def profileDerivativeUiApprox(self):
k = 10
U = numpy.random.rand(self.m, k)
V = numpy.random.rand(self.n, k)
indPtr, colInds = SparseUtils.getOmegaListPtr(self.X)
gp = numpy.random.rand(self.n)
gp /= gp.sum()
gq = numpy.random.rand(self.n)
gq /= gq.sum()
j = 3
numRowSamples = 100
numAucSamples = 10
permutedRowInds = numpy.array(numpy.random.permutation(self.m), numpy.uint32)
permutedColInds = numpy.array(numpy.random.permutation(self.n), numpy.uint32)
maxLocalAuc = MaxLocalAUC(k, w=0.9)
normGp, normGq = maxLocalAuc.computeNormGpq(indPtr, colInds, gp, gq, self.m)
lmbda = 0.001
normalise = True
learner = MaxLocalAUCCython()
def run():
numRuns = 10
for j in range(numRuns):
for i in range(self.m):
learner.derivativeUiApprox(indPtr, colInds, U, V, gp, gq, permutedColInds, i)
ProfileUtils.profile("run()", globals(), locals())
def testScale(self):
"""
Look at the scales of the unnormalised gradients.
"""
m = 100
n = 400
k = 3
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, csarray=True)
w = 0.1
eps = 0.001
learner = MaxAUCTanh(k, w)
learner.normalise = False
learner.lmbdaU = 1.0
learner.lmbdaV = 1.0
learner.rho = 1.0
learner.numAucSamples = 100
indPtr, colInds = SparseUtils.getOmegaListPtr(X)
r = numpy.random.rand(m)
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
gi = numpy.random.rand(m)
gi /= gi.sum()
gp = numpy.random.rand(n)
gp /= gp.sum()
gq = numpy.random.rand(n)
gq /= gq.sum()
permutedRowInds = numpy.array(numpy.random.permutation(m),
numpy.uint32)
permutedColInds = numpy.array(numpy.random.permutation(n),
numpy.uint32)
maxLocalAuc = MaxLocalAUC(k, w)
normGp, normGq = maxLocalAuc.computeNormGpq(indPtr, colInds, gp, gq, m)
normDui = 0
for i in range(m):
du = learner.derivativeUi(indPtr, colInds, U, V, r, gi, gp, gq, i)
normDui += numpy.linalg.norm(du)
normDui /= float(m)
print(normDui)
normDvi = 0
for i in range(n):
dv = learner.derivativeVi(indPtr, colInds, U, V, r, gi, gp, gq, i)
normDvi += numpy.linalg.norm(dv)
normDvi /= float(n)
print(normDvi)
def testOverfit(self):
"""
See if we can get a zero objective on the hinge loss
"""
m = 10
n = 20
k = 5
u = 0.5
w = 1 - u
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, w, csarray=True)
eps = 0.001
k = 10
maxLocalAuc = MaxLocalAUC(k, u, eps=eps, stochastic=True)
maxLocalAuc.rate = "constant"
maxLocalAuc.maxIterations = 500
maxLocalAuc.numProcesses = 1
maxLocalAuc.loss = "hinge"
maxLocalAuc.validationUsers = 0
maxLocalAuc.lmbda = 0
print("Overfit example")
U, V, trainMeasures, testMeasures, iterations, time = maxLocalAuc.learnModel(
X, verbose=True)
self.assertAlmostEquals(trainMeasures[-1, 0], 0, 3)
def testScale(self):
"""
Look at the scales of the unnormalised gradients.
"""
m = 100
n = 400
k = 3
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, csarray=True)
w = 0.1
eps = 0.001
learner = MaxAUCTanh(k, w)
learner.normalise = False
learner.lmbdaU = 1.0
learner.lmbdaV = 1.0
learner.rho = 1.0
learner.numAucSamples = 100
indPtr, colInds = SparseUtils.getOmegaListPtr(X)
r = numpy.random.rand(m)
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
gi = numpy.random.rand(m)
gi /= gi.sum()
gp = numpy.random.rand(n)
gp /= gp.sum()
gq = numpy.random.rand(n)
gq /= gq.sum()
permutedRowInds = numpy.array(numpy.random.permutation(m), numpy.uint32)
permutedColInds = numpy.array(numpy.random.permutation(n), numpy.uint32)
maxLocalAuc = MaxLocalAUC(k, w)
normGp, normGq = maxLocalAuc.computeNormGpq(indPtr, colInds, gp, gq, m)
normDui = 0
for i in range(m):
du = learner.derivativeUi(indPtr, colInds, U, V, r, gi, gp, gq, i)
normDui += numpy.linalg.norm(du)
normDui /= float(m)
print(normDui)
normDvi = 0
for i in range(n):
dv = learner.derivativeVi(indPtr, colInds, U, V, r, gi, gp, gq, i)
normDvi += numpy.linalg.norm(dv)
normDvi /= float(n)
print(normDvi)
def testParallelLearnModel(self):
numpy.random.seed(21)
m = 500
n = 200
k = 5
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, csarray=True)
from wallhack.rankingexp.DatasetUtils import DatasetUtils
X, U, V = DatasetUtils.syntheticDataset1()
u = 0.1
w = 1-u
eps = 0.05
maxLocalAuc = MaxLocalAUC(k, w, alpha=1.0, eps=eps, stochastic=True)
maxLocalAuc.maxIterations = 3
maxLocalAuc.recordStep = 1
maxLocalAuc.rate = "optimal"
maxLocalAuc.t0 = 2.0
maxLocalAuc.validationUsers = 0.0
maxLocalAuc.numProcesses = 4
os.system('taskset -p 0xffffffff %d' % os.getpid())
print(X.nnz/maxLocalAuc.numAucSamples)
U, V = maxLocalAuc.parallelLearnModel(X)
def testOverfit(self):
"""
See if we can get a zero objective on the hinge loss
"""
m = 10
n = 20
k = 5
u = 0.5
w = 1-u
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, w, csarray=True)
eps = 0.001
k = 10
maxLocalAuc = MaxLocalAUC(k, u, eps=eps, stochastic=True)
maxLocalAuc.rate = "constant"
maxLocalAuc.maxIterations = 500
maxLocalAuc.numProcesses = 1
maxLocalAuc.loss = "hinge"
maxLocalAuc.validationUsers = 0
maxLocalAuc.lmbda = 0
print("Overfit example")
U, V, trainMeasures, testMeasures, iterations, time = maxLocalAuc.learnModel(X, verbose=True)
self.assertAlmostEquals(trainMeasures[-1, 0], 0, 3)
def testModelSelectLmbda(self):
m = 10
n = 20
k = 5
u = 0.5
w = 1 - u
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, w, csarray=True)
os.system('taskset -p 0xffffffff %d' % os.getpid())
eps = 0.001
k = 5
maxLocalAuc = MaxLocalAUC(k, w, eps=eps, stochastic=True)
maxLocalAuc.maxIterations = 5
maxLocalAuc.recordStep = 1
maxLocalAuc.validationSize = 3
maxLocalAuc.metric = "f1"
maxLocalAuc.rate = "constant"
maxLocalAuc.ks = numpy.array([4, 8])
maxLocalAuc.modelSelectLmbda(X)
def profileDerivativeVjApprox(self):
k = 10
U = numpy.random.rand(self.m, k)
V = numpy.random.rand(self.n, k)
indPtr, colInds = SparseUtils.getOmegaListPtr(self.X)
gp = numpy.random.rand(self.n)
gp /= gp.sum()
gq = numpy.random.rand(self.n)
gq /= gq.sum()
j = 3
numRowSamples = 100
numAucSamples = 10
permutedRowInds = numpy.array(numpy.random.permutation(self.m),
numpy.uint32)
permutedColInds = numpy.array(numpy.random.permutation(self.n),
numpy.uint32)
maxLocalAuc = MaxLocalAUC(k, w=0.9)
normGp, normGq = maxLocalAuc.computeNormGpq(indPtr, colInds, gp, gq,
self.m)
lmbda = 0.001
normalise = True
learner = MaxLocalAUCCython()
def run():
numRuns = 1
for i in range(numRuns):
for j in range(self.n):
learner.derivativeViApprox(indPtr, colInds, U, V, gp, gq,
normGp, normGq, permutedRowInds,
permutedColInds, i)
ProfileUtils.profile('run()', globals(), locals())
def testModelSelectLmbda(self):
m = 10
n = 20
k = 5
u = 0.5
w = 1-u
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, w, csarray=True)
os.system('taskset -p 0xffffffff %d' % os.getpid())
eps = 0.001
k = 5
maxLocalAuc = MaxLocalAUC(k, w, eps=eps, stochastic=True)
maxLocalAuc.maxIterations = 5
maxLocalAuc.recordStep = 1
maxLocalAuc.validationSize = 3
maxLocalAuc.metric = "f1"
maxLocalAuc.rate = "constant"
maxLocalAuc.ks = numpy.array([4, 8])
maxLocalAuc.modelSelectLmbda(X)
def profileLearnModel2(self):
#Profile stochastic case
#X = DatasetUtils.flixster()
#X = Sampling.sampleUsers(X, 1000)
X, U, V = DatasetUtils.syntheticDataset1(u=0.001, m=10000, n=1000)
rho = 0.00
u = 0.2
w = 1 - u
eps = 10**-6
alpha = 0.5
k = self.k
maxLocalAuc = MaxLocalAUC(k, w, alpha=alpha, eps=eps, stochastic=True)
maxLocalAuc.numRowSamples = 2
maxLocalAuc.numAucSamples = 10
maxLocalAuc.maxIterations = 1
maxLocalAuc.numRecordAucSamples = 100
maxLocalAuc.recordStep = 10
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.rate = "optimal"
#maxLocalAuc.parallelSGD = True
trainTestX = Sampling.shuffleSplitRows(X, maxLocalAuc.folds, 5)
trainX, testX = trainTestX[0]
def run():
U, V, trainMeasures, testMeasures, iterations, time = maxLocalAuc.learnModel(
trainX, True)
#logging.debug("Train Precision@5=" + str(MCEvaluator.precisionAtK(trainX, U, V, 5)))
#logging.debug("Train Precision@10=" + str(MCEvaluator.precisionAtK(trainX, U, V, 10)))
#logging.debug("Train Precision@20=" + str(MCEvaluator.precisionAtK(trainX, U, V, 20)))
#logging.debug("Train Precision@50=" + str(MCEvaluator.precisionAtK(trainX, U, V, 50)))
#logging.debug("Test Precision@5=" + str(MCEvaluator.precisionAtK(testX, U, V, 5)))
#logging.debug("Test Precision@10=" + str(MCEvaluator.precisionAtK(testX, U, V, 10)))
#logging.debug("Test Precision@20=" + str(MCEvaluator.precisionAtK(testX, U, V, 20)))
#logging.debug("Test Precision@50=" + str(MCEvaluator.precisionAtK(testX, U, V, 50)))
ProfileUtils.profile('run()', globals(), locals())
def testLearningRateSelect(self):
m = 10
n = 20
k = 5
u = 0.5
w = 1 - u
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, w, csarray=True)
eps = 0.001
maxLocalAuc = MaxLocalAUC(k, u, eps=eps, stochastic=True)
maxLocalAuc.rate = "optimal"
maxLocalAuc.maxIterations = 5
maxLocalAuc.numProcesses = 1
maxLocalAuc.learningRateSelect(X)
def testParallelLearnModel(self):
numpy.random.seed(21)
m = 500
n = 200
k = 5
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, csarray=True)
from wallhack.rankingexp.DatasetUtils import DatasetUtils
X, U, V = DatasetUtils.syntheticDataset1()
u = 0.1
w = 1 - u
eps = 0.05
maxLocalAuc = MaxLocalAUC(k, w, alpha=1.0, eps=eps, stochastic=True)
maxLocalAuc.maxIterations = 3
maxLocalAuc.recordStep = 1
maxLocalAuc.rate = "optimal"
maxLocalAuc.t0 = 2.0
maxLocalAuc.validationUsers = 0.0
maxLocalAuc.numProcesses = 4
os.system('taskset -p 0xffffffff %d' % os.getpid())
print(X.nnz / maxLocalAuc.numAucSamples)
U, V = maxLocalAuc.parallelLearnModel(X)
def testLearnModel(self):
m = 50
n = 20
k = 5
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, csarray=True)
u = 0.1
w = 1 - u
eps = 0.05
maxLocalAuc = MaxLocalAUC(k, w, alpha=5.0, eps=eps, stochastic=False)
U, V = maxLocalAuc.learnModel(X)
maxLocalAuc.stochastic = True
U, V = maxLocalAuc.learnModel(X)
#Test case where we do not have validation set
maxLocalAuc.validationUsers = 0.0
U, V = maxLocalAuc.learnModel(X)
def testLearningRateSelect(self):
m = 10
n = 20
k = 5
u = 0.5
w = 1-u
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, w, csarray=True)
eps = 0.001
maxLocalAuc = MaxLocalAUC(k, u, eps=eps, stochastic=True)
maxLocalAuc.rate = "optimal"
maxLocalAuc.maxIterations = 5
maxLocalAuc.numProcesses = 1
maxLocalAuc.learningRateSelect(X)
def testLearnModel(self):
m = 50
n = 20
k = 5
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, csarray=True)
u = 0.1
w = 1-u
eps = 0.05
maxLocalAuc = MaxLocalAUC(k, w, alpha=5.0, eps=eps, stochastic=False)
U, V = maxLocalAuc.learnModel(X)
maxLocalAuc.stochastic = True
U, V = maxLocalAuc.learnModel(X)
#Test case where we do not have validation set
maxLocalAuc.validationUsers = 0.0
U, V = maxLocalAuc.learnModel(X)
dataset = "synthetic"
saveResults = True
prefix = "Exp2"
outputFile = PathDefaults.getOutputDir() + "ranking/" + prefix + dataset.title() + "Results.npz"
X = DatasetUtils.getDataset(dataset)
testSize = 5
folds = 2
trainTestXs = Sampling.shuffleSplitRows(X, folds, testSize)
u = 0.1
w2 = 1-u
k = 64
eps = 10**-8
maxLocalAuc = MaxLocalAUC(k, w2, eps=eps, stochastic=True)
maxLocalAuc.maxIterations = 50
maxLocalAuc.numRowSamples = 30
maxLocalAuc.numAucSamples = 10
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.recordStep = 10
maxLocalAuc.rate = "optimal"
maxLocalAuc.alpha = 1.0
maxLocalAuc.t0 = 0.1
maxLocalAuc.lmbdaU = 0.0
maxLocalAuc.lmbdaV = 1.0
maxLocalAuc.rho = 0.5
maxItems = 10
chunkSize = 1
startAverages = numpy.array([2, 5, 10, 20, 30, 40])
X = DatasetUtils.getDataset(dataset)
m, n = X.shape
testSize = 5
trainTestXs = Sampling.shuffleSplitRows(X, 1, testSize)
trainX, testX = trainTestXs[0]
logging.debug("Number of non-zero elements: " + str((trainX.nnz, testX.nnz)))
k2 = 32
u2 = 0.1
w2 = 1-u2
eps = 10**-8
lmbda = 1.0
maxLocalAuc = MaxLocalAUC(k2, w2, eps=eps, lmbdaU=0.0, lmbdaV=lmbda, stochastic=True)
maxLocalAuc.alpha = 1.0
maxLocalAuc.alphas = 2.0**-numpy.arange(-5, 5, 1)
maxLocalAuc.folds = 5
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.itemExpP = 1.0
maxLocalAuc.itemExpQ = 1.0
maxLocalAuc.lmbdas = numpy.linspace(0.5, 2.0, 7)
maxLocalAuc.maxIterations = 100
maxLocalAuc.metric = "f1"
maxLocalAuc.normalise = True
maxLocalAuc.numAucSamples = 10
#maxLocalAuc.numProcesses = 1
maxLocalAuc.numRecordAucSamples = 100
maxLocalAuc.numRowSamples = 30
maxLocalAuc.rate = "optimal"
m, n = X.shape u = 0.1 w = 1-u testSize = 5 folds = 5 trainTestXs = Sampling.shuffleSplitRows(X, folds, testSize) numRecordAucSamples = 200 k2 = 8 u2 = 0.5 w2 = 1-u2 eps = 10**-8 lmbda = 0.1 maxLocalAuc = MaxLocalAUC(k2, w2, eps=eps, lmbdaU=lmbda, lmbdaV=lmbda, stochastic=True) maxLocalAuc.maxIterations = 100 maxLocalAuc.numRowSamples = 30 maxLocalAuc.numAucSamples = 10 maxLocalAuc.numRecordAucSamples = 100 maxLocalAuc.recordStep = 10 maxLocalAuc.initialAlg = "rand" maxLocalAuc.rate = "constant" maxLocalAuc.alpha = 0.1 maxLocalAuc.t0 = 1.0 maxLocalAuc.folds = 1 maxLocalAuc.rho = 1.0 maxLocalAuc.ks = numpy.array([k2]) maxLocalAuc.validationSize = 3 maxLocalAuc.lmbdas = numpy.linspace(0.5, 2.0, 7) maxLocalAuc.normalise = True
def testDerivativeViApprox(self):
"""
We'll test the case in which we apprormate using a large number of samples
for the AUC and see if we get close to the exact derivative
"""
m = 20
n = 30
k = 3
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, csarray=True)
for i in range(m):
X[i, 0] = 1
X[i, 1] = 0
w = 0.1
eps = 0.001
learner = MaxAUCSigmoid(k, w)
learner.normalise = False
learner.lmbdaU = 0
learner.lmbdaV = 0
learner.numAucSamples = n
indPtr, colInds = SparseUtils.getOmegaListPtr(X)
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
gp = numpy.random.rand(n)
gp /= gp.sum()
gq = numpy.random.rand(n)
gq /= gq.sum()
permutedRowInds = numpy.array(numpy.random.permutation(m),
numpy.uint32)
permutedColInds = numpy.array(numpy.random.permutation(n),
numpy.uint32)
maxLocalAuc = MaxLocalAUC(k, w)
normGp, normGq = maxLocalAuc.computeNormGpq(indPtr, colInds, gp, gq, m)
numRuns = 200
numTests = 5
#Let's compare against using the exact derivative
for i in numpy.random.permutation(m)[0:numTests]:
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
dv1 = numpy.zeros(k)
for j in range(numRuns):
dv1 += learner.derivativeViApprox(indPtr, colInds, U, V, gp,
gq, normGp, normGq,
permutedRowInds,
permutedColInds, i)
dv1 /= numRuns
dv2 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
dv3 = numpy.zeros(k)
for j in range(k):
eps = 10**-6
tempV = V.copy()
tempV[i, j] += eps
obj1 = learner.objective(indPtr, colInds, indPtr, colInds, U,
tempV, gp, gq)
tempV = V.copy()
tempV[i, j] -= eps
obj2 = learner.objective(indPtr, colInds, indPtr, colInds, U,
tempV, gp, gq)
dv3[j] = (obj1 - obj2) / (2 * eps)
print(dv1, dv2, dv3)
nptst.assert_array_almost_equal(dv1, dv2, 3)
learner.lmbdaV = 0.5
learner.rho = 0.5
for i in numpy.random.permutation(m)[0:numTests]:
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
dv1 = numpy.zeros(k)
for j in range(numRuns):
dv1 += learner.derivativeViApprox(indPtr, colInds, U, V, gp,
gq, normGp, normGq,
permutedRowInds,
permutedColInds, i)
dv1 /= numRuns
dv2 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
print(dv1, dv2)
nptst.assert_array_almost_equal(dv1, dv2, 3)
learner.numRowSamples = 10
numRuns = 1000
for i in numpy.random.permutation(m)[0:numTests]:
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
dv1 = numpy.zeros(k)
for j in range(numRuns):
dv1 += learner.derivativeViApprox(indPtr, colInds, U, V, gp,
gq, normGp, normGq,
permutedRowInds,
permutedColInds, i)
dv1 /= numRuns
dv2 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
print(dv1, dv2)
nptst.assert_array_almost_equal(dv1, dv2, 3)
maxLocalAuc.numRowSamples = m
maxLocalAuc.numAucSamples = 20
maxLocalAuc.lmbdaV = 0
numRuns = 1000
print("Final test")
#for i in numpy.random.permutation(m)[0:numTests]:
for i in range(m):
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
dv1 = numpy.zeros(k)
for j in range(numRuns):
dv1 += learner.derivativeViApprox(indPtr, colInds, U, V, gp,
gq, normGp, normGq,
permutedRowInds,
permutedColInds, i)
dv1 /= numRuns
#dv1 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
dv2 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
print(i, dv1, dv2)
nptst.assert_array_almost_equal(dv1, dv2, 3)
def testCopy(self):
u= 0.1
eps = 0.001
k = 10
maxLocalAuc = MaxLocalAUC(k, u, alpha=5.0, eps=eps)
maxLocalAuc.copy()
saveResults = True
prefix = "Regularisation5"
outputFile = PathDefaults.getOutputDir() + "ranking/" + prefix + dataset.title() + "Results.npz"
X = DatasetUtils.getDataset(dataset)
testSize = 5
trainTestXs = Sampling.shuffleSplitRows(X, 1, testSize)
trainX, testX = trainTestXs[0]
logging.debug("Number of non-zero elements: " + str((trainX.nnz, testX.nnz)))
u = 0.1
w = 1-u
k2 = 64
eps = 10**-6
maxLocalAuc = MaxLocalAUC(k2, w, eps=eps, stochastic=True)
maxLocalAuc.alpha = 0.1
maxLocalAuc.alphas = 2.0**-numpy.arange(0, 5, 1)
maxLocalAuc.folds = 1
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.itemExpP = 0.0
maxLocalAuc.itemExpQ = 0.0
maxLocalAuc.ks = numpy.array([k2])
maxLocalAuc.lmbdaU = 0.0
maxLocalAuc.lmbdaV = 0.0
maxLocalAuc.lmbdas = 2.0**-numpy.arange(0, 8)
maxLocalAuc.loss = "hinge"
maxLocalAuc.maxIterations = 500
maxLocalAuc.maxNorms = 2.0**numpy.arange(-2, 5, 0.5)
maxLocalAuc.metric = "f1"
maxLocalAuc.normalise = True
def profileLearnModel(self):
#Profile full gradient descent
X, U, V = DatasetUtils.syntheticDataset1(u=0.01, m=1000, n=2000)
#X, U, V = DatasetUtils.syntheticDataset1()
#X, U, V = DatasetUtils.syntheticDataset1(u=0.2, sd=0.2)
#X = DatasetUtils.flixster()
u = 0.2
w = 1 - u
eps = 10**-6
alpha = 0.5
maxLocalAuc = MaxLocalAUC(self.k,
w,
alpha=alpha,
eps=eps,
stochastic=True)
maxLocalAuc.maxNormU = 10
maxLocalAuc.maxNormV = 10
maxLocalAuc.maxIterations = 100
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.rate = "constant"
maxLocalAuc.parallelSGD = True
maxLocalAuc.numProcesses = 8
maxLocalAuc.numAucSamples = 10
maxLocalAuc.numRowSamples = 30
maxLocalAuc.scaleAlpha = False
maxLocalAuc.loss = "hinge"
maxLocalAuc.validationUsers = 0.0
print(maxLocalAuc)
ProfileUtils.profile('maxLocalAuc.learnModel(X)', globals(), locals())
def testCopy(self):
u = 0.1
eps = 0.001
k = 10
maxLocalAuc = MaxLocalAUC(k, u, alpha=5.0, eps=eps)
maxLocalAuc.copy()
u = 0.1
w = 1-u
logging.debug("Sampled X shape: " + str(X.shape))
testSize = 5
folds = 5
trainTestXs = Sampling.shuffleSplitRows(X, folds, testSize)
numRecordAucSamples = 200
k2 = 64
u2 = 0.5
w2 = 1-u2
eps = 10**-8
maxLocalAuc = MaxLocalAUC(k2, w2, eps=eps,stochastic=True)
maxLocalAuc.alpha = 0.05
maxLocalAuc.alphas = 2.0**-numpy.arange(0, 9, 1)
maxLocalAuc.folds = 1
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.itemExpP = 0.0
maxLocalAuc.itemExpQ = 0.0
maxLocalAuc.ks = numpy.array([k2])
maxLocalAuc.maxNorms = 2.0**numpy.arange(-2, 2)
maxLocalAuc.maxIterations = 500
maxLocalAuc.lmbdaU = 0
maxLocalAuc.lmbdaV = 0
maxLocalAuc.loss = "hinge"
maxLocalAuc.maxIterations = 500
maxLocalAuc.maxNormU = 100
maxLocalAuc.maxNormV = 100
def testStr(self):
k = 10
u = 0.1
eps = 0.001
maxLocalAuc = MaxLocalAUC(k, u, eps=eps)
dataset = "movielens"
saveResults = True
prefix = "Rademacher"
outputFile = PathDefaults.getOutputDir() + "ranking/" + prefix + dataset.title() + "Results.npz"
X = DatasetUtils.getDataset(dataset, nnz=20000)
m, n = X.shape
k2 = 16
u2 = 5/float(n)
w2 = 1-u2
eps = 10**-8
lmbda = 0.0
maxLocalAuc = MaxLocalAUC(k2, w2, eps=eps, lmbdaU=lmbda, lmbdaV=lmbda, stochastic=True)
maxLocalAuc.alpha = 0.1
maxLocalAuc.alphas = 2.0**-numpy.arange(0, 5, 1)
maxLocalAuc.bound = True
maxLocalAuc.delta = 0.1
maxLocalAuc.eta = 0
maxLocalAuc.folds = 2
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.itemExpP = 0.0
maxLocalAuc.itemExpQ = 0.0
maxLocalAuc.ks = numpy.array([4, 8, 16, 32, 64, 128])
maxLocalAuc.lmbdas = 2.0**-numpy.arange(1, 10)
maxLocalAuc.loss = "hinge"
maxLocalAuc.maxIterations = 100
maxLocalAuc.maxNorm = 1/numpy.sqrt(2)
maxLocalAuc.metric = "f1"
softImpute.maxIterations = maxIterations
softImpute.metric = "f1"
softImpute.q = 3
softImpute.p = 10
softImpute.rho = 0.1
softImpute.eps = 10**-4
softImpute.numProcesses = args.processes
wrmf = WeightedMf(k=k, maxIterations=maxIterations, alpha=1.0)
wrmf.ks = ks
wrmf.folds = folds
wrmf.lmbdas = 2.0**-numpy.arange(-1, 12, 2)
wrmf.metric = "f1"
wrmf.numProcesses = args.processes
maxLocalAuc = MaxLocalAUC(k=k, w=0.9, maxIterations=50, lmbdaU=0.1, lmbdaV=0.1, stochastic=True)
maxLocalAuc.numRowSamples = 10
maxLocalAuc.parallelSGD = True
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.ks = ks
maxLocalAuc.folds = folds
maxLocalAuc.metric = "f1"
maxLocalAuc.numProcesses = args.processes
kNeighbours = 25
knn = CosineKNNRecommender(kNeighbours)
numFeatures = 200
slim = SLIM(num_selected_features=numFeatures)
learners = [("SoftImpute", softImpute), ("WRMF", wrmf), ("KNN", knn), ("MLAUC", maxLocalAuc), ("SLIM", slim)]
dataset = sys.argv[1]
else:
dataset = "movielens"
saveResults = False
prefix = "LearningRate2"
outputFile = PathDefaults.getOutputDir() + "ranking/" + prefix + dataset.title() + "Results.npz"
X = DatasetUtils.getDataset(dataset)
m, n = X.shape
k2 = 64
u2 = 5/float(n)
w2 = 1-u2
eps = 10**-8
lmbda = 0.01
maxLocalAuc = MaxLocalAUC(k2, w2, eps=eps, lmbdaU=0.1, lmbdaV=0.1, stochastic=True)
maxLocalAuc.alpha = 0.5
maxLocalAuc.alphas = 2.0**-numpy.arange(2, 9, 2)
maxLocalAuc.beta = 2
maxLocalAuc.bound = False
maxLocalAuc.delta = 0.1
maxLocalAuc.eta = 20
maxLocalAuc.folds = 2
maxLocalAuc.initialAlg = "svd"
maxLocalAuc.itemExpP = 0.0
maxLocalAuc.itemExpQ = 0.0
maxLocalAuc.ks = numpy.array([4, 8, 16, 32, 64, 128])
maxLocalAuc.lmbdas = 2.0**-numpy.arange(1, 5)
maxLocalAuc.loss = "hinge"
maxLocalAuc.maxIterations = 500
maxLocalAuc.maxNorm = 100
trainX, testX = trainTestXs[0]
trainOmegaPtr = SparseUtils.getOmegaListPtr(trainX)
testOmegaPtr = SparseUtils.getOmegaListPtr(testX)
allOmegaPtr = SparseUtils.getOmegaListPtr(X)
numRecordAucSamples = 200
logging.debug("Number of non-zero elements: " + str((trainX.nnz, testX.nnz)))
#w = 1.0
k2 = 8
u2 = 5/float(n)
w2 = 1-u2
eps = 10**-8
lmbda = 0.01
maxLocalAuc = MaxLocalAUC(k2, w2, eps=eps, lmbdaU=0.1, lmbdaV=0.1, stochastic=True)
maxLocalAuc.alpha = 0.1
maxLocalAuc.alphas = 2.0**-numpy.arange(0, 5, 1)
maxLocalAuc.beta = 2
maxLocalAuc.bound = False
maxLocalAuc.delta = 0.1
maxLocalAuc.eta = 0
maxLocalAuc.folds = 2
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.itemExpP = 0.0
maxLocalAuc.itemExpQ = 0.0
maxLocalAuc.ks = numpy.array([4, 8, 16, 32, 64, 128])
maxLocalAuc.lmbdas = numpy.linspace(0.5, 2.0, 7)
maxLocalAuc.loss = "hinge"
maxLocalAuc.maxIterations = 100
maxLocalAuc.maxNorm = 100
m, n = X.shape u = 0.1 w = 1-u testSize = 5 folds = 5 trainTestXs = Sampling.shuffleSplitRows(X, folds, testSize) numRecordAucSamples = 200 k2 = 8 u2 = 0.5 w2 = 1-u2 eps = 10**-4 lmbda = 0.0 maxLocalAuc = MaxLocalAUC(k2, w2, eps=eps, lmbdaU=lmbda, lmbdaV=lmbda, stochastic=True) maxLocalAuc.alpha = 0.05 maxLocalAuc.alphas = 2.0**-numpy.arange(0, 5, 1) maxLocalAuc.folds = 1 maxLocalAuc.initialAlg = "rand" maxLocalAuc.itemExpP = 0.0 maxLocalAuc.itemExpQ = 0.0 maxLocalAuc.ks = numpy.array([k2]) maxLocalAuc.lmbdas = numpy.linspace(0.5, 2.0, 7) maxLocalAuc.maxIterations = 500 maxLocalAuc.metric = "f1" maxLocalAuc.normalise = True maxLocalAuc.numAucSamples = 10 maxLocalAuc.numProcesses = 1 maxLocalAuc.numRecordAucSamples = 100 maxLocalAuc.numRowSamples = 30
"""
if len(sys.argv) > 1:
dataset = sys.argv[1]
else:
dataset = "flixster"
saveResults = True
prefix = "ModelSelect"
outputFile = PathDefaults.getOutputDir() + "ranking/" + prefix + dataset.title() + "Results.npz"
u = 0.1
w = 1-u
k2 = 64
eps = 10**-6
maxLocalAuc = MaxLocalAUC(k2, w, eps=eps, stochastic=True)
maxLocalAuc.alpha = 0.1
maxLocalAuc.alphas = 2.0**-numpy.arange(1, 7, 1)
maxLocalAuc.folds = 1
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.itemExpP = 0.0
maxLocalAuc.itemExpQ = 0.0
maxLocalAuc.ks = numpy.array([k2])
maxLocalAuc.lmbdas = 2.0**-numpy.arange(1, 6)
maxLocalAuc.lmbdaU = 0.25
maxLocalAuc.lmbdaV = 0.25
maxLocalAuc.loss = "hinge"
maxLocalAuc.maxIterations = 500
maxLocalAuc.maxNormU = 100
maxLocalAuc.maxNormV = 100
maxLocalAuc.metric = "f1"
def testDerivativeViApprox(self):
"""
We'll test the case in which we apprormate using a large number of samples
for the AUC and see if we get close to the exact derivative
"""
m = 20
n = 30
k = 3
X = SparseUtils.generateSparseBinaryMatrix((m, n), k, csarray=True)
for i in range(m):
X[i, 0] = 1
X[i, 1] = 0
w = 0.1
eps = 0.001
learner = MaxAUCSigmoid(k, w)
learner.normalise = False
learner.lmbdaU = 0
learner.lmbdaV = 0
learner.numAucSamples = n
indPtr, colInds = SparseUtils.getOmegaListPtr(X)
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
gp = numpy.random.rand(n)
gp /= gp.sum()
gq = numpy.random.rand(n)
gq /= gq.sum()
permutedRowInds = numpy.array(numpy.random.permutation(m), numpy.uint32)
permutedColInds = numpy.array(numpy.random.permutation(n), numpy.uint32)
maxLocalAuc = MaxLocalAUC(k, w)
normGp, normGq = maxLocalAuc.computeNormGpq(indPtr, colInds, gp, gq, m)
numRuns = 200
numTests = 5
#Let's compare against using the exact derivative
for i in numpy.random.permutation(m)[0:numTests]:
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
dv1 = numpy.zeros(k)
for j in range(numRuns):
dv1 += learner.derivativeViApprox(indPtr, colInds, U, V, gp, gq, normGp, normGq, permutedRowInds, permutedColInds, i)
dv1 /= numRuns
dv2 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
dv3 = numpy.zeros(k)
for j in range(k):
eps = 10**-6
tempV = V.copy()
tempV[i,j] += eps
obj1 = learner.objective(indPtr, colInds, indPtr, colInds, U, tempV, gp, gq)
tempV = V.copy()
tempV[i,j] -= eps
obj2 = learner.objective(indPtr, colInds, indPtr, colInds, U, tempV, gp, gq)
dv3[j] = (obj1-obj2)/(2*eps)
print(dv1, dv2, dv3)
nptst.assert_array_almost_equal(dv1, dv2, 3)
learner.lmbdaV = 0.5
learner.rho = 0.5
for i in numpy.random.permutation(m)[0:numTests]:
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
dv1 = numpy.zeros(k)
for j in range(numRuns):
dv1 += learner.derivativeViApprox(indPtr, colInds, U, V, gp, gq, normGp, normGq, permutedRowInds, permutedColInds, i)
dv1 /= numRuns
dv2 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
print(dv1, dv2)
nptst.assert_array_almost_equal(dv1, dv2, 3)
learner.numRowSamples = 10
numRuns = 1000
for i in numpy.random.permutation(m)[0:numTests]:
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
dv1 = numpy.zeros(k)
for j in range(numRuns):
dv1 += learner.derivativeViApprox(indPtr, colInds, U, V, gp, gq, normGp, normGq, permutedRowInds, permutedColInds, i)
dv1 /= numRuns
dv2 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
print(dv1, dv2)
nptst.assert_array_almost_equal(dv1, dv2, 3)
maxLocalAuc.numRowSamples = m
maxLocalAuc.numAucSamples = 20
maxLocalAuc.lmbdaV = 0
numRuns = 1000
print("Final test")
#for i in numpy.random.permutation(m)[0:numTests]:
for i in range(m):
U = numpy.random.rand(X.shape[0], k)
V = numpy.random.rand(X.shape[1], k)
dv1 = numpy.zeros(k)
for j in range(numRuns):
dv1 += learner.derivativeViApprox(indPtr, colInds, U, V, gp, gq, normGp, normGq, permutedRowInds, permutedColInds, i)
dv1 /= numRuns
#dv1 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
dv2 = learner.derivativeVi(indPtr, colInds, U, V, gp, gq, i)
print(i, dv1, dv2)
nptst.assert_array_almost_equal(dv1, dv2, 3)
trainOmegaList = SparseUtils.getOmegaList(trainX)
trainOmegaPtr = SparseUtils.getOmegaListPtr(trainX)
testOmegaList = SparseUtils.getOmegaList(testX)
testOmegaPtr = SparseUtils.getOmegaListPtr(testX)
allOmegaPtr = SparseUtils.getOmegaListPtr(X)
numRecordAucSamples = 200
logging.debug("Number of non-zero elements: " + str((trainX.nnz, testX.nnz)))
k2 = 64
u2 = 5/float(n)
w2 = 1-u2
eps = 10**-8
lmbda = 0.01
maxLocalAuc = MaxLocalAUC(k2, w2, eps=eps, lmbdaU=0.1, lmbdaV=0.1, stochastic=True)
maxLocalAuc.alpha = 32
maxLocalAuc.alphas = 2.0**-numpy.arange(0, 5, 1)
maxLocalAuc.beta = 2
maxLocalAuc.bound = False
maxLocalAuc.delta = 0.1
maxLocalAuc.eta = 0
maxLocalAuc.folds = 2
maxLocalAuc.initialAlg = "rand"
maxLocalAuc.itemExpP = 0.0
maxLocalAuc.itemExpQ = 0.0
maxLocalAuc.ks = numpy.array([4, 8, 16, 32, 64, 128])
maxLocalAuc.lmbdas = numpy.linspace(0.5, 2.0, 7)
maxLocalAuc.loss = "hinge"
maxLocalAuc.maxIterations = 100
maxLocalAuc.maxNorm = 100
