def exercise1_3(N=200,P=50,tests=100):
print "Exercise 1.3:"
h = HopfieldNetwork()
error = zeros((P,tests),float)
for p in range(P):
for q in range(tests):
if p > 0:
r = np.random.randint(0,p)
else:
r = 0
error[p,q]=(h.run(N, P=p+1, mu=r, flip_ratio=0.1, bPlot=False))
print error;
stdev=np.std(error,axis=1)
meanvalues=np.mean(error,axis=1)
fig = figure()
ax1 = fig.add_subplot(111)
lp = errorbar(range(1,P+1),meanvalues,stdev,color='g')
bp = boxplot(transpose(error))
axhline(y=0.02,linewidth=1, color='r')
xticks(arange(0,P+1,P/10))
setp(lp,color='black')
ax1.set_ylim(0,0.5)
ax1.yaxis.grid(True,linestyle='-',which='major',
color=(0.2,0.2,0.2),alpha=0.5)
ax1.set_axisbelow(True)
ax1.set_title('Mean retrieval error over different'
'network realizations (N=%d, tests=%d)' % (N,tests) )
ax1.set_ylabel('Mean retrieval error')
ax1.set_xlabel('Dictonary size P')
savefig('../tex/dat/ex1_3-error_avg-N%d-P%d-Q%d.png' % (N,P,tests))
close()
def show_one(hopfield_net, rows_no, cols_no, patterns, args):
# Choose a pattern
original_pattern = choice(patterns)
hopfield_net.reset(original_pattern)
hopfield_net.display_as_matrix(rows_no, cols_no, msg='Original pattern')
# hopfield_net.display_as_image(rows_no, cols_no)
# Apply noise to it
noisy_pattern = apply_noise(original_pattern, args.noise) # apply noise
hopfield_net.reset(noisy_pattern) # The net starts from the noisy pattern
hopfield_net.display_as_matrix(rows_no, cols_no, msg='Noisy pattern')
# hopfield_net.display_as_image(rows_no, cols_no)
# Let the network reach an energetic mimimum
hopfield_net.convergence_test = False
while not hopfield_net.is_energy_minimal():
hopfield_net.single_update()
hopfield_net.display_as_matrix(rows_no, cols_no, msg='After update')
print("Energy: %f" % hopfield_net.energy())
# hopfield_net.display_as_image(rows_no, cols_no)
if original_pattern == HopfieldNetwork.state_to_string(hopfield_net.state):
print("Recovered OK")
else:
print("NOT RECOVERED OK")
print("Done!")
def __init__(self, interpreter=None, parent=None):
# initialization of the superclass
super(DesignerMainWindow, self).__init__(parent)
# setup the GUI --> function generated by pyuic4
self.setupUi(self)
self.hop = HopfieldNetwork(100)
self.hop.setClocksHopfield()
self.defaultButtonState()
self.mem1 = []
self.mem2 = []
self.mem3 = []
self.mem4 = []
self.input = []
self.connectElements()
self.runCount = 0
self.runPushButton.setEnabled(False)
self.statusbar.showMessage('Hopfield Network Demo')
self.inputChanged = False
def __init__(self, interpreter=None, parent=None):
# initialization of the superclass
super(DesignerMainWindow, self).__init__(parent)
# setup the GUI --> function generated by pyuic4
self.setupUi(self)
self.hop = HopfieldNetwork(100)
self.hop.setClocksHopfield()
self.defaultButtonState()
self.mem1 = []
self.mem2 = []
self.mem3 = []
self.mem4 = []
self.input = []
self.connectElements()
self.runCount = 0
self.runPushButton.setEnabled(False)
self.statusbar.showMessage("Hopfield Network Demo")
self.inputChanged = False
def test_hopfield(hopfield_net, patterns, args):
########################## TASK 2: Testarea retelei #######################
# args.test_no - numarul de sabloane ce vor fi testate
# intoarce acuratetea
count_correct = 0
for original_pattern in patterns:
hopfield_net.reset(original_pattern)
hopfield_net.display_as_matrix(rows_no,
cols_no,
msg='Original pattern')
# hopfield_net.display_as_image(rows_no, cols_no)
# Apply noise to it
noisy_pattern = apply_noise(original_pattern,
args.noise) # apply noise
hopfield_net.reset(
noisy_pattern) # The net starts from the noisy pattern
hopfield_net.display_as_matrix(rows_no, cols_no, msg='Noisy pattern')
# hopfield_net.display_as_image(rows_no, cols_no)
# Let the network reach an energetic mimimum
hopfield_net.convergence_test = False
while not hopfield_net.is_energy_minimal():
hopfield_net.single_update()
hopfield_net.display_as_matrix(rows_no,
cols_no,
msg='After update')
print("Energy: %f" % hopfield_net.energy())
# hopfield_net.display_as_image(rows_no, cols_no)
if original_pattern == HopfieldNetwork.state_to_string(
hopfield_net.state):
count_correct += 1
print("Recovered OK")
return count_correct / len(patterns)
length = len(v1)
for i in range(length):
if v1[i] != v2[i]:
diff += 1
diff /= length
return diff
countOfPatterns = 2
size = 100
#vectors = csv_parser.read_input('SN_projekt3/animals-14x9.csv')
vectors = PngReader.getSomeBitmaps(countOfPatterns, 'pokemon' + str(size))
width = size
height = size
hn = HopfieldNetwork(height * width, countOfPatterns, False, 0.1, 1)
result = hn.train(vectors)
print("Result: ", result)
f, axarr = plt.pyplot.subplots(countOfPatterns, 2)
for i in range(countOfPatterns):
axarr[i, 0].imshow(Img.vector_to_img(
vectors[i],
height,
width,
False,
))
axarr[i, 1].imshow(Img.vector_to_img(hn.Z[i], height, width, False))
diff = calculateDiff(vectors[i], hn.Z[i])
plt.pyplot.sca(axarr[i, 1])
class DesignerMainWindow(QtGui.QMainWindow, Ui_MainWindow):
"""Customization for Qt Designer created window"""
def __init__(self, interpreter=None, parent=None):
# initialization of the superclass
super(DesignerMainWindow, self).__init__(parent)
# setup the GUI --> function generated by pyuic4
self.setupUi(self)
self.hop = HopfieldNetwork(100)
self.hop.setClocksHopfield()
self.defaultButtonState()
self.mem1 = []
self.mem2 = []
self.mem3 = []
self.mem4 = []
self.input = []
self.connectElements()
self.runCount = 0
self.runPushButton.setEnabled(False)
self.statusbar.showMessage("Hopfield Network Demo")
self.inputChanged = False
def defaultButtonState(self):
self.mem1PushButton.setEnabled(False)
self.mem2PushButton.setEnabled(False)
self.mem3PushButton.setEnabled(False)
self.mem4PushButton.setEnabled(False)
self.inputPushButton.setEnabled(False)
def connectElements(self):
self.connect(self.clearDisplayPushButton, QtCore.SIGNAL("clicked()"), self.clearDisplay)
self.connect(self.memorizePushButton, QtCore.SIGNAL("clicked()"), self.memorizePattern)
self.connect(self.randomizePushButton, QtCore.SIGNAL("clicked()"), self.randomizePattern)
self.connect(self.clearMemPushButton, QtCore.SIGNAL("clicked()"), self.clearMemory)
self.connect(self.runPushButton, QtCore.SIGNAL("clicked()"), self.runStep)
self.connect(self.mem1PushButton, QtCore.SIGNAL("clicked()"), self.retriveMem1)
self.connect(self.mem2PushButton, QtCore.SIGNAL("clicked()"), self.retriveMem2)
self.connect(self.mem3PushButton, QtCore.SIGNAL("clicked()"), self.retriveMem3)
self.connect(self.mem4PushButton, QtCore.SIGNAL("clicked()"), self.retriveMem4)
self.connect(self.inputPushButton, QtCore.SIGNAL("clicked()"), self.retriveInput)
self.connect(self.saveInputPushButton, QtCore.SIGNAL("clicked()"), self.saveInput)
self.connect(self.aPushButton, QtCore.SIGNAL("clicked()"), self.retriveA)
self.connect(self.bPushButton, QtCore.SIGNAL("clicked()"), self.retriveB)
self.connect(self.cPushButton, QtCore.SIGNAL("clicked()"), self.retriveC)
self.connect(self.dPushButton, QtCore.SIGNAL("clicked()"), self.retriveD)
self.connect(self.computeSynWeightsPushButton, QtCore.SIGNAL("clicked()"), self.computeAllWeights)
def saveInput(self):
print "saving current pattern as input"
inpList = []
for i in range(100):
exec (("inpList.append(int(self.pushButton_%s.isChecked()))" % i))
self.input = inpList
self.inputPushButton.setEnabled(True)
self.inputChanged = True
self.runCount = 0
def computeAllWeights(self):
self.statusbar.showMessage("Computing synaptic weights")
self.hop.assignAllSynapticWeights() # weights are assigned
self.runPushButton.setEnabled(True)
def runStep(self):
self.statusbar.showMessage("Running")
if self.inputChanged:
inputPattern = self.input
self.hop.updateInputs(inputPattern) # input updating
self.hop.mooseReinit()
self.hop.runMooseHopfield(0.0310)
if self.runCount == 0:
for i in range(100):
if np.any((self.hop.allSpikes[i].vector > 0.0) & (self.hop.allSpikes[i].vector < 0.0310)):
exec (("self.pushButton_%s.setChecked(True)" % i))
else:
exec (("self.pushButton_%s.setChecked(False)" % i))
else:
self.hop.runMooseHopfield(0.02)
# print 0.0310+(0.02*(self.runCount-1)),(0.0310+(0.02*self.runCount))
for i in range(100):
if np.any(
(self.hop.allSpikes[i].vector > (0.0301 + (0.02 * (self.runCount - 1))))
& (self.hop.allSpikes[i].vector < (0.0301 + (0.02 * self.runCount)))
):
exec (("self.pushButton_%s.setChecked(True)" % i))
else:
exec (("self.pushButton_%s.setChecked(False)" % i))
self.runCount += 1
self.statusbar.showMessage("Done Running")
def retriveMem1(self):
for i in range(100):
exec (("self.pushButton_%s.setChecked(self.mem1[int(%s)])" % (i, i)))
self.statusbar.showMessage("Showing Mem1")
def retriveMem2(self):
for i in range(100):
exec (("self.pushButton_%s.setChecked(self.mem2[int(%s)])" % (i, i)))
self.statusbar.showMessage("Showing Mem2")
def retriveMem3(self):
for i in range(100):
exec (("self.pushButton_%s.setChecked(self.mem3[int(%s)])" % (i, i)))
self.statusbar.showMessage("Showing Mem3")
def retriveMem4(self):
for i in range(100):
exec (("self.pushButton_%s.setChecked(self.mem4[int(%s)])" % (i, i)))
self.statusbar.showMessage("Showing Mem4")
def retriveInput(self):
for i in range(100):
exec (("self.pushButton_%s.setChecked(self.input[int(%s)])" % (i, i)))
self.statusbar.showMessage("Showing Input")
def retriveA(self):
a = [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
1,
0,
0,
0,
0,
0,
1,
0,
0,
0,
1,
0,
0,
0,
0,
0,
1,
1,
1,
1,
1,
0,
0,
0,
0,
0,
1,
0,
0,
0,
1,
0,
0,
0,
0,
0,
1,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
]
for i in range(100):
exec (("self.pushButton_%s.setChecked(a[int(%s)])" % (i, i)))
self.statusbar.showMessage("Showing Sample A")
def retriveB(self):
b = [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
1,
1,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
1,
0,
0,
0,
0,
0,
0,
1,
1,
1,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
1,
0,
0,
0,
0,
0,
0,
1,
1,
1,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
]
for i in range(100):
exec (("self.pushButton_%s.setChecked(b[int(%s)])" % (i, i)))
self.statusbar.showMessage("Showing Sample B")
def retriveC(self):
c = [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
1,
1,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
1,
1,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
]
for i in range(100):
exec (("self.pushButton_%s.setChecked(c[int(%s)])" % (i, i)))
self.statusbar.showMessage("Showing Sample C")
def retriveD(self):
d = [
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
1,
1,
1,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
1,
0,
0,
0,
0,
0,
0,
1,
0,
0,
1,
0,
0,
0,
0,
0,
0,
1,
1,
1,
1,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
0,
]
for i in range(100):
exec (("self.pushButton_%s.setChecked(d[int(%s)])" % (i, i)))
self.statusbar.showMessage("Showing Sample D")
def clearMemory(self):
self.hop.clearAllMemory()
self.clearDisplay()
self.defaultButtonState()
self.mem1 = []
self.mem2 = []
self.mem3 = []
self.input = []
self.statusbar.showMessage("Cleared all Memory")
self.runPushButton.setEnabled(False)
self.inputPushButton.setEnabled(False)
self.runCount = 0
def randomizePattern(self):
r = np.random.randint(2, size=100)
for i in range(100):
exec (("self.pushButton_%s.setChecked(r[int(%s)])" % (i, i)))
def memorizePattern(self):
pattern = []
for i in range(100):
exec (("pattern.append(int(self.pushButton_%s.isChecked()))" % i))
self.hop.updateWeights(pattern)
for k in range(100): # very weird that I have to do this!
if pattern[k] == -1:
pattern[k] = 0
savedAlready = self.hop.numMemories
if savedAlready == 1:
self.mem1 = pattern
self.mem1PushButton.setEnabled(True)
# print pattern
elif savedAlready == 2:
self.mem2 = pattern
self.mem2PushButton.setEnabled(True)
# print pattern
elif savedAlready == 3:
self.mem3 = pattern
self.mem3PushButton.setEnabled(True)
elif savedAlready == 4:
self.mem4 = pattern
self.mem4PushButton.setEnabled(True)
self.runPushButton.setEnabled(False)
self.statusbar.showMessage("New Pattern Memorised")
def clearDisplay(self):
for i in range(100):
exec (("self.pushButton_%s.setChecked(False)" % i))
self.statusbar.showMessage("Cleared display")
class DesignerMainWindow(QtGui.QMainWindow, Ui_MainWindow):
"""Customization for Qt Designer created window"""
def __init__(self, interpreter=None, parent=None):
# initialization of the superclass
super(DesignerMainWindow, self).__init__(parent)
# setup the GUI --> function generated by pyuic4
self.setupUi(self)
self.hop = HopfieldNetwork(100)
self.hop.setClocksHopfield()
self.defaultButtonState()
self.mem1 = []
self.mem2 = []
self.mem3 = []
self.mem4 = []
self.input = []
self.connectElements()
self.runCount = 0
self.runPushButton.setEnabled(False)
self.statusbar.showMessage('Hopfield Network Demo')
self.inputChanged = False
def defaultButtonState(self):
self.mem1PushButton.setEnabled(False)
self.mem2PushButton.setEnabled(False)
self.mem3PushButton.setEnabled(False)
self.mem4PushButton.setEnabled(False)
self.inputPushButton.setEnabled(False)
def connectElements(self):
self.connect(self.clearDisplayPushButton, QtCore.SIGNAL('clicked()'),
self.clearDisplay)
self.connect(self.memorizePushButton, QtCore.SIGNAL('clicked()'),
self.memorizePattern)
self.connect(self.randomizePushButton, QtCore.SIGNAL('clicked()'),
self.randomizePattern)
self.connect(self.clearMemPushButton, QtCore.SIGNAL('clicked()'),
self.clearMemory)
self.connect(self.runPushButton, QtCore.SIGNAL('clicked()'),
self.runStep)
self.connect(self.mem1PushButton, QtCore.SIGNAL('clicked()'),
self.retriveMem1)
self.connect(self.mem2PushButton, QtCore.SIGNAL('clicked()'),
self.retriveMem2)
self.connect(self.mem3PushButton, QtCore.SIGNAL('clicked()'),
self.retriveMem3)
self.connect(self.mem4PushButton, QtCore.SIGNAL('clicked()'),
self.retriveMem4)
self.connect(self.inputPushButton, QtCore.SIGNAL('clicked()'),
self.retriveInput)
self.connect(self.saveInputPushButton, QtCore.SIGNAL('clicked()'),
self.saveInput)
self.connect(self.aPushButton, QtCore.SIGNAL('clicked()'),
self.retriveA)
self.connect(self.bPushButton, QtCore.SIGNAL('clicked()'),
self.retriveB)
self.connect(self.cPushButton, QtCore.SIGNAL('clicked()'),
self.retriveC)
self.connect(self.dPushButton, QtCore.SIGNAL('clicked()'),
self.retriveD)
self.connect(self.computeSynWeightsPushButton,
QtCore.SIGNAL('clicked()'), self.computeAllWeights)
def saveInput(self):
print 'saving current pattern as input'
inpList = []
for i in range(100):
exec(('inpList.append(int(self.pushButton_%s.isChecked()))' % i))
self.input = inpList
self.inputPushButton.setEnabled(True)
self.inputChanged = True
self.runCount = 0
def computeAllWeights(self):
self.statusbar.showMessage('Computing synaptic weights')
self.hop.assignAllSynapticWeights() #weights are assigned
self.runPushButton.setEnabled(True)
def runStep(self):
self.statusbar.showMessage('Running')
if self.inputChanged:
inputPattern = self.input
self.hop.updateInputs(inputPattern) #input updating
self.hop.mooseReinit()
self.hop.runMooseHopfield(0.0310)
if self.runCount == 0:
for i in range(100):
if np.any((self.hop.allSpikes[i].vector > 0.0)
& (self.hop.allSpikes[i].vector < 0.0310)):
exec(('self.pushButton_%s.setChecked(True)' % i))
else:
exec(('self.pushButton_%s.setChecked(False)' % i))
else:
self.hop.runMooseHopfield(0.02)
#print 0.0310+(0.02*(self.runCount-1)),(0.0310+(0.02*self.runCount))
for i in range(100):
if np.any((self.hop.allSpikes[i].vector >
(0.0301 + (0.02 * (self.runCount - 1))))
& (self.hop.allSpikes[i].vector <
(0.0301 + (0.02 * self.runCount)))):
exec(('self.pushButton_%s.setChecked(True)' % i))
else:
exec(('self.pushButton_%s.setChecked(False)' % i))
self.runCount += 1
self.statusbar.showMessage('Done Running')
def retriveMem1(self):
for i in range(100):
exec(('self.pushButton_%s.setChecked(self.mem1[int(%s)])' %
(i, i)))
self.statusbar.showMessage('Showing Mem1')
def retriveMem2(self):
for i in range(100):
exec(('self.pushButton_%s.setChecked(self.mem2[int(%s)])' %
(i, i)))
self.statusbar.showMessage('Showing Mem2')
def retriveMem3(self):
for i in range(100):
exec(('self.pushButton_%s.setChecked(self.mem3[int(%s)])' %
(i, i)))
self.statusbar.showMessage('Showing Mem3')
def retriveMem4(self):
for i in range(100):
exec(('self.pushButton_%s.setChecked(self.mem4[int(%s)])' %
(i, i)))
self.statusbar.showMessage('Showing Mem4')
def retriveInput(self):
for i in range(100):
exec(('self.pushButton_%s.setChecked(self.input[int(%s)])' %
(i, i)))
self.statusbar.showMessage('Showing Input')
def retriveA(self):
a = [
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0,
0, 0, 1, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 1,
0, 0, 0, 1, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 0, 0, 0, 0, 0, 1, 0, 0,
0, 1, 0, 0, 0, 0, 0, 1, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
]
for i in range(100):
exec(('self.pushButton_%s.setChecked(a[int(%s)])' % (i, i)))
self.statusbar.showMessage('Showing Sample A')
def retriveB(self):
b = [
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1,
0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0,
1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
]
for i in range(100):
exec(('self.pushButton_%s.setChecked(b[int(%s)])' % (i, i)))
self.statusbar.showMessage('Showing Sample B')
def retriveC(self):
c = [
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1,
0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0, 0,
0, 0, 0, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
]
for i in range(100):
exec(('self.pushButton_%s.setChecked(c[int(%s)])' % (i, i)))
self.statusbar.showMessage('Showing Sample C')
def retriveD(self):
d = [
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0,
0, 1, 1, 1, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1,
0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 0, 0,
1, 0, 0, 0, 0, 0, 0, 1, 0, 0, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 0,
0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0
]
for i in range(100):
exec(('self.pushButton_%s.setChecked(d[int(%s)])' % (i, i)))
self.statusbar.showMessage('Showing Sample D')
def clearMemory(self):
self.hop.clearAllMemory()
self.clearDisplay()
self.defaultButtonState()
self.mem1 = []
self.mem2 = []
self.mem3 = []
self.input = []
self.statusbar.showMessage('Cleared all Memory')
self.runPushButton.setEnabled(False)
self.inputPushButton.setEnabled(False)
self.runCount = 0
def randomizePattern(self):
r = np.random.randint(2, size=100)
for i in range(100):
exec(('self.pushButton_%s.setChecked(r[int(%s)])' % (i, i)))
def memorizePattern(self):
pattern = []
for i in range(100):
exec(('pattern.append(int(self.pushButton_%s.isChecked()))' % i))
self.hop.updateWeights(pattern)
for k in range(100): #very weird that I have to do this!
if pattern[k] == -1:
pattern[k] = 0
savedAlready = self.hop.numMemories
if savedAlready == 1:
self.mem1 = pattern
self.mem1PushButton.setEnabled(True)
#print pattern
elif savedAlready == 2:
self.mem2 = pattern
self.mem2PushButton.setEnabled(True)
#print pattern
elif savedAlready == 3:
self.mem3 = pattern
self.mem3PushButton.setEnabled(True)
elif savedAlready == 4:
self.mem4 = pattern
self.mem4PushButton.setEnabled(True)
self.runPushButton.setEnabled(False)
self.statusbar.showMessage('New Pattern Memorised')
def clearDisplay(self):
for i in range(100):
exec(('self.pushButton_%s.setChecked(False)' % i))
self.statusbar.showMessage('Cleared display')
def train_hopfield(rows_no, cols_no, patterns, args):
hopfield_net = HopfieldNetwork(rows_no * cols_no)
hopfield_net.learn_patterns(patterns, args.train_lr)
return hopfield_net
def exercise1_1(N=200,P=5,c=0.2):
print "Exercise 1.1:"
h = HopfieldNetwork()
h.run(N=N, P=P, flip_ratio=c, bPlot=True)
savefig('../tex/dat/ex1_1-energy_overlap-N%d-P%d-c%d.png' % (N,P,c*100))
close()
def testAllNetworks():
fileHandle = open(RESULT_FILE, 'w', 0)
appendHtmlHeader(fileHandle)
path = 'signs/'
images = [path+f for f in listdir(path)]
print images
netHebb = HopfieldNetwork(SIDE_OF_ARRAY*SIDE_OF_ARRAY)
netPseudoInv = HopfieldNetwork(SIDE_OF_ARRAY*SIDE_OF_ARRAY)
netDelta = HopfieldNetwork(SIDE_OF_ARRAY*SIDE_OF_ARRAY)
for image in images:
print "Learning image " + image
imageName = image
imgProc = ImgPreprocessor(image,SIDE_OF_ARRAY,127)
image = imgProc.getImageForHopfield().copy()
cv2.imshow('test',imgProc.getImage())
ImageSaver(imageName.replace('signs','results'), imgProc.getImage(),SIDE_OF_ARRAY)
netHebb.trainHebb(image)
netDelta.trainDelta(image)
netPseudoInv.appendTrainingVectorForPseudoInversion(image)
netPseudoInv.trainPseudoInversion()
print "Learning done"
#noise processing
for image in images:
tmpImage = image
for noiseFactor in [0.05, 0.10, 0.15]:
image = tmpImage
fileHandle.write('<tr>\n')
start = time.time()
print "Noise factor: " + str(int(noiseFactor*100)) + "%"
print "Processing image " + image
imageName = image
fileHandle.write('<td>\n')
fileHandle.write(imageName.replace('signs/','').replace('.png','') +' '+ '<img src=' + imageName.replace('signs','results') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
fileHandle.write('<td>\n')
fileHandle.write('Szum: ' + str(int(noiseFactor*100)) +'%')
fileHandle.write('\n')
fileHandle.write('</td>\n')
imgProc = ImgPreprocessor(image,SIDE_OF_ARRAY,127)
image = imgProc.getImageForHopfield().copy()
image = ImageNoise(image, noiseFactor).get()
ImageSaver(imageName.replace('signs/','results/rec_noise_' + str(int(noiseFactor*100)) + '_'), image,SIDE_OF_ARRAY)
fileHandle.write('<td>\n')
fileHandle.write('<img src=' + imageName.replace('signs/','results/rec_noise_' + str(int(noiseFactor*100)) + '_') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
netPseudoInv.initNeurons(image)
netDelta.initNeurons(image)
netHebb.initNeurons(image)
netHebb.updateUntilSatisfied()
ImageSaver(imageName.replace('signs/','results/hebb_noise_' + str(int(noiseFactor*100)) + '_'), netHebb.getNeuronsMatrix(),SIDE_OF_ARRAY)
netDelta.updateUntilSatisfied()
ImageSaver(imageName.replace('signs/','results/delta_noise_' + str(int(noiseFactor*100)) + '_'), netDelta.getNeuronsMatrix(),SIDE_OF_ARRAY)
netPseudoInv.updateUntilSatisfied()
ImageSaver(imageName.replace('signs/','results/pinv_noise_' + str(int(noiseFactor*100)) + '_'), netPseudoInv.getNeuronsMatrix(),SIDE_OF_ARRAY)
end = time.time()
fileHandle.write('<td>\n')
fileHandle.write('<img src=' + imageName.replace('signs/','results/hebb_noise_' + str(int(noiseFactor*100)) + '_') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
fileHandle.write('<td>\n')
fileHandle.write('<img src=' + imageName.replace('signs/','results/delta_noise_' + str(int(noiseFactor*100)) + '_') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
fileHandle.write('<td>\n')
fileHandle.write('<img src=' + imageName.replace('signs/','results/pinv_noise_' + str(int(noiseFactor*100)) + '_') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
print "Time elapsed " + str(int(end-start)) + " [s]"
fileHandle.write('</tr>\n')
#lines processing
for image in images:
tmpImage = image
for linesEvery in [16,8]:
image = tmpImage
fileHandle.write('<tr>\n')
start = time.time()
print "Lines every: " + str(int(linesEvery))
print "Processing image " + image
imageName = image
fileHandle.write('<td>\n')
fileHandle.write(imageName.replace('signs/','').replace('.png','') +' '+ '<img src=' + imageName.replace('signs','results') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
fileHandle.write('<td>\n')
fileHandle.write('Linie co: ' + str(int(linesEvery)))
fileHandle.write('\n')
fileHandle.write('</td>\n')
imgProc = ImgPreprocessor(image,SIDE_OF_ARRAY,127)
image = imgProc.getImageForHopfield().copy()
image = ImageLines(image, linesEvery).get()
ImageSaver(imageName.replace('signs/','results/rec_lines_' + str(linesEvery) + '_'), image,SIDE_OF_ARRAY)
fileHandle.write('<td>\n')
fileHandle.write('<img src=' + imageName.replace('signs/','results/rec_lines_' + str(linesEvery) + '_') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
netPseudoInv.initNeurons(image)
netDelta.initNeurons(image)
netHebb.initNeurons(image)
netHebb.updateUntilSatisfied()
ImageSaver(imageName.replace('signs/','results/hebb_lines_' + str(linesEvery) + '_'), netHebb.getNeuronsMatrix(),SIDE_OF_ARRAY)
netDelta.updateUntilSatisfied()
ImageSaver(imageName.replace('signs/','results/delta_lines_' + str(linesEvery) + '_'), netDelta.getNeuronsMatrix(),SIDE_OF_ARRAY)
netPseudoInv.updateUntilSatisfied()
ImageSaver(imageName.replace('signs/','results/pinv_lines_' + str(linesEvery) + '_'), netPseudoInv.getNeuronsMatrix(),SIDE_OF_ARRAY)
end = time.time()
fileHandle.write('<td>\n')
fileHandle.write('<img src=' + imageName.replace('signs/','results/hebb_lines_' + str(linesEvery) + '_') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
fileHandle.write('<td>\n')
fileHandle.write('<img src=' + imageName.replace('signs/','results/delta_lines_' + str(linesEvery) + '_') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
fileHandle.write('<td>\n')
fileHandle.write('<img src=' + imageName.replace('signs/','results/pinv_lines_' + str(linesEvery) + '_') + '>')
fileHandle.write('\n')
fileHandle.write('</td>\n')
print "Time elapsed " + str(int(end-start)) + " [s]"
fileHandle.write('</tr>\n')
appendHtmlFooter(fileHandle)
fileHandle.close()
def main(self):
SIDE_OF_ARRAY = 32
F = npyscreen.Form(name = "Road signs recognition via Hopfield Network",)
F.add(npyscreen.TitleFixedText, name = "Using dimensions: " + str(SIDE_OF_ARRAY) + "x" + str(SIDE_OF_ARRAY),)
netHebb = HopfieldNetwork(SIDE_OF_ARRAY*SIDE_OF_ARRAY)
netPseudoInv = HopfieldNetwork(SIDE_OF_ARRAY*SIDE_OF_ARRAY)
netDelta = HopfieldNetwork(SIDE_OF_ARRAY*SIDE_OF_ARRAY)
F.add(npyscreen.TitleFixedText, name = "Network instances initialized",)
F.display()
signsList, imgToRecogn = self.loadParameters(sys.argv)
F.add(npyscreen.TitleFixedText, name = "Image to recognize: " + imgToRecogn,)
images = list()
for image in signsList:
imageName = image
imgProc = ImgPreprocessor(image,SIDE_OF_ARRAY,127)
image = imgProc.getImageForHopfield().copy()
#print image
F.add(npyscreen.TitleFixedText, name = "Image to learn: " + imageName,)
F.display()
images.append(image)
netHebb.trainHebb(image)
netDelta.trainDelta(image)
netPseudoInv.appendTrainingVectorForPseudoInversion(image)
netPseudoInv.trainPseudoInversion()
F.add(npyscreen.TitleFixedText, name = "Networks teached by above images",)
F.display()
imgProc = ImgPreprocessor(imgToRecogn,SIDE_OF_ARRAY,127)
imageRec = imgProc.getImageForHopfield()
netHebb.initNeurons(imageRec)
netDelta.initNeurons(imageRec)
netPseudoInv.initNeurons(imageRec)
hebbProgressWidget = F.add(npyscreen.TitleSlider, out_of=100, name = "Hebb")
F.add(npyscreen.TitleFixedText, name = " ",)
deltaProgressWidget = F.add(npyscreen.TitleSlider, out_of=100, name = "Delta")
F.add(npyscreen.TitleFixedText, name = " ",)
pInvProgressWidget = F.add(npyscreen.TitleSlider, out_of=100, name = "PInv")
for i in range(50):
netHebb.update(100)
hebbProgressWidget.value = 2 + i*2
F.display()
for i in range(50):
netDelta.update(100)
deltaProgressWidget.value = 2 + i*2
F.display()
for i in range(50):
netPseudoInv.update(100)
pInvProgressWidget.value = 2 + i*2
F.display()
networkResultHebb = netHebb.getNeuronsMatrix().copy()
networkResultDelta = netDelta.getNeuronsMatrix().copy()
networkResultPseudoInv = netPseudoInv.getNeuronsMatrix().copy()
#netHebb.saveNetworkConfigToFile('networkConfig.txt');
"""
#print "================================"
matchFound = False
for idx, image in enumerate(images):
if np.array_equiv(networkResultHebb, image):
#print "Found matching image as sample " + str(idx)
matchFound = True
break
elif np.array_equiv(self.invertMatrix(networkResultHebb), image):
#print "Found inverse matching image as sample" + str(idx)
matchFound = True
break
if(not matchFound):
pass
#print "Unfortunately no match!"
#print "================================"
"""
subplotIndex = 201
subplotIndex += 10 * (len(images)+1)
for idx, image in enumerate(images):
plt.subplot(subplotIndex+idx)
image2d = np.reshape(image, (SIDE_OF_ARRAY,SIDE_OF_ARRAY))
plt.imshow(image2d, cmap='gray', interpolation = 'nearest')
plt.xticks([]), plt.yticks([]), plt.title("Sample %d" % idx)
newLineSubplotIndex = subplotIndex + len(images)
plt.subplot(newLineSubplotIndex)
image2d = np.reshape(imageRec, (SIDE_OF_ARRAY,SIDE_OF_ARRAY))
plt.imshow(image2d, cmap='gray', interpolation = 'nearest')
plt.xticks([]), plt.yticks([]), plt.title("To recognize")
plt.subplot(newLineSubplotIndex+1)
image2dResult = np.reshape(networkResultHebb , (SIDE_OF_ARRAY,SIDE_OF_ARRAY))
plt.imshow(image2dResult, cmap='gray', interpolation = 'nearest')
plt.xticks([]), plt.yticks([]), plt.title("Returned by net (HEBB)")
plt.subplot(newLineSubplotIndex+2)
image2dResult = np.reshape(networkResultDelta, (SIDE_OF_ARRAY,SIDE_OF_ARRAY))
plt.imshow(image2dResult, cmap='gray', interpolation = 'nearest')
plt.xticks([]), plt.yticks([]), plt.title("Returned by net (DELTA)")
plt.subplot(newLineSubplotIndex+3)
image2dResult = np.reshape(networkResultPseudoInv, (SIDE_OF_ARRAY,SIDE_OF_ARRAY))
plt.imshow(image2dResult, cmap='gray', interpolation = 'nearest')
plt.xticks([]), plt.yticks([]), plt.title("Returned by net (PINVERSE)")
plt.show()
