def testPredict(self):
#Create a set of indices
lmbda = 0.0
iterativeSoftImpute = IterativeSoftImpute(lmbda, k=10)
matrixIterator = iter(self.matrixList)
ZList = iterativeSoftImpute.learnModel(matrixIterator)
XhatList = iterativeSoftImpute.predict(ZList, self.indsList)
#Check we get the exact matrices returned
for i, Xhat in enumerate(XhatList):
nptst.assert_array_almost_equal(numpy.array(Xhat.todense()), self.matrixList[i].todense())
self.assertEquals(Xhat.nnz, self.indsList[i].shape[0])
self.assertAlmostEquals(MCEvaluator.meanSqError(Xhat, self.matrixList[i]), 0)
self.assertAlmostEquals(MCEvaluator.rootMeanSqError(Xhat, self.matrixList[i]), 0)
#Try moderate lambda
lmbda = 0.1
iterativeSoftImpute = IterativeSoftImpute(lmbda, k=10)
matrixIterator = iter(self.matrixList)
ZList = list(iterativeSoftImpute.learnModel(matrixIterator))
XhatList = iterativeSoftImpute.predict(iter(ZList), self.indsList)
for i, Xhat in enumerate(XhatList):
for ind in self.indsList[i]:
U, s, V = ZList[i]
Z = (U*s).dot(V.T)
self.assertEquals(Xhat[numpy.unravel_index(ind, Xhat.shape)], Z[numpy.unravel_index(ind, Xhat.shape)])
self.assertEquals(Xhat.nnz, self.indsList[i].shape[0])
def testPostProcess(self):
lmbda = 0.0
eps = 0.1
k = 20
matrixIterator = iter(self.matrixList)
iterativeSoftImpute = IterativeSoftImpute(lmbda, k=k, eps=eps, svdAlg="rsvd", postProcess=True)
ZList = iterativeSoftImpute.learnModel(matrixIterator)
for i, Z in enumerate(ZList):
U, s, V = Z
Xhat = (U*s).dot(V.T)
nptst.assert_array_almost_equal(Xhat, numpy.array(self.matrixList[i].todense()))
#Try case with iterativeSoftImpute.postProcessSamples < X.nnz
matrixIterator = iter(self.matrixList)
iterativeSoftImpute.postProcessSamples = int(self.matrixList[0].nnz/2)
ZList = iterativeSoftImpute.learnModel(matrixIterator)
for i, Z in enumerate(ZList):
U, s, V = Z
Xhat = (U*s).dot(V.T)
nptst.assert_array_almost_equal(Xhat, self.matrixList[i].todense(), 2)
#Try for larger lambda
iterativeSoftImpute.setRho(0.2)
ZList = iterativeSoftImpute.learnModel(matrixIterator)
for i, Z in enumerate(ZList):
U, s, V = Z
Xhat = (U*s).dot(V.T)
def initUV(self, X):
m = X.shape[0]
n = X.shape[1]
if self.initialAlg == "rand":
U = numpy.random.randn(m, self.k) * 0.1
V = numpy.random.randn(n, self.k) * 0.1
elif self.initialAlg == "svd":
logging.debug("Initialising with Randomised SVD")
U, s, V = RandomisedSVD.svd(X, self.k, self.p, self.q)
U = U * s
elif self.initialAlg == "softimpute":
logging.debug("Initialising with softimpute")
trainIterator = iter([X.toScipyCsc()])
rho = 0.01
learner = IterativeSoftImpute(rho, k=self.k, svdAlg="propack", postProcess=True)
ZList = learner.learnModel(trainIterator)
U, s, V = ZList.next()
U = U * s
elif self.initialAlg == "wrmf":
logging.debug("Initialising with wrmf")
learner = WeightedMf(self.k, w=self.w)
U, V = learner.learnModel(X.toScipyCsr())
else:
raise ValueError("Unknown initialisation: " + str(self.initialAlg))
U = numpy.ascontiguousarray(U)
V = numpy.ascontiguousarray(V)
return U, V
def initUV(self, X):
m = X.shape[0]
n = X.shape[1]
if self.initialAlg == "rand":
U = numpy.random.randn(m, self.k) * 0.1
V = numpy.random.randn(n, self.k) * 0.1
elif self.initialAlg == "svd":
logging.debug("Initialising with Randomised SVD")
U, s, V = RandomisedSVD.svd(X, self.k, self.p, self.q)
U = U * s
elif self.initialAlg == "softimpute":
logging.debug("Initialising with softimpute")
trainIterator = iter([X.toScipyCsc()])
rho = 0.01
learner = IterativeSoftImpute(rho,
k=self.k,
svdAlg="propack",
postProcess=True)
ZList = learner.learnModel(trainIterator)
U, s, V = ZList.next()
U = U * s
elif self.initialAlg == "wrmf":
logging.debug("Initialising with wrmf")
learner = WeightedMf(self.k, w=self.w)
U, V = learner.learnModel(X.toScipyCsr())
else:
raise ValueError("Unknown initialisation: " + str(self.initialAlg))
U = numpy.ascontiguousarray(U)
V = numpy.ascontiguousarray(V)
return U, V
def testWeightedLearning(self):
#See if the weighted learning has any effect
shape = (20, 20)
r = 20
numInds = 100
noise = 0.2
X = ExpSU.SparseUtils.generateSparseLowRank(shape, r, numInds, noise)
rho = 0.0
iterativeSoftImpute = IterativeSoftImpute(rho, k=10, weighted=True)
iterX = iter([X])
resultIter = iterativeSoftImpute.learnModel(iterX)
Z = resultIter.next()
iterativeSoftImpute = IterativeSoftImpute(rho, k=10, weighted=False)
iterX = iter([X])
resultIter = iterativeSoftImpute.learnModel(iterX)
Z2 = resultIter.next()
#Check results when rho=0
nptst.assert_array_almost_equal((Z[0]*Z[1]).dot(Z[2].T), (Z2[0]*Z2[1]).dot(Z2[2].T))
nptst.assert_array_almost_equal(Z[1], Z2[1])
#Then check non-uniform matrix - entries clustered around middle indices
shape = (20, 15)
numInds = 200
maxInd = (shape[0]*shape[1]-1)
nzInds = numpy.array(numpy.random.randn(numInds)*maxInd/4 + maxInd/2, numpy.int)
trainInds = nzInds[0:int(nzInds.shape[0]/2)]
testInds = nzInds[int(nzInds.shape[0]/2):]
trainInds = numpy.unique(numpy.clip(trainInds, 0, maxInd))
testInds = numpy.unique(numpy.clip(testInds, 0, maxInd))
trainX = ExpSU.SparseUtils.generateSparseLowRank(shape, r, trainInds, noise)
testX = ExpSU.SparseUtils.generateSparseLowRank(shape, r, testInds, noise)
#Error using weighted soft impute
#print("Running weighted soft impute")
rho = 0.5
iterativeSoftImpute = IterativeSoftImpute(rho, k=10, weighted=True)
iterX = iter([trainX])
resultIter = iterativeSoftImpute.learnModel(iterX)
Z = resultIter.next()
iterTestX = iter([testX])
predX = iterativeSoftImpute.predictOne(Z, testX.nonzero())
error = MCEvaluator.rootMeanSqError(testX, predX)
#print(error)
iterativeSoftImpute = IterativeSoftImpute(rho, k=10, weighted=False)
iterX = iter([trainX])
resultIter = iterativeSoftImpute.learnModel(iterX)
Z = resultIter.next()
iterTestX = iter([testX])
predX = iterativeSoftImpute.predictOne(Z, testX.nonzero())
error = MCEvaluator.rootMeanSqError(testX, predX)
def testLearnModel2(self):
#Test the SVD updating solution in the case where we get an exact solution
lmbda = 0.0
eps = 0.1
k = 20
matrixIterator = iter(self.matrixList)
iterativeSoftImpute = IterativeSoftImpute(lmbda, k=k, eps=eps, svdAlg="rsvd")
ZList = iterativeSoftImpute.learnModel(matrixIterator)
#Check that ZList is the same as XList
for i, Z in enumerate(ZList):
U, s, V = Z
Xhat = (U*s).dot(V.T)
nptst.assert_array_almost_equal(Xhat, self.matrixList[i].todense())
#Compare solution with that of SoftImpute class
rhoList = [0.1, 0.2, 0.5, 1.0]
for rho in rhoList:
iterativeSoftImpute = IterativeSoftImpute(rho, k=k, eps=eps, svdAlg="rsvd", updateAlg="zero")
matrixIterator = iter(self.matrixList)
ZList = iterativeSoftImpute.learnModel(matrixIterator)
rhos = numpy.array([rho])
softImpute = SoftImpute(rhos, k=k, eps=eps)
Z1 = softImpute.learnModel(self.matrixList[0])
Z2 = softImpute.learnModel(self.matrixList[1])
Z3 = softImpute.learnModel(self.matrixList[2])
ZList2 = [Z1, Z2, Z3]
for j, Zhat in enumerate(ZList):
U, s, V = Zhat
Z = (U*s).dot(V.T)
nptst.assert_array_almost_equal(Z, ZList2[j].todense())
#Also test with true solution Z = S_lambda(X + Z^\bot_\omega)
Zomega = numpy.zeros(self.matrixList[j].shape)
rowInds, colInds = self.matrixList[j].nonzero()
for i in range(self.matrixList[j].nonzero()[0].shape[0]):
Zomega[rowInds[i], colInds[i]] = Z[rowInds[i], colInds[i]]
U, s, V = ExpSU.SparseUtils.svdArpack(self.matrixList[j], 1, kmax=20)
lmbda = rho*numpy.max(s)
U, s, V = ExpSU.SparseUtils.svdSoft(numpy.array(self.matrixList[j]-Zomega+Z), lmbda)
tol = 0.1
self.assertTrue(numpy.linalg.norm(Z -(U*s).dot(V.T))**2 < tol)
def testLearnModel(self):
lmbda = 0.0
eps = 0.1
k = 10
matrixIterator = iter(self.matrixList)
iterativeSoftImpute = IterativeSoftImpute(lmbda, k=k, eps=eps, svdAlg="propack")
ZList = iterativeSoftImpute.learnModel(matrixIterator)
#Check that ZList is the same as XList
for i, Z in enumerate(ZList):
U, s, V = Z
Xhat = (U*s).dot(V.T)
nptst.assert_array_almost_equal(Xhat, numpy.array(self.matrixList[i].todense()))
#Compare solution with that of SoftImpute class
lmbdaList = [0.1, 0.2, 0.5, 1.0]
for lmbda in lmbdaList:
iterativeSoftImpute = IterativeSoftImpute(lmbda, k=k, eps=eps, svdAlg="propack", updateAlg="zero")
matrixIterator = iter(self.matrixList)
ZList = iterativeSoftImpute.learnModel(matrixIterator)
lmbdas = numpy.array([lmbda])
softImpute = SoftImpute(lmbdas, k=k, eps=eps)
Z1 = softImpute.learnModel(self.matrixList[0])
Z2 = softImpute.learnModel(self.matrixList[1])
Z3 = softImpute.learnModel(self.matrixList[2])
ZList2 = [Z1, Z2, Z3]
for j, Zhat in enumerate(ZList):
U, s, V = Zhat
Z = (U*s).dot(V.T)
nptst.assert_array_almost_equal(Z, ZList2[j].todense())
#Also test with true solution Z = S_lambda(X + Z^\bot_\omega)
Zomega = numpy.zeros(self.matrixList[j].shape)
rowInds, colInds = self.matrixList[j].nonzero()
for i in range(self.matrixList[j].nonzero()[0].shape[0]):
Zomega[rowInds[i], colInds[i]] = Z[rowInds[i], colInds[i]]
U, s, V = ExpSU.SparseUtils.svdSoft(numpy.array(self.matrixList[j]-Zomega+Z), lmbda)
tol = 0.1
self.assertTrue(numpy.linalg.norm(Z -(U*s).dot(V.T))**2 < tol)
def testModelSelect2(self):
rho = 0.1
shape = (20, 20)
r = 20
numInds = 100
noise = 0.2
X = ExpSU.SparseUtils.generateSparseLowRank(shape, r, numInds, noise)
X = X.tocsc()
U, s, V = numpy.linalg.svd(X.todense())
k = 15
iterativeSoftImpute = IterativeSoftImpute(rho, k=None, svdAlg="propack", updateAlg="initial")
rhos = numpy.linspace(0.5, 0.001, 5)
ks = numpy.array([5, 10, 15], numpy.int)
folds = 3
cvInds = []
for i in range(folds):
cvInds.append((numpy.arange(X.nnz), numpy.arange(X.nnz)))
meanTestErrors, stdTestErrors = iterativeSoftImpute.modelSelect(X, rhos, ks, cvInds)
self.assertAlmostEquals(numpy.linalg.norm(stdTestErrors), 0, 3)
meanTestErrors2 = numpy.zeros((rhos.shape[0], ks.shape[0]))
#Now compute errors manually
for j, k in enumerate(ks):
iterativeSoftImpute.setK(k)
for i, rho in enumerate(rhos):
iterativeSoftImpute.setRho(rho)
ZIter = iterativeSoftImpute.learnModel(iter([X]))
indList = [X.nonzero()]
outIterator = iterativeSoftImpute.predict(ZIter, indList)
Xhat = outIterator.next()
meanTestErrors2[i, j] = MCEvaluator.rootMeanSqError(X, Xhat)
nptst.assert_array_almost_equal(meanTestErrors, meanTestErrors2, 2)
def runExperiment(self):
"""
Run the selected clustering experiments and save results
"""
if self.algoArgs.runSoftImpute:
logging.debug("Running soft impute")
for svdAlg in self.algoArgs.svdAlgs:
if svdAlg == "rsvd" or svdAlg == "rsvdUpdate" or svdAlg == "rsvdUpdate2":
resultsFileName = self.resultsDir + "ResultsSoftImpute_alg=" + svdAlg + "_p=" + str(self.algoArgs.p)+ "_q=" + str(self.algoArgs.q) + "_updateAlg=" + self.algoArgs.updateAlg + ".npz"
else:
resultsFileName = self.resultsDir + "ResultsSoftImpute_alg=" + svdAlg + "_updateAlg=" + self.algoArgs.updateAlg + ".npz"
fileLock = FileLock(resultsFileName)
if not fileLock.isLocked() and not fileLock.fileExists():
fileLock.lock()
try:
learner = IterativeSoftImpute(svdAlg=svdAlg, logStep=self.logStep, kmax=self.algoArgs.kmax, postProcess=self.algoArgs.postProcess, weighted=self.algoArgs.weighted, p=self.algoArgs.p, q=self.algoArgs.q, verbose=self.algoArgs.verbose, updateAlg=self.algoArgs.updateAlg)
if self.algoArgs.modelSelect:
trainIterator = self.getTrainIterator()
#Let's find the optimal lambda using the first matrix
X = trainIterator.next()
logging.debug("Performing model selection, taking subsample of entries of size " + str(self.sampleSize))
X = SparseUtils.submatrix(X, self.sampleSize)
cvInds = Sampling.randCrossValidation(self.algoArgs.folds, X.nnz)
meanErrors, stdErrors = learner.modelSelect(X, self.algoArgs.rhos, self.algoArgs.ks, cvInds)
logging.debug("Mean errors = " + str(meanErrors))
logging.debug("Std errors = " + str(stdErrors))
modelSelectFileName = resultsFileName.replace("Results", "ModelSelect")
numpy.savez(modelSelectFileName, meanErrors, stdErrors)
logging.debug("Saved model selection grid as " + modelSelectFileName)
rho = self.algoArgs.rhos[numpy.unravel_index(numpy.argmin(meanErrors), meanErrors.shape)[0]]
k = self.algoArgs.ks[numpy.unravel_index(numpy.argmin(meanErrors), meanErrors.shape)[1]]
else:
rho = self.algoArgs.rhos[0]
k = self.algoArgs.ks[0]
learner.setK(k)
learner.setRho(rho)
logging.debug(learner)
trainIterator = self.getTrainIterator()
ZIter = learner.learnModel(trainIterator)
self.recordResults(ZIter, learner, resultsFileName)
finally:
fileLock.unlock()
else:
logging.debug("File is locked or already computed: " + resultsFileName)
if self.algoArgs.runSgdMf:
logging.debug("Running SGD MF")
resultsFileName = self.resultsDir + "ResultsSgdMf.npz"
fileLock = FileLock(resultsFileName)
if not fileLock.isLocked() and not fileLock.fileExists():
fileLock.lock()
try:
learner = IterativeSGDNorm2Reg(k=self.algoArgs.ks[0], lmbda=self.algoArgs.lmbdas[0], gamma=self.algoArgs.gammas[0], eps=self.algoArgs.eps)
if self.algoArgs.modelSelect:
# Let's find optimal parameters using the first matrix
learner.modelSelect(self.getTrainIterator().next(), self.algoArgs.ks, self.algoArgs.lmbdas, self.algoArgs.gammas, self.algoArgs.folds)
trainIterator = self.getTrainIterator()
trainIterator = self.getTrainIterator()
ZIter = learner.learnModel(trainIterator)
self.recordResults(ZIter, learner, resultsFileName)
finally:
fileLock.unlock()
else:
logging.debug("File is locked or already computed: " + resultsFileName)
logging.info("All done: see you around!")
