def test_composite(self):
ls = MS.GradientDescent(lr=0.1)
cost = MC.NegativeLogLikelihood()
inp = ML.Input(2, 'inp')
h1 = ML.Hidden(5, activation=MA.Tanh(), name="h1")
h2 = ML.Hidden(5, activation=MA.Tanh(), name="h2")
o = ML.SoftmaxClassifier(2,
learningScenario=ls,
costObject=cost,
name="out")
c = ML.Composite(name="Comp")
inp > h1 > c
inp > h2 > c
mlp = c > o
for i in xrange(10000):
ii = i % len(self.xor_ins)
mlp.train(o, inp=[self.xor_ins[ii]], targets=[self.xor_outs[ii]])
self.assertEqual(mlp.predict(o, inp=[self.xor_ins[0]])["class"], 0)
self.assertEqual(mlp.predict(o, inp=[self.xor_ins[1]])["class"], 1)
self.assertEqual(mlp.predict(o, inp=[self.xor_ins[2]])["class"], 1)
self.assertEqual(mlp.predict(o, inp=[self.xor_ins[3]])["class"], 0)
def test_concatenation(self):
ls = MS.GradientDescent(lr=0.1)
cost = MC.NegativeLogLikelihood()
inp = ML.Input(2, 'inp')
h1 = ML.Hidden(5, activation=MA.Tanh(), name="h1")
h2 = ML.Hidden(5, activation=MA.Tanh(), name="h2")
o = ML.SoftmaxClassifier(nbClasses=2,
cost=cost,
learningScenari=[ls],
name="out")
inp > h1
inp > h2
c = ML.C([h1, h2], name="concat")
mlp = c > o
mlp.init()
self.assertEqual(c.getIntrinsicShape()[0],
h1.getIntrinsicShape()[0] + h2.getIntrinsicShape()[0])
for i in xrange(10000):
ii = i % len(self.xor_ins)
miniBatch = [self.xor_ins[ii]]
targets = [self.xor_outs[ii]]
mlp["out"].train({
"inp.inputs": miniBatch,
"out.targets": targets
})["out.drive.train"]
for i in xrange(len(self.xor_ins)):
self.assertEqual(
mlp["out"].predict["test"]({
"inp.inputs": [self.xor_ins[i]]
})["out.predict.test"], self.xor_outs[i])
def __init__(self, ls, cost):
maxPool = MCONV.MaxPooling2D(2, 2)
#The input channeler will take regular layers and arrange them into several channels
i = MCONV.Input(nbChannels=3, height=256, width=256, name='inp')
#ichan = MCONV.InputChanneler(256, 256, name = 'inpChan')
c1 = MCONV.Convolution2D(nbFilters=3,
filterHeight=5,
filterWidth=5,
activation=MA.Tanh(),
pooler=maxPool,
name="conv1")
c2 = MCONV.Convolution2D(nbFilters=3,
filterHeight=5,
filterWidth=5,
activation=MA.Tanh(),
pooler=maxPool,
name="conv2")
f = MCONV.Flatten(name="flat")
h = ML.Hidden(5,
activation=MA.Tanh(),
decorators=[MD.GlorotTanhInit()],
regularizations=[MR.L1(0), MR.L2(0.0001)],
name="hid")
o = ML.SoftmaxClassifier(2,
decorators=[],
learningScenario=ls,
costObject=cost,
name="out",
regularizations=[])
self.model = i > c1 > c2 > f > h > o
def ae2(data):
"""This one uses an Autoencode layer. This layer is a part of the graph and does not need a specific traget"""
miniBatchSize = 1
ls = MS.GradientDescent(lr=0.1)
cost = MC.MeanSquaredError()
i = ML.Input(8, name='inp')
h = ML.Hidden(3,
activation=MA.ReLU(),
initializations=[MI.SmallUniformWeights(),
MI.ZeroBias()],
name="hid")
o = ML.Autoencode(
i.name,
activation=MA.ReLU(),
initializations=[MI.SmallUniformWeights(),
MI.ZeroBias()],
learningScenario=ls,
costObject=cost,
name="out")
ae = i > h > o
# ae.init()
# o.train.printGraph()
for e in xrange(2000):
for i in xrange(0, len(data), miniBatchSize):
ae.train(o, inp=data[i:i + miniBatchSize])
return ae, o
def getModel(inpSize, filterWidth):
ls = MS.GradientDescent(lr=0.5)
cost = MC.NegativeLogLikelihood()
i = ML.Input((1, 1, inpSize), name='inp')
c1 = MCONV.Convolution2D(numFilters=5,
filterHeight=1,
filterWidth=filterWidth,
activation=MA.ReLU(),
name="conv1")
pool1 = MSAMP.MaxPooling2D(poolHeight=1, poolWidth=2, name="pool1")
c2 = MCONV.Convolution2D(numFilters=10,
filterHeight=1,
filterWidth=filterWidth,
activation=MA.ReLU(),
name="conv2")
pool2 = MSAMP.MaxPooling2D(poolHeight=1, poolWidth=2, name="pool2")
h = ML.Hidden(5, activation=MA.ReLU(), name="hid")
o = ML.SoftmaxClassifier(nbClasses=2,
cost=cost,
learningScenari=[ls],
name="out")
model = i > c1 > pool1 > c2 > pool2 > h > o
model.init()
return model
def test_multiinputs(self):
ls = MS.GradientDescent(lr=0.1)
inpA = ML.Embedding(2, 2, 2, name="IA")
inpB = ML.Input(2, name="IB")
inpNexus = ML.Composite(name="InputNexus")
h1 = ML.Hidden(32,
activation=MA.ReLU(),
decorators=[],
regularizations=[],
name="Fully-connected1")
o = ML.Regression(2,
decorators=[],
activation=MA.ReLU(),
learningScenario=ls,
costObject=MC.CrossEntropy(),
name="Out",
regularizations=[])
inpA > inpNexus
inpB > inpNexus
m = inpNexus > h1 > o
m.init()
def __init__(self, ls, cost):
maxPool = MCONV.MaxPooling2D(2, 2)
i = MCONV.Input(nbChannels=1, height=28, width=28, name='inp')
c1 = MCONV.Convolution2D(nbFilters=20,
filterHeight=5,
filterWidth=5,
activation=MA.Tanh(),
pooler=maxPool,
name="conv1")
c2 = MCONV.Convolution2D(nbFilters=50,
filterHeight=5,
filterWidth=5,
activation=MA.Tanh(),
pooler=maxPool,
name="conv2")
#needed for the transition to a fully connected layer
f = MCONV.Flatten(name="flat")
h = ML.Hidden(500,
activation=MA.Tanh(),
decorators=[],
regularizations=[],
name="hid")
o = ML.SoftmaxClassifier(10,
decorators=[],
learningScenario=ls,
costObject=cost,
name="out",
regularizations=[])
self.model = i > c1 > c2 > f > h > o
print self.model
def test_ae(self) :
data = []
for i in xrange(8) :
zeros = numpy.zeros(8)
zeros[i] = 1
data.append(zeros)
ls = MS.GradientDescent(lr = 0.1)
cost = MC.MeanSquaredError()
i = ML.Input(8, name = 'inp')
h = ML.Hidden(3, activation = MA.ReLU(), initializations=[MI.SmallUniformWeights(), MI.ZeroBias()], name = "hid")
o = ML.Autoencode(targetLayerName = "inp", activation = MA.ReLU(), initializations=[MI.SmallUniformWeights(), MI.ZeroBias()], learningScenario = ls, costObject = cost, name = "out" )
ae = i > h > o
miniBatchSize = 1
for e in xrange(2000) :
for i in xrange(0, len(data), miniBatchSize) :
ae.train(o, inp = data[i:i+miniBatchSize])
res = ae.propagate(o, inp = data)["outputs"]
for i in xrange(len(res)) :
self.assertEqual( numpy.argmax(data[i]), numpy.argmax(res[i]))
def test_ae(self):
data = []
for i in xrange(8):
zeros = numpy.zeros(8)
zeros[i] = 1
data.append(zeros)
ls = MS.GradientDescent(lr=0.1)
cost = MC.MeanSquaredError()
i = ML.Input(8, name='inp')
h = ML.Hidden(3, activation=MA.ReLU(), name="hid")
o = ML.Regression(8,
activation=MA.ReLU(),
learningScenario=ls,
costObject=cost,
name="out")
ae = i > h > o
miniBatchSize = 2
for e in xrange(2000):
for i in xrange(0, len(data), miniBatchSize):
ae.train(o,
inp=data[i:i + miniBatchSize],
targets=data[i:i + miniBatchSize])
res = ae.propagate(o, inp=data)["outputs"]
for i in xrange(len(res)):
self.assertEqual(numpy.argmax(data[i]), numpy.argmax(res[i]))
def getModel(inpSize, filterWidth) :
ls = MS.GradientDescent(lr = 0.5)
cost = MC.NegativeLogLikelihood()
pooler = MCONV.MaxPooling2D(1, 2)
i = ML.Input(inpSize, name = 'inp')
ichan = MCONV.InputChanneler(1, inpSize, name = 'inpChan')
c1 = MCONV.Convolution2D(
nbFilters = 5,
filterHeight = 1,
filterWidth = filterWidth,
activation = MA.ReLU(),
pooler = pooler,
name = "conv1"
)
c2 = MCONV.Convolution2D(
nbFilters = 10,
filterHeight = 1,
filterWidth = filterWidth,
activation = MA.ReLU(),
pooler = pooler,
name = "conv2"
)
f = MCONV.Flatten(name = "flat")
h = ML.Hidden(5, activation = MA.ReLU(), decorators = [], regularizations = [ ], name = "hid" )
o = ML.SoftmaxClassifier(2, decorators = [], learningScenario = ls, costObject = cost, name = "out", regularizations = [ ] )
model = i > ichan > c1 > c2 > f > h > o
return model
def ae1(data):
'''Using a regression layer. This layer needs an explicit target'''
miniBatchSize = 2
ls = MS.GradientDescent(lr=0.1)
cost = MC.MeanSquaredError()
i = ML.Input(8, name='inp')
h = ML.Hidden(3,
activation=MA.ReLU(),
initializations=[MI.SmallUniformWeights(),
MI.ZeroBias()],
name="hid")
o = ML.Regression(
8,
activation=MA.ReLU(),
initializations=[MI.SmallUniformWeights(),
MI.ZeroBias()],
learningScenario=ls,
costObject=cost,
name="out")
ae = i > h > o
for e in xrange(1000):
for i in xrange(0, len(data), miniBatchSize):
ae.train(o,
inp=data[i:i + miniBatchSize],
targets=data[i:i + miniBatchSize])
return ae, o
def test_save_load_64h(self):
import os
import Mariana.network as MN
ls = MS.GradientDescent(lr=0.1)
cost = MC.NegativeLogLikelihood()
i = ML.Input(2, 'inp')
o = ML.SoftmaxClassifier(nbClasses=2,
cost=cost,
learningScenari=[ls],
name="out")
prev = i
for i in xrange(64):
h = ML.Hidden(size=10, activation=MA.ReLU(), name="Hidden_%s" % i)
prev > h
prev = h
mlp = prev > o
mlp.init()
mlp.save("test_save")
mlp2 = MN.loadModel("test_save.mar")
mlp2.init()
v1 = mlp["out"].propagate["test"]({
"inp.inputs": self.xor_ins
})["out.propagate.test"]
v2 = mlp2["out"].propagate["test"]({
"inp.inputs": self.xor_ins
})["out.propagate.test"]
self.assertTrue((v1 == v2).all())
os.remove('test_save.mar')
def test_optimizer_override(self):
ls = MS.GradientDescent(lr=0.5)
cost = MC.NegativeLogLikelihood()
inp = ML.Input(1, 'inp')
h = ML.Hidden(5,
activation=MA.Tanh(),
learningScenari=[MS.Fixed("b")],
name="h")
o = ML.SoftmaxClassifier(
2,
learningScenari=[MS.GradientDescent(lr=0.5),
MS.Fixed("W")],
cost=cost,
name="out")
net = inp > h > o
net.init()
ow = o.getP('W').getValue()
ob = o.getP('b').getValue()
hw = h.getP('W').getValue()
hb = h.getP('b').getValue()
for x in xrange(1, 10):
net["out"].train({
"inp.inputs": [[1]],
"out.targets": [1]
})["out.drive.train"]
self.assertTrue(sum(ow[0]) == sum(o.getP('W').getValue()[0]))
self.assertTrue(sum(ob) != sum(o.getP('b').getValue()))
self.assertTrue(sum(hb) == sum(h.getP('b').getValue()))
self.assertTrue(sum(hw[0]) != sum(h.getP('W').getValue()[0]))
def getMLP(self, nbInputs=2, nbClasses=2):
ls = MS.GradientDescent(lr=0.1)
cost = MC.NegativeLogLikelihood()
i = ML.Input(nbInputs, 'inp')
h = ML.Hidden(size=6, activation=MA.ReLU(), name="Hidden_0.500705866892")
o = ML.SoftmaxClassifier(nbClasses=nbClasses,
cost=cost,
learningScenari=[ls],
name="out")
mlp = i > h > o
mlp.init()
return mlp
def test_ae_reg(self):
powerOf2 = 3
nbUnits = 2**powerOf2
data = []
for i in xrange(nbUnits):
zeros = numpy.zeros(nbUnits)
zeros[i] = 1
data.append(zeros)
ls = MS.GradientDescent(lr=0.1)
cost = MC.MeanSquaredError()
i = ML.Input(nbUnits, name='inp')
h = ML.Hidden(powerOf2,
activation=MA.ReLU(),
initializations=[
MI.Uniform('W', small=True),
MI.SingleValue('b', 0)
],
name="hid")
o = ML.Regression(nbUnits,
activation=MA.ReLU(),
initializations=[
MI.Uniform('W', small=True),
MI.SingleValue('b', 0)
],
learningScenari=[ls],
cost=cost,
name="out")
ae = i > h > o
ae.init()
miniBatchSize = 1
for e in xrange(2000):
for i in xrange(0, len(data), miniBatchSize):
miniBatch = data[i:i + miniBatchSize]
ae["out"].train({
"inp.inputs": miniBatch,
"out.targets": miniBatch
})["out.drive.train"]
res = ae["out"].propagate["test"]({
"inp.inputs": data
})["out.propagate.test"]
for i in xrange(len(res)):
self.assertEqual(numpy.argmax(data[i]), numpy.argmax(res[i]))
def trainMLP_xor(self) :
ls = MS.GradientDescent(lr = 0.1)
cost = MC.NegativeLogLikelihood()
i = ML.Input(2, 'inp')
h = ML.Hidden(10, activation = MA.ReLU(), name = "Hidden_0.500705866892")
o = ML.SoftmaxClassifier(2, learningScenario = ls, costObject = cost, name = "out")
mlp = i > h > o
self.xor_ins = numpy.array(self.xor_ins)
self.xor_outs = numpy.array(self.xor_outs)
for i in xrange(1000) :
mlp.train(o, inp = self.xor_ins, targets = self.xor_outs )
return mlp
def MLP(ls, cost):
i = ML.Input(28 * 28, name='inp')
h = ML.Hidden(500,
activation=MA.Tanh(),
decorators=[MD.GlorotTanhInit()],
regularizations=[MR.L1(0), MR.L2(0.0001)],
name="hid")
o = ML.SoftmaxClassifier(10,
decorators=[MD.ZerosInit()],
learningScenario=ls,
costObject=cost,
name="out",
regularizations=[MR.L1(0),
MR.L2(0.0001)])
mlp = i > h > o
return mlp
def __init__(self, ls, cost):
maxPool = MCONV.MaxPooling2D(3, 3)
i = MCONV.Input(nbChannels=1, height=100, width=100, name='inp')
c1 = MCONV.Convolution2D(nbFilters=10,
filterHeight=3,
filterWidth=3,
activation=MA.Max_norm(),
pooler=maxPool,
name="conv1")
c3 = MCONV.Convolution2D(nbFilters=20,
filterHeight=3,
filterWidth=3,
activation=MA.Max_norm(),
pooler=maxPool,
name="conv3")
c2 = MCONV.Convolution2D(nbFilters=10,
filterHeight=3,
filterWidth=3,
activation=MA.Max_norm(),
pooler=maxPool,
name="conv2")
f = MCONV.Flatten(name="flat")
h = ML.Hidden(2048,
activation=MA.Max_norm(),
decorators=[MD.BinomialDropout(0.7)],
regularizations=[],
name="hid")
o = ML.SoftmaxClassifier(2,
decorators=[],
learningScenario=ls,
costObject=cost,
name="out",
regularizations=[])
self.model = i > c1 > c3 > c2 > f > h > o
def trainMLP_xor(self):
ls = MS.GradientDescent(lr=0.1)
cost = MC.NegativeLogLikelihood()
i = ML.Input(2, 'inp')
h = ML.Hidden(4,
activation=MA.Tanh(),
decorators=[dec.GlorotTanhInit()],
regularizations=[MR.L1(0), MR.L2(0)])
o = ML.SoftmaxClassifier(2,
learningScenario=ls,
costObject=cost,
name="out")
mlp = i > h > o
self.xor_ins = numpy.array(self.xor_ins)
self.xor_outs = numpy.array(self.xor_outs)
for i in xrange(1000):
ii = i % len(self.xor_ins)
mlp.train(o, inp=[self.xor_ins[ii]], targets=[self.xor_outs[ii]])
return mlp
def __init__(self, inputSize, dictSize, patternSize, embSize, ls, cost):
# pooler = MCONV.NoPooling()
pooler = MCONV.MaxPooling2D(1, 2)
emb = MCONV.Embedding(size=inputSize,
nbDimentions=embSize,
dictSize=dictSize,
name='Emb')
c1 = MCONV.Convolution2D(nbFilters=1,
filterHeight=1,
filterWidth=patternSize / 2,
activation=MA.ReLU(),
pooler=pooler,
name="conv1")
c2 = MCONV.Convolution2D(nbFilters=4,
filterHeight=1,
filterWidth=patternSize / 2,
activation=MA.ReLU(),
pooler=MCONV.NoPooling(),
name="conv2")
f = MCONV.Flatten(name="flat")
h = ML.Hidden(5,
activation=MA.ReLU(),
decorators=[],
regularizations=[],
name="hid")
o = ML.SoftmaxClassifier(2,
decorators=[],
learningScenario=ls,
costObject=cost,
name="out",
regularizations=[])
self.model = emb > c1 > c2 > f > h > o
def __init__(self, inputSize, patternSize, ls, cost):
# pooler = MCONV.NoPooling()
pooler = MCONV.MaxPooling2D(1, 2)
#The input channeler will take regular layers and arrange them into several channels
i = ML.Input(inputSize, name='inp')
ichan = MCONV.InputChanneler(1, inputSize, name='inpChan')
c1 = MCONV.Convolution2D(nbFilters=5,
filterHeight=1,
filterWidth=patternSize / 2,
activation=MA.ReLU(),
pooler=pooler,
name="conv1")
c2 = MCONV.Convolution2D(nbFilters=10,
filterHeight=1,
filterWidth=patternSize / 2,
activation=MA.ReLU(),
pooler=MCONV.NoPooling(),
name="conv2")
f = MCONV.Flatten(name="flat")
h = ML.Hidden(5,
activation=MA.ReLU(),
decorators=[],
regularizations=[],
name="hid")
o = ML.SoftmaxClassifier(2,
decorators=[],
learningScenario=ls,
costObject=cost,
name="out",
regularizations=[])
self.model = i > ichan > c1 > c2 > f > h > o
* automatically saves the model if the training halts because of an error or if the process is killed
* saves a log if the process dies unexpectedly
* training results and hyper parameters values are recorded to a file
* allows you to define custom stop criteria
* training info is printed at each epoch, including best scores and at which epoch they were achieved
"""
if __name__ == "__main__":
# Let's define the network
ls = MS.GradientDescent(lr=0.01)
cost = MC.NegativeLogLikelihood()
i = ML.Input(28 * 28, name='inp')
h = ML.Hidden(500, activation=MA.Tanh(), decorators=[MD.GlorotTanhInit()], regularizations=[MR.L1(0), MR.L2(0.0001)], name="hid")
o = ML.SoftmaxClassifier(10, learningScenario=ls, costObject=cost, name="out", regularizations=[MR.L1(0), MR.L2(0.0001)])
mlp = i > h > o
mlp.saveDOT("mnist_mlp")
mlp.saveHTML("mnist_mlp")
# And then map sets to the inputs and outputs of our network
train_set, validation_set, test_set = load_mnist()
trainData = MDM.Series(images=train_set[0], numbers=train_set[1])
trainMaps = MDM.DatasetMapper()
trainMaps.mapInput(i, trainData.images)
trainMaps.mapOutput(o, trainData.numbers)
testData = MDM.Series(images=test_set[0], numbers=test_set[1])
data[i][start:start+patternSize] = patternC2
targets.append(2)
else :
targets.append(0)
targets = numpy.asarray(targets, dtype=theano.config.floatX)
return data, targets
if __name__ == "__main__" :
examples, targets = makeDataset(300, 100, 10)
ls = MS.GradientDescent(lr = 0.01)
cost = MC.NegativeLogLikelihood()
i = ML.Input(100, 'inp')
h1 = ML.Hidden(50, activation = MA.ReLU(), decorators = [MD.BatchNormalization()])
h2 = ML.Hidden(2, activation = MA.Softmax())
o = ML.SoftmaxClassifier(3, learningScenario = ls, costObject = cost, name = "out")
mlp = i > h1 > h2 > o
for k in xrange(100) :
for example, target in zip(examples, targets) :
mlp.train(o, inp=[example], targets=[target])
nbErr = 0
for example, target in zip(examples, targets) :
if target != mlp.classify(o, inp=[example])["class"] :
nbErr += 1
print "Nb Errors: %s/%s (%s%%) " % (nbErr, len(targets), float(nbErr)/len(targets) * 100)
* automatically saves the model if the training halts because of an error or if the process is killed
* saves a log if the process dies unexpectedly
* training results and hyper parameters values are recorded to a file
* allows you to define custom stop criteria
* training info is printed at each epoch, including best scores and at which epoch they were achieved
"""
if __name__ == "__main__":
# Let's define the network
ls = MS.GradientDescent(lr=0.01)
cost = MC.NegativeLogLikelihood()
i = ML.Input(28 * 28, name='inp')
h = ML.Hidden(500, activation=MA.Tanh(), initializations=[MI.GlorotTanhInit(), MI.ZeroBias()], regularizations=[MR.L1(0), MR.L2(0.0001)], name="hid")
o = ML.SoftmaxClassifier(10, learningScenario=ls, costObject=cost, name="out", regularizations=[MR.L1(0), MR.L2(0.0001)])
mlp = i > h > o
mlp.saveDOT("mnist_mlp")
mlp.saveHTML("mnist_mlp")
# And then map sets to the inputs and outputs of our network
train_set, validation_set, test_set = load_mnist()
trainData = MDM.Series(images=train_set[0], numbers=train_set[1])
trainMaps = MDM.DatasetMapper("train", miniBatchSize=500)
trainMaps.mapInput(i, trainData.images)
trainMaps.mapOutput(o, trainData.numbers)
testData = MDM.Series(images=test_set[0], numbers=test_set[1])
name = "conv3" ) c2 = MCONV.Convolution2D( nbFilters = 15, filterHeight = 3, filterWidth = 3, activation = MA.Max_norm(), pooler = maxPool, name = "conv2" ) fa = MCONV.Flatten(name="flata") fb = MCONV.Flatten(name="flatb") f = MCONV.Flatten(name = "flat") h = ML.Hidden(2048, activation = MA.Max_norm(), decorators = [MD.BinomialDropout(0.75)], regularizations = [], name = "hid" ) passa = ML.Hidden(1500, activation = MA.Pass(), decorators = [MD.BinomialDropout(0.5)], regularizations = [], name = "pass1" ) passb = ML.Hidden(1500, activation = MA.Pass(), decorators = [MD.BinomialDropout(0.5)], regularizations = [], name = "pass2" ) h2 = ML.Hidden(2048, activation = MA.Max_norm(), decorators = [MD.BinomialDropout(0.75)], regularizations = [], name = "hid2" ) o = ML.SoftmaxClassifier(2, decorators = [], learningScenario = ls, costObject = cost, name = "out", regularizations = [] ) model = i > c1 > c3 > c2 > f > h > h2 > o c1 > fa > passa > h > h2 > o c2 > fb > passb >h > h2 > o tscore = [] vscore = [] tdata = load_data(trainfile) vdata = load_data(validfile) vdata = (center(vdata[0]),vdata[1]) test = load_data(testfile)
