def importCatDogANN(fileName=root.path() + "/res/recCatDogANN"):
n = FeedForwardNetwork()
n.addInputModule(LinearLayer(7500, name='in'))
n.addModule(SigmoidLayer(9000, name='hidden'))
n.addOutputModule(LinearLayer(2, name='out'))
n.addConnection(FullConnection(n['in'], n['hidden'], name='c1'))
n.addConnection(FullConnection(n['hidden'], n['out'], name='c2'))
n.sortModules()
params = np.load(root.path() + '/res/cat_dog_params.txt.npy')
n._setParameters(params)
return n
def importCatDogANN(fileName = root.path()+"/res/recCatDogANN"):
n = FeedForwardNetwork()
n.addInputModule(LinearLayer(7500, name='in'))
n.addModule(SigmoidLayer(9000, name='hidden'))
n.addOutputModule(LinearLayer(2, name='out'))
n.addConnection(FullConnection(n['in'], n['hidden'], name='c1'))
n.addConnection(FullConnection(n['hidden'], n['out'], name='c2'))
n.sortModules()
params = np.load(root.path()+'/res/cat_dog_params.txt.npy')
n._setParameters(params)
return n
def trainedANN():
n = FeedForwardNetwork()
n.addInputModule(LinearLayer(4, name='in'))
n.addModule(SigmoidLayer(6, name='hidden'))
n.addOutputModule(LinearLayer(2, name='out'))
n.addConnection(FullConnection(n['in'], n['hidden'], name='c1'))
n.addConnection(FullConnection(n['hidden'], n['out'], name='c2'))
n.sortModules()
draw_connections(n)
# d = generateTrainingData()
d = getDatasetFromFile(root.path() + "/res/dataSet")
t = BackpropTrainer(n, d, learningrate=0.001, momentum=0.75)
t.trainOnDataset(d)
# FIXME: I'm not sure the recurrent ANN is going to converge
# so just training for fixed number of epochs
count = 0
while True:
globErr = t.train()
print globErr
if globErr < 0.01:
break
count += 1
if count == 20:
return trainedANN()
exportANN(n)
draw_connections(n)
return n
def trainedRNN():
n = RecurrentNetwork()
n.addInputModule(LinearLayer(4, name='in'))
n.addModule(SigmoidLayer(6, name='hidden'))
n.addOutputModule(LinearLayer(2, name='out'))
n.addConnection(FullConnection(n['in'], n['hidden'], name='c1'))
n.addConnection(FullConnection(n['hidden'], n['out'], name='c2'))
n.addRecurrentConnection(NMConnection(n['out'], n['out'], name='nmc'))
# n.addRecurrentConnection(FullConnection(n['out'], n['hidden'], inSliceFrom = 0, inSliceTo = 1, outSliceFrom = 0, outSliceTo = 3))
n.sortModules()
draw_connections(n)
d = getDatasetFromFile(root.path() + "/res/dataSet")
t = BackpropTrainer(n, d, learningrate=0.001, momentum=0.75)
t.trainOnDataset(d)
count = 0
while True:
globErr = t.train()
print globErr
if globErr < 0.01:
break
count += 1
if count == 50:
return trainedRNN()
# exportRNN(n)
draw_connections(n)
return n
def trainedRNN():
n = RecurrentNetwork()
n.addInputModule(LinearLayer(4, name='in'))
n.addModule(SigmoidLayer(6, name='hidden'))
n.addOutputModule(LinearLayer(2, name='out'))
n.addConnection(FullConnection(n['in'], n['hidden'], name='c1'))
n.addConnection(FullConnection(n['hidden'], n['out'], name='c2'))
n.addRecurrentConnection(NMConnection(n['out'], n['out'], name='nmc'))
# n.addRecurrentConnection(FullConnection(n['out'], n['hidden'], inSliceFrom = 0, inSliceTo = 1, outSliceFrom = 0, outSliceTo = 3))
n.sortModules()
draw_connections(n)
d = getDatasetFromFile(root.path()+"/res/dataSet")
t = BackpropTrainer(n, d, learningrate=0.001, momentum=0.75)
t.trainOnDataset(d)
count = 0
while True:
globErr = t.train()
print globErr
if globErr < 0.01:
break
count += 1
if count == 50:
return trainedRNN()
# exportRNN(n)
draw_connections(n)
return n
def trainedANN():
n = FeedForwardNetwork()
n.addInputModule(LinearLayer(4, name='in'))
n.addModule(SigmoidLayer(6, name='hidden'))
n.addOutputModule(LinearLayer(2, name='out'))
n.addConnection(FullConnection(n['in'], n['hidden'], name='c1'))
n.addConnection(FullConnection(n['hidden'], n['out'], name='c2'))
n.sortModules()
draw_connections(n)
# d = generateTrainingData()
d = getDatasetFromFile(root.path()+"/res/dataSet")
t = BackpropTrainer(n, d, learningrate=0.001, momentum=0.75)
t.trainOnDataset(d)
# FIXME: I'm not sure the recurrent ANN is going to converge
# so just training for fixed number of epochs
count = 0
while True:
globErr = t.train()
print globErr
if globErr < 0.01:
break
count += 1
if count == 20:
return trainedANN()
exportANN(n)
draw_connections(n)
return n
def exportCatDogRFCNN(net, fileName = root.path()+"/res/cat_dog_fc_params"):
# arr = net.params
# np.save(fileName, arr)
# fileObject = open(fileName+'.pickle', 'w')
# pickle.dump(net, fileObject)
# fileObject.close()
NetworkWriter.writeToFile(net, fileName+'.xml')
def exportCatDogRNN(net, fileName=root.path() + "/res/cat_dog_nm_params"):
# arr = net.params
# np.save(fileName, arr)
# fileObject = open(fileName+'.pickle', 'w')
# pickle.dump(net, fileObject)
# fileObject.close()
NetworkWriter.writeToFile(net, fileName + '.xml')
def load_mnist():
"""
Loads MNIST files into 3D numpy arrays
Adapted from: http://abel.ee.ucla.edu/cvxopt/_downloads/mnist.py
"""
path = str(root.path()) + '/res/mnist.pkl.gz'
f = gzip.open(path, 'rb')
train_set, valid_set, test_set = cPickle.load(f)
f.close()
return train_set, valid_set, test_set
def create1OrderDataSet():
lab_images = get_train_set(instance=False, number_of_instances=10)
ds = SupervisedDataSet(100, 1)
for i in range(len(lab_images)):
data = np.zeros((100))
for j in range(100):
data[j] = lab_images[i][0][j]
ds.addSample(data, lab_images[i][1])
print "creating dataset, iteration:",i,"of",len(lab_images)
ds.saveToFile(root.path() + '/res/dataset1')
return ds
def create1OrderDataSet():
lab_images = get_train_set(instance=False, number_of_instances=10)
ds = SupervisedDataSet(100, 1)
for i in range(len(lab_images)):
data = np.zeros((100))
for j in range(100):
data[j] = lab_images[i][0][j]
ds.addSample(data, lab_images[i][1])
print "creating dataset, iteration:", i, "of", len(lab_images)
ds.saveToFile(root.path() + '/res/dataset1')
return ds
def load_mnist():
"""
Loads MNIST files into 3D numpy arrays
Adapted from: http://abel.ee.ucla.edu/cvxopt/_downloads/mnist.py
"""
path = str(root.path()) + '/res/mnist.pkl.gz'
f = gzip.open(path, 'rb')
train_set, valid_set, test_set = cPickle.load(f)
f.close()
return train_set, valid_set, test_set
def create2OrderDataSet():
lab_images = get_train_set(instance=True)
ds = SupervisedDataSet(5150, 1)
for i in range(len(lab_images)):
data = np.zeros((5150))
for j in range(100):
data[j] = lab_images[i][0][j]
count = 100
for x1 in range(100):
for x2 in range(x1, 100):
# print count
data[count] = lab_images[i][0][x1]*lab_images[i][0][x2]
count += 1
ds.addSample(data, lab_images[i][1])
print "creating dataset, iteration:",i,"of",len(lab_images)
ds.saveToFile(root.path() + '/res/dataset2')
return ds
def create2OrderDataSet():
lab_images = get_train_set(instance=True)
ds = SupervisedDataSet(5150, 1)
for i in range(len(lab_images)):
data = np.zeros((5150))
for j in range(100):
data[j] = lab_images[i][0][j]
count = 100
for x1 in range(100):
for x2 in range(x1, 100):
# print count
data[count] = lab_images[i][0][x1] * lab_images[i][0][x2]
count += 1
ds.addSample(data, lab_images[i][1])
print "creating dataset, iteration:", i, "of", len(lab_images)
ds.saveToFile(root.path() + '/res/dataset2')
return ds
def generateTrainingData(size=10000, saveAfter=False):
"""
Creates a set of training data with 4-dimensioanal input and 2-dimensional output
with `size` samples
"""
np.random.seed()
data = SupervisedDataSet(4, 2)
for i in xrange(1, int(size / 2)):
[a, b] = np.random.random_integers(1, 100, 2)
[c, d] = np.random.random_integers(100, 500, 2)
data.addSample((a, b, c, d), (-1, 1))
for i in xrange(1, int(size / 2)):
[a, b] = np.random.random_integers(100, 500, 2)
[c, d] = np.random.random_integers(1, 100, 2)
data.addSample((a, b, c, d), (1, -1))
if saveAfter:
data.saveToFile(root.path() + "/res/dataSet")
return data
def generateTrainingData(size=10000, saveAfter = False):
"""
Creates a set of training data with 4-dimensioanal input and 2-dimensional output
with `size` samples
"""
np.random.seed()
data = SupervisedDataSet(4,2)
for i in xrange(1, int(size/2)):
[a, b] = np.random.random_integers(1, 100, 2)
[c, d] = np.random.random_integers(100, 500, 2)
data.addSample((a, b, c, d), (-1, 1))
for i in xrange(1, int(size/2)):
[a, b] = np.random.random_integers(100, 500, 2)
[c, d] = np.random.random_integers(1, 100, 2)
data.addSample((a, b, c, d), (1, -1))
if saveAfter:
data.saveToFile(root.path()+"/res/dataSet")
return data
def get_cat_dog_testset():
count = 0
images = os.listdir(root.path() + '/res/cats_proc/')
shape = cv2.imread(root.path() + '/res/cats_proc/'+images[0],0).shape
ds = SupervisedDataSet(shape[0]*shape[1], 2)
for image in os.listdir(root.path() + '/res/cats_proc/'):
img = cv2.imread(root.path() + '/res/cats_proc/'+image,0)
inp = np.reshape(img, shape[0]*shape[1])
target = [1,0]
ds.addSample(inp, target)
count += 1
for image in os.listdir(root.path() + '/res/dogs_proc/'):
img = cv2.imread(root.path() + '/res/dogs_proc/'+image,0)
img = cv2.resize(img, img.shape, fx=0.5, fy=0.5)
inp = np.reshape(img, shape[0]*shape[1])
target = [0,1]
ds.addSample(inp, target)
count += 1
return ds
def load3OrderDataSet():
ds = SupervisedDataSet.loadFromFile(root.path() + '/res/dataset3')
return ds
def exportCatDogANN(net, fileName = root.path()+"/res/cat_dog_params"):
arr = net.params
np.save(fileName, arr)
def importRFCNN(fileName = root.path()+"/res/recRFCNN"):
fileObject = open(fileName, 'r')
net = pickle.load(fileObject)
fileObject.close()
return net
def exportRFCNN(net, fileName = root.path()+"/res/recRFCNN"):
fileObject = open(fileName, 'w')
pickle.dump(net, fileObject)
fileObject.close()
def load3OrderDataSet():
ds = SupervisedDataSet.loadFromFile(root.path() + '/res/dataset3')
return ds
def importCatDogRNN(fileName=root.path() + "/res/recCatDogANN"):
n = NetworkReader.readFrom(root.path() + "/res/cat_dog_nm_params.xml")
return n
def importCatDogRNN(fileName = root.path()+"/res/recCatDogANN"):
n = NetworkReader.readFrom(root.path()+"/res/cat_dog_nm_params.xml")
return n
def exportRFCNN(net, fileName=root.path() + "/res/recRFCNN"):
fileObject = open(fileName, 'w')
pickle.dump(net, fileObject)
fileObject.close()
def exportCatDogANN(net, fileName=root.path() + "/res/cat_dog_params"):
arr = net.params
np.save(fileName, arr)
def importRFCNN(fileName=root.path() + "/res/recRFCNN"):
fileObject = open(fileName, 'r')
net = pickle.load(fileObject)
fileObject.close()
return net
