]]
# lets also interpolate missing kwh using facebook prophet (or we could simply drop them)
# now turn to kwh and make the format compatible with prophet
power = ColumnNames.POWER.value
interpolated_df = facebook_prophet_filter(df, power,
Constants.FORECASTED_POWER_FILE.value)
interpolated_df.index = df.index
df[[power]] = interpolated_df[[ColumnNames.FORECAST.value]]
df = df.rename(columns={power: ColumnNames.LABEL.value})
df.drop(columns=[ColumnNames.DATE.value,
ColumnNames.TIME.value,
ColumnNames.DAY_OF_WEEK.value,
ColumnNames.MONTH.value],
inplace=True
)
if upsample_freq is not None:
df = df.resample(upsample_freq).mean()
# for any regression or forecasting it is better to work with normalized data
self.transformer = QuantileTransformer() # handle outliers better than MinMaxScalar
features = ColumnNames.FEATURES.value
normalized = normalize(df, features, transformer=self.transformer)
# we use the last part (after 12/1/2013) that doesnt have temperature for testing
cutoff_date = Constants.CUTOFF_DATE.value
self.df = normalized[normalized.index < cutoff_date]
self.testing = normalized[normalized.index >= cutoff_date]
self.df[ColumnNames.DATE_STAMP.value] = self.df.index
self.df_blocked = None
self.train_test_split_ratio = train_test_split_ratio
self.model_type = model
self.train_X, self.test_X, self.train_test_split_index = self.train_test_split(self.df[features])
self.train_y, self.test_y, _ = self.train_test_split(self.df[ColumnNames.LABELS.value])
self.model_fit = None
self.epochs = epochs
self.initial_epoch = initial_epoch
self.batch_size = batch_size
self.history = None
# following is defines in sliding_window
self.do_shuffle = do_shuffle
self.val_idx = None
self.shuffled_X = None
self.shuffled_y = None
self.train = None
self.label = None
self.train_size = None
self.val_size = None
if logging.getLogger().isEnabledFor(logging.INFO):
explore_data(self.df)
def identifyKeyFrame(self, SIFTFeatures, indices, threshold = 0.15):
if len(indices) in [1,2]:
lst = []
lst.append(indices[0])
return lst
# build up graph structure
numberOfNodes = len(SIFTFeatures)
graph = Graph(numberOfNodes)
for i in range(numberOfNodes):
for j in range(i+1, numberOfNodes, 1):
one = SIFTFeatures[i]
two = SIFTFeatures[j]
# check whether one and two are near duplicate
pcaFeatures1 = pca.project(one, self.V, 36)
pcaFeatures2 = pca.project(two, self.V, 36)
# normalize features
util.normalize(pcaFeatures1)
util.normalize(pcaFeatures2)
np.savetxt("pcafeature1", pcaFeatures1, delimiter="\t")
np.savetxt("pcafeature2", pcaFeatures2, delimiter="\t")
# interface with Java program to do matching
rowX = pcaFeatures1.shape[0]
rowY = pcaFeatures2.shape[0]
column = pcaFeatures1.shape[1]
matchCommand = "java match " +str(rowX)+ " " +str(rowY)+ " " +str(column)
print matchCommand
os.system(matchCommand)
# plot according to match stored in "data" folder
matchFile = open("data/match", 'r')
lines = matchFile.readlines()
matchSize = len(lines)
oneSize = one.shape[0]
twoSize = two.shape[0]
ratio = matchSize / float(min(oneSize, twoSize))
if ratio > threshold:
graph.connect(i,j)
# Find nodes with largest edges in each connected component
connectedComponents = graph.connectedComponent()
tempkeyFrames = []
for component in connectedComponents:
edges = []
for node in component:
edges.append(graph.getNumOfEdges(node))
maxIndice = 0
maxValue = edges[0]
for i in range(1, len(edges), 1):
if maxValue < edges[i]:
maxValue = edges[i]
maxIndice = i
# random choose one if there are many within one component
maxEdges = []
for i in range(len(edges)):
if edges[i] == maxValue:
maxEdges.append(i)
if len(maxEdges) == 1:
tempkeyFrames.append(component[maxIndice])
else:
maxEdgeSize = len(maxEdges)
randomNumber = randint(0, maxEdgeSize - 1)
tempkeyFrames.append(component[randomNumber])
keyFrames = []
for indice in tempkeyFrames:
keyFrames.append(indices[indice])
return keyFrames
def encrypt(plaintext, key):
return normalize(plaintext).translate(maketrans('ABCDEFGHIJKLMNOPQRSTUVWXYZ', normalize(key)))
def decrypt(cipher, key):
return normalize(cipher).translate(
maketrans(normalize(key), 'ABCDEFGHIJKLMNOPQRSTUVWXYZ'))
def encrypt(plaintext, key):
return normalize(plaintext).translate(
maketrans('ABCDEFGHIJKLMNOPQRSTUVWXYZ', normalize(key)))
a = [1, 2, 3]
b = [4, 5, 6]
c = [7, 8, 9]
mt1 = np.array([a, b, c], dtype=int) # 这个完美解决ndarray之中存储的是list的问题，但是要求每行长度相等
print(mt1)
mt1.transpose()
print("转置后", mt1)
mt2 = np.array([[1.0, 2.0], [3.0, 4.0]]).transpose()
print(mt2)
mt2.transpose()
print(mt2.transpose()) # 该矩阵的转置矩阵，这里的transpose()并不是modification函数，所以不会修改原始数据
mt3 = np.vstack((a, b))
print(mt3)
mt3 = np.vstack((mt3, c))
print(mt3)
mt0 = np.array(a)
print(mt0)
matrix1 = np.matrix([1, 2, 3])
matrix2 = np.matrix([3, 2, 3])
m = matrix1 - matrix2
print(np.sqrt(np.sum(np.square(m))))
list1 = [[1, 2, 3], [4, 5, 6]]
list1[1][2] = 0
utility.normalize(list1[0])
print(list1)
locations1, features1 = sift.siftFeature(img1)
locations2, features2 = sift.siftFeature(img2)
# use PCAac to reduce dimensions
numberOfFeaturesOne = locations1.shape[0]
numberOfFeaturesTwo = locations2.shape[0]
features = np.vstack((features1, features2))
V, S, mean = pca.pca(features)
pcaFeatures1 = pca.project(features1, V, 36)
pcaFeatures2 = pca.project(features2, V, 36)
# normalize features
util.normalize(pcaFeatures1)
util.normalize(pcaFeatures2)
np.savetxt("pcafeature1", pcaFeatures1, delimiter="\t")
np.savetxt("pcafeature2", pcaFeatures2, delimiter="\t")
# interface with Java program to do matching
rowX = pcaFeatures1.shape[0]
rowY = pcaFeatures2.shape[0]
column = pcaFeatures1.shape[1]
matchCommand = "java match " + str(rowX) + " " + str(rowY) + " " + str(
column)
print matchCommand
os.system(matchCommand)
def decrypt(cipher, key):
return normalize(cipher).translate(maketrans(normalize(key),'ABCDEFGHIJKLMNOPQRSTUVWXYZ'))
def DataGenBB(self, train_start, train_end, DataStrs=None):
generateFunc = ["original"]
# generateFunc = ["original", "resize", "rotate", "mirror", "translate", "brightnessAndContrast" ]
InputData = np.zeros(
[self.batch_size * len(generateFunc), self.imSize, self.imSize, 3],
dtype=np.float32)
# InputLabel = np.zeros([self.batch_size * len(generateFunc), 7], dtype = np.float32)
InputLabel = np.zeros([self.batch_size * len(generateFunc), 3],
dtype=np.float32)
InputNames = []
# InputNames = np.zeros([self.batch_size * len(generateFunc), 1], dtype = np.float32)
count = 0
print "train_start,train_end: ", train_start, train_end
# for i in range(train_start,train_end):
i = train_start
while (i < train_end):
if self.ifMenpo39DataSet:
imgName = self.imgs[i]
imgNameHeader = imgName.split('.')[0]
index = imgNameHeader[imgNameHeader.find('e') + 1:]
labelsPTS = np.loadtxt(self.PTSDir + 'pts' + index + ".txt")
img = cv2.imread(self.ImgDir + imgName)
elif self.ifpreProcessedSemifrontal:
imgName = self.imgs[i]
# print "imgName: ", imgName
imgNameHeader = imgName.split('.')[0]
print "self.ImgDir + imgName: ", self.ImgDir + imgName
img = cv2.imread(self.ImgDir + imgName)
print "self.labelDir + imgNameHeader + .txt: ", self.labelDir + imgNameHeader + ".txt"
label = np.loadtxt(self.labelDir + imgNameHeader + ".txt")
else:
strLine = DataStrs[i]
strCells = strLine.rstrip(' \n').split(' ')
imgName = strCells[0]
label = np.array(strCells[1:]).astype(np.float)
labelsPTS = labels[:136].reshape([68, 2])
img = cv2.imread(imgName)
if self.debug:
cv2.imshow("original", img)
cv2.waitKey(0)
if img != None:
# print "img.shape: ", img.shape
img = cv2.resize(img, (self.imSize, self.imSize))
# (w, h, _) = img.shape
if self.ifMenpo39DataSet:
# x, y = self.unpackLandmarks(labelsPTS)
x, y = None, None
elif self.ifpreProcessedSemifrontal:
# x, y = None, None
# labels = [None, None, None]
newImg = img
# else:
# x, y = ut.unpackLandmarks(labelsPTS, self.imSize)
# for index in range(len(generateFunc)):
# method = generateFunc[index]
# # tag = random.choice(generateFunc)
# if method == "resize":
# newImg, newX, newY = ut.resize(img, x, y, xMaxBound = w, yMaxBound = h, random = True)
# elif method == "rotate":
# newImg, newX, newY = ut.rotate(img, x, y, w = w, h = h)
# elif method == "mirror":
# newImg, newX, newY = ut.mirror(img, x, y, w = w, h = h)
# elif method == "translate":
# newImg, newX, newY = ut.translate(img, x, y, w = w, h = h)
# elif method == "brightnessAndContrast":
# newImg, newX, newY = ut.contrastBrightess(img, x, y)
# elif method == "original":
# newImg, newX, newY = img, x, y
# else:
# raise "not existing function"
if self.ifMenpo39DataSet:
label = labelsPTS
elif self.ifpreProcessedSemifrontal:
pass
else:
newXMin = min(newX)
newYMin = min(newY)
newXMax = max(newX)
newYMax = max(newY)
newXMean = (newXMax + newXMin) / 2.0
newYMean = (newYMax + newYMin) / 2.0
newEdge = max(newYMax - newYMin, newXMax - newXMin)
normX = ut.normalize(newX, self.imSize)
normY = ut.normalize(newY, self.imSize)
normXMean, normYMean, normEdge = ut.normalize(
newXMean, self.imSize), ut.normalize(
newYMean,
self.imSize), ut.normalize(newEdge, self.imSize)
label = np.array([normXMean, normYMean, normEdge])
InputData[count, ...] = newImg
InputLabel[count, ...] = label
InputNames.append(imgName)
# InputNames[count,...] = imgName
# print "InputNames.shape: ", len(InputNames)
else:
print "cannot : ", imgName
print "count: ", count
count += 1
i += 1
return InputData, InputLabel, np.asarray(InputNames)