def traingNetwork(self, nn, numberOfSet):
print "Training starting:"
imageNumberList = loadData.getTrainingImageNumberList(numberOfSet)
d, t = loadData.getImageAndTarget(imageNumberList[0])
for i in range(1, len(imageNumberList)):
#print "Forwardpass"
nn.Calculate(d)
if (i % (numberOfSet / 10) == 0):
print "Number of iterations:", i
nn.learningRate -= 0.000001
nn.Backpropagate(nn.outputVector, t)
d, t = loadData.getImageAndTarget(imageNumberList[i])
saveWeights("1", str(numberOfSet), nn.layers[1].weights)
saveWeights("2", str(numberOfSet), nn.layers[2].weights)
saveWeights("3", str(numberOfSet), nn.layers[3].weights)
saveWeights("4", str(numberOfSet), nn.layers[4].weights)
print "Training completed. Weights are saved.\n"
return nn
def traingNetwork(nn,numberOfSet):
print "Training starting:"
imageNumberList = loadData.getTrainingImageNumberList(numberOfSet)
d,t = loadData.getImageAndTarget(imageNumberList[0])
for i in range(1,len(imageNumberList)):
#print "Forwardpass"
nn.ForwardPass(d)
if(i%(numberOfSet/10)==0):
print "Number of iterations:",i
nn.learningRate -=0.00001
nn.Backpropagate(nn.outputVector,t)
d,t = loadData.getImageAndTarget(imageNumberList[i])
saveWeights("1",str(numberOfSet),nn.layers[1].weights)
saveWeights("2",str(numberOfSet),nn.layers[2].weights)
saveWeights("3",str(numberOfSet),nn.layers[3].weights)
saveWeights("4",str(numberOfSet),nn.layers[4].weights)
print "Training completed. Weights are saved.\n"
return nn
def runCNN(self, image):
d, t = loadData.getImageAndTarget(100)
# Forward-pass
nn.Calculate(d)
return nn.outputVector.index(max(nn.outputVector))
def testNetwork(self, nn, numberOfSet, numberOfTest):
print "Testing starting:"
# Set weights from file
nn = setWeights(nn, numberOfSet)
imageNumberList = loadData.getTestImageNumberList(numberOfTest)
correct = 0
for i in range(0, len(imageNumberList)):
# Get random picture
d, t = loadData.getImageAndTarget(imageNumberList[i])
# Forward-pass
nn.Calculate(d)
correctGuess = False
# Check if result is correct
if (nn.outputVector.index(max(nn.outputVector)) == t.index(
max(t))):
correct += 1
correctGuess = True
print "CNN:", nn.outputVector.index(max(
nn.outputVector)), "Target:", t.index(max(t)), correctGuess
print "\nNumber of correct:", correct
print "Number of pictures", numberOfTest
print "Percentage", (correct * 1.0 / numberOfTest) * 100
def visualiseNetwork(nn, numberOfSet):
nn = setWeights(nn, numberOfSet)
d, t = loadData.getImageAndTarget(random.randint(0, 60000))
nn.Calculate(d)
visualise.visualise(nn)
def visualiseNetwork(nn,numberOfSet): nn = setWeights(nn,numberOfSet) d,t = loadData.getImageAndTarget(random.randint(0,60000)) nn.ForwardPass(d) visualise.visualise(nn)
def testNetwork(nn,numberOfSet,numberOfTest,modification=False): print "Testing starting:" # Set weights from file nn = setWeights(nn,numberOfSet) imageNumberList = loadData.getTestImageNumberList(numberOfTest) correct = 0 for i in range(0,len(imageNumberList)): # Get random picture d,t = loadData.getImageAndTarget(imageNumberList[i]) # Forward-pass nn.ForwardPass(d) correctGuess = False # Check if result is correct if(nn.outputVector.index(max(nn.outputVector))==t.index(max(t))): correct+=1 correctGuess = True print "CNN:",nn.outputVector.index(max(nn.outputVector)),"Target:",t.index(max(t)),correctGuess print "\nNumber of correct:",correct print "Number of pictures",numberOfTest print "Percentage",(correct*1.0/numberOfTest) * 100
def runCNN(self,image): d,t = loadData.getImageAndTarget(100) # Forward-pass nn.Calculate(d) return nn.outputVector.index(max(nn.outputVector))
def getNetworkImage(nn,number): d,t = loadData.getImageAndTarget(number) nn.ForwardPass(d) out = visualise.getNeuronOutputs(nn) out = np.array(out[0]) for i in range(0,len(out)): for j in range(0,len(out[i])): out[i][j] = out[i][j]*100 s = base64.b64encode(out) s = "data:image/png;base64,"+s return s
def getNetworkImage(nn, number):
d, t = loadData.getImageAndTarget(number)
nn.ForwardPass(d)
out = visualise.getNeuronOutputs(nn)
out = np.array(out[0])
for i in range(0, len(out)):
for j in range(0, len(out[i])):
out[i][j] = out[i][j] * 100
s = base64.b64encode(out)
s = "data:image/png;base64," + s
return s
def runCNN(nn,image):
print image
#
# This method takes an image and returns the classified label of that image
#
#
# Converts the image to png and saves it localy
#
image = image.replace('data:image/png;base64,','')
fh = open("imageToSave.png", "wb")
image = image.decode('base64')
fh.write(image)
fh.close()
#
# Downsample image to 29x29 pixels and saves it
#
import PIL
from PIL import Image
basewidth = 29
img = Image.open('imageToSave.png')
wpercent = (basewidth/float(img.size[0]))
hsize = int((float(img.size[1])*float(wpercent)))
img = img.resize((basewidth,hsize), PIL.Image.ANTIALIAS)
img.save('imageToSave.png')
# Load the image
img = matplotlib.image.imread("imageToSave.png")
#
# Imported images is on wrong format. [[0,0,0],[0,0,0],[0,0,0]], have to have it in one list
#
c = 0
newimage= []
for i in range(len(img)):
for j in range(len(img[i])):
c = 0
for k in range(len(img[i][j])):
c+= img[i][j][k]
newimage.append(c)
#
# Since the image created digital, the image don't have the "natural" features as a handwritten image have
# Have to do some stuff to make it more applicable.
#
# Input picture is something like this:[[0,0,0]
# [0,1,0]
# [0,0,0]]
#
#
# Output will be something like this: [[0, 52, 0]
# [102, 250, 69]
# [0, 94, 0]]
#
#
# For every pixel with value in, the neighbours on all sides of that pixel will have a random value from 50 to 150.
#
padd = [0]*len(newimage)
for i in range(29*2,len(newimage)-29*2):
if(newimage[i]>0.12):
padd[i-1] = random.randint(50,150)
padd[i+1] = random.randint(50,150)
#padd[i-2] = 100
#padd[i+2] = 100
padd[i-29] = random.randint(50,150)
padd[i+29] = random.randint(50,150)
#padd[i-29*2] = 100
#padd[i+29*2] = 100
for i in range(len(newimage)):
if(newimage[i]>0.12 and newimage[i] != 150):
padd[i] = random.randint(240,255)
newimage = padd
#
# This part is just to look how the created picture is compared to one from the training/testing set
#
d,t = loadData.getImageAndTarget(random.randint(0,60000))
#
# Get better printing format for both images
#
#for i in range(0,29):
# a = []
# for j in range(0,29):
# a.append(d[i*29+j])
# print a
#print t.index(max(t))
#print " "
#for i in range(0,29):
# a = []
# for j in range(0,29):
# a.append(newimage[i*29+j])
# print a
# Forward-pass
nn.ForwardPass(newimage)
#
# Return dictionary to server
#
sort = sorted(nn.outputVector[:])[::-1]
ranked = {}
for i in range(0,len(nn.outputVector)):
ranked[i] = (nn.outputVector[nn.outputVector.index(sort[i])],(nn.outputVector.index(sort[i])))
return ranked
def runCNN(nn, image):
print image
#
# This method takes an image and returns the classified label of that image
#
#
# Converts the image to png and saves it localy
#
image = image.replace('data:image/png;base64,', '')
fh = open("imageToSave.png", "wb")
image = image.decode('base64')
fh.write(image)
fh.close()
#
# Downsample image to 29x29 pixels and saves it
#
import PIL
from PIL import Image
basewidth = 29
img = Image.open('imageToSave.png')
wpercent = (basewidth / float(img.size[0]))
hsize = int((float(img.size[1]) * float(wpercent)))
img = img.resize((basewidth, hsize), PIL.Image.ANTIALIAS)
img.save('imageToSave.png')
# Load the image
img = matplotlib.image.imread("imageToSave.png")
#
# Imported images is on wrong format. [[0,0,0],[0,0,0],[0,0,0]], have to have it in one list
#
c = 0
newimage = []
for i in range(len(img)):
for j in range(len(img[i])):
c = 0
for k in range(len(img[i][j])):
c += img[i][j][k]
newimage.append(c)
#
# Since the image created digital, the image don't have the "natural" features as a handwritten image have
# Have to do some stuff to make it more applicable.
#
# Input picture is something like this:[[0,0,0]
# [0,1,0]
# [0,0,0]]
#
#
# Output will be something like this: [[0, 52, 0]
# [102, 250, 69]
# [0, 94, 0]]
#
#
# For every pixel with value in, the neighbours on all sides of that pixel will have a random value from 50 to 150.
#
padd = [0] * len(newimage)
for i in range(29 * 2, len(newimage) - 29 * 2):
if (newimage[i] > 0.12):
padd[i - 1] = random.randint(50, 150)
padd[i + 1] = random.randint(50, 150)
#padd[i-2] = 100
#padd[i+2] = 100
padd[i - 29] = random.randint(50, 150)
padd[i + 29] = random.randint(50, 150)
#padd[i-29*2] = 100
#padd[i+29*2] = 100
for i in range(len(newimage)):
if (newimage[i] > 0.12 and newimage[i] != 150):
padd[i] = random.randint(240, 255)
newimage = padd
#
# This part is just to look how the created picture is compared to one from the training/testing set
#
d, t = loadData.getImageAndTarget(random.randint(0, 60000))
#
# Get better printing format for both images
#
#for i in range(0,29):
# a = []
# for j in range(0,29):
# a.append(d[i*29+j])
# print a
#print t.index(max(t))
#print " "
#for i in range(0,29):
# a = []
# for j in range(0,29):
# a.append(newimage[i*29+j])
# print a
# Forward-pass
nn.ForwardPass(newimage)
#
# Return dictionary to server
#
sort = sorted(nn.outputVector[:])[::-1]
ranked = {}
for i in range(0, len(nn.outputVector)):
ranked[i] = (nn.outputVector[nn.outputVector.index(sort[i])],
(nn.outputVector.index(sort[i])))
return ranked
