def monitoring(matrix, rate, roundId, para):
startTime = time.clock()
logger.info('rate=%.2f, %2d-round starts.'%(rate, roundId + 1))
# sampling
logger.info('CS-PCA sampling...')
startTime = time.clock() # to record the running time for one round
(trainMatrix, observedMatrix, testMatrix) = CS_PCA.sampling(matrix, rate, roundId, para)
# CS-PCA algorithm
logger.info('CS-PCA estimation...')
recoveredMatrix = CS_PCA.recover(trainMatrix, observedMatrix, para)
runningTime = float(time.clock() - startTime)
# evaluate the estimation error
evalResult = evallib.evaluate(testMatrix, recoveredMatrix, para)
result = (evalResult, runningTime)
# dump the result at each rate
outFile = '%s%s%s_result_%.2f%s.tmp'%(para['outPath'], para['dataName'],
('_%s'%para['dataType'] if ('dataType' in para.keys()) else ''), rate,
'_round%02d'%(roundId + 1) if (para['rounds'] > 1) else '')
evallib.dumpresult(outFile, result)
logger.info('rate=%.2f, %2d-round done.'%(rate, roundId + 1))
logger.info('----------------------------------------------')
def executeOneSetting(matrix, density, roundId, sliceId, para):
(numUser, numService) = matrix.shape
dim = para['dimension']
# remove data entries to generate trainMatrix and testMatrix
seedID = roundId + sliceId * 100
(trainMatrix, testMatrix) = evallib.removeEntries(matrix, density, seedID)
(testVecX, testVecY) = np.where(testMatrix)
testVec = testMatrix[testVecX, testVecY]
# invocation to the prediction function
startTime = time.clock()
predictedMatrix = PMF.predict(trainMatrix, para)
runningTime = float(time.clock() - startTime)
# evaluate the prediction error
predVec = predictedMatrix[testVecX, testVecY]
evalResult = evallib.errMetric(testVec, predVec, para['metrics'])
result = (evalResult, runningTime)
# dump the result at each density
outFile = '%s%s_%s_result_%02d_%.2f_round%02d.tmp' % (
para['outPath'], para['dataName'], para['dataType'], sliceId + 1,
density, roundId + 1)
evallib.dumpresult(outFile, result)
logger.info('density=%.2f, sliceId=%02d, %2d-round done.'\
%(density, sliceId + 1, roundId + 1))
def executeOneSetting(matrix, density, roundId, sliceId, para):
(numUser, numService) = matrix.shape
dim = para['dimension']
# remove data entries to generate trainMatrix and testMatrix
seedID = roundId + sliceId * 100
(trainMatrix, testMatrix) = evallib.removeEntries(matrix, density, seedID)
(testVecX, testVecY) = np.where(testMatrix)
testVec = testMatrix[testVecX, testVecY]
# invocation to the prediction function
startTime = time.clock()
predictedMatrix = PMF.predict(trainMatrix, para)
runningTime = float(time.clock() - startTime)
# evaluate the prediction error
predVec = predictedMatrix[testVecX, testVecY]
evalResult = evallib.errMetric(testVec, predVec, para['metrics'])
result = (evalResult, runningTime)
# dump the result at each density
outFile = '%s%s_%s_result_%02d_%.2f_round%02d.tmp'%(para['outPath'],
para['dataName'], para['dataType'], sliceId + 1, density, roundId + 1)
evallib.dumpresult(outFile, result)
logger.info('density=%.2f, sliceId=%02d, %2d-round done.'\
%(density, sliceId + 1, roundId + 1))
def monitoring(matrix, rate, para):
startTime = time.clock()
logger.info("rate=%.2f starts." % rate)
# JGD algorithm
startTime = time.clock() # to record the running time for one round
trainingPeriod = para["trainingPeriod"]
trainMatrix = matrix[:, 0:trainingPeriod]
testMatrix = matrix[:, trainingPeriod:]
logger.info("monitor selection...")
(selectedMonitors, toEstimateNodes) = JGD.selectMonitor(trainMatrix, rate, para)
observedMatrix = testMatrix[selectedMonitors, :]
logger.info("JGD estimation...")
recoveredMatrix = JGD.recover(trainMatrix, observedMatrix, selectedMonitors, toEstimateNodes)
runningTime = float(time.clock() - startTime)
# evaluate the estimation error
evalResult = evallib.evaluate(testMatrix, recoveredMatrix, para)
result = (evalResult, runningTime)
# dump the result at each rate
outFile = "%s%s%s_result_%.2f.tmp" % (
para["outPath"],
para["dataName"],
"_%s" % para["dataType"] if ("dataType" in para.keys()) else "",
rate,
)
evallib.dumpresult(outFile, result)
logger.info("rate=%.2f done." % rate)
logger.info("----------------------------------------------")
def monitoring(matrix, rate, roundId, para):
startTime = time.clock()
logger.info('rate=%.2f, %2d-round starts.'%(rate, roundId + 1))
# sampling
logger.info('CS sampling...')
startTime = time.clock() # to record the running time for one round
# sorting in ascending and permutation
idx = np.argsort(-matrix[:, 0])
matrix = matrix[idx, :]
(trainMatrix, observedMatrix, testMatrix) = CS.sampling(matrix, rate, roundId, para)
# CS algorithm
logger.info('CS estimation...')
recoveredMatrix = CS.recover(trainMatrix, observedMatrix, para)
plot((matrix[:,30]), '-o')
plot(recoveredMatrix[:,30], '-x')
show()
# sys.exit()
runningTime = float(time.clock() - startTime)
# evaluate the estimation error
evalResult = evallib.evaluate(testMatrix, recoveredMatrix, para)
result = (evalResult, runningTime)
# dump the result at each rate
outFile = '%s%s%s_result_%.2f%s.tmp'%(para['outPath'], para['dataName'],
('_%s'%para['dataType'] if ('dataType' in para.keys()) else ''), rate,
'_round%02d'%(roundId + 1) if (para['rounds'] > 1) else '')
evallib.dumpresult(outFile, result)
logger.info('rate=%.2f, %2d-round done.'%(rate, roundId + 1))
logger.info('----------------------------------------------')
def executeOneSetting(tensor, density, roundId, para):
logger.info('density=%.2f, %2d-round starts.'%(density, roundId + 1))
(numUser, numService, numTime) = tensor.shape
# remove the entries of data to generate trainTensor and testTensor
(trainTensor, testTensor) = evallib.removeTensor(tensor, density, roundId, para)
# invocation to the prediction function
startTime = time.clock() # to record the running time for one round
predictedTensor = Average.predict(trainTensor, para)
runningTime = float(time.clock() - startTime) / numTime
# evaluate the prediction error
for sliceId in xrange(numTime):
testMatrix = testTensor[:, :, sliceId]
predictedMatrix = predictedTensor[:, :, sliceId]
(testVecX, testVecY) = np.where(testMatrix)
testVec = testMatrix[testVecX, testVecY]
predVec = predictedMatrix[testVecX, testVecY]
evalResult = evallib.errMetric(testVec, predVec, para['metrics'])
result = (evalResult, runningTime)
# dump the result at each density
outFile = '%s%s_%s_result_%02d_%.2f_round%02d.tmp'%(para['outPath'],
para['dataName'], para['dataType'], sliceId + 1, density, roundId + 1)
evallib.dumpresult(outFile, result)
logger.info('density=%.2f, %2d-round done.'%(density, roundId + 1))
logger.info('----------------------------------------------')
def executeOneSetting(tensor, density, roundId, para):
logger.info('density=%.2f, %2d-round starts.' % (density, roundId + 1))
(numUser, numService, numTime) = tensor.shape
dim = para['dimension']
# remove the entries of data to generate trainTensor and testTensor
(trainTensor, testTensor) = evallib.removeTensor(tensor, density, roundId,
para)
# invocation to the prediction function
startTime = time.clock() # to record the running time for one round
predictedTensor = NTF.predict(trainTensor, para)
runningTime = float(time.clock() - startTime) / numTime
# evaluate the prediction error
for sliceId in xrange(numTime):
testMatrix = testTensor[:, :, sliceId]
predictedMatrix = predictedTensor[:, :, sliceId]
(testVecX, testVecY) = np.where(testMatrix)
testVec = testMatrix[testVecX, testVecY]
predVec = predictedMatrix[testVecX, testVecY]
evalResult = evallib.errMetric(testVec, predVec, para['metrics'])
result = (evalResult, runningTime)
# dump the result at each density
outFile = '%s%s_%s_result_%02d_%.2f_round%02d.tmp' % (
para['outPath'], para['dataName'], para['dataType'], sliceId + 1,
density, roundId + 1)
evallib.dumpresult(outFile, result)
logger.info('density=%.2f, %2d-round done.' % (density, roundId + 1))
logger.info('----------------------------------------------')
def executeOneSetting(matrix, density, roundId, para):
logger.info('density=%.2f, %2d-round starts.' % (density, roundId + 1))
# remove data matrix
(trainMatrix, testMatrix) = evallib.removeEntries(matrix, density, roundId)
# QoS prediction
startTime = time.clock() # to record the running time for one round
predictedMatrix1 = UIPCC.UPCC(trainMatrix, para)
predictedMatrix2 = UIPCC.IPCC(trainMatrix, para)
predictedMatrix = UIPCC.UIPCC(trainMatrix, predictedMatrix1,
predictedMatrix2, para)
runningTime = float(time.clock() - startTime)
# evaluate the estimation error
evalResult = evallib.evaluate(testMatrix, predictedMatrix, para)
result = (evalResult, runningTime)
# dump the result at each density
outFile = '%s%s_%s_result_%.2f_round%02d.tmp' % (
para['outPath'], para['dataName'], para['dataType'], density,
roundId + 1)
evallib.dumpresult(outFile, result)
logger.info('density=%.2f, %2d-round done.' % (density, roundId + 1))
logger.info('----------------------------------------------')
def executeOneSetting(matrix, density, roundId, para):
logger.info('density=%.2f, %2d-round starts.'%(density, roundId + 1))
# remove data matrix
(trainMatrix, testMatrix) = evallib.removeEntries(matrix, density, roundId)
# QoS prediction
startTime = time.clock() # to record the running time for one round
predictedMatrix = NMF.predict(trainMatrix, para)
runningTime = float(time.clock() - startTime)
# evaluate the estimation error
evalResult = evallib.evaluate(testMatrix, predictedMatrix, para)
result = (evalResult, runningTime)
# dump the result at each density
outFile = '%s%s_%s_result_%.2f_round%02d.tmp'%(para['outPath'], para['dataName'],
para['dataType'], density, roundId + 1)
evallib.dumpresult(outFile, result)
logger.info('density=%.2f, %2d-round done.'%(density, roundId + 1))
logger.info('----------------------------------------------')
'topK': 10, # the parameter of TopK similar users or services, the default
# value is topK = 10 as in the reference paper
'dimension': 10, # dimenisionality of the latent factors
'etaInit': 0.01, # inital learning rate. We use line search
# to find the best eta at each iteration
'lambda': 30, # L2 regularization parameter
'alpha': 0.4, # the parameter of combination, 0.4 as in the reference paper
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('NIMF: Neighbourhood Integrated Matrix Factorization')
# load the dataset
dataMatrix = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE',
'NPRE'], # delete where appropriate
'density': np.arange(0.05, 0.31, 0.05), # matrix density
'rounds': 20, # how many runs are performed at each matrix density
'dimension': 10, # dimenisionality of the latent factors
'eta': 0.8, # learning rate
'lambda': 0.0002, # regularization parameter
'maxIter': 50, # the max iterations
'convergeThreshold': 5e-2, # stopping criteria for convergence
'beta': 0.3, # the controlling weight of exponential moving average
'saveTimeInfo': True, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('AMF: Adaptive Matrix Factorization [TPDS]')
# load the dataset
dataTensor = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataTensor, para)
logger.info('All done. Elaspsed time: ' +
utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'dataName': 'google-cluster-data', # set the dataset name
'dataType': 'cpu', # data type: cpu or memory
'dataSample': 'day-sample', # choose 'day-sample', 'week-sample', or 'all-data'
'outPath': 'result/', # output path for results
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE', 'NNPRE', 'SNR'], # evaluation metrics
'samplingRate': np.arange(0.1, 0.11, 0.05), # sampling rate
'rounds': 1, # how many runs to perform at each sampling rate
'transform': 'DCT', # transform base: 'DCT', 'DWT-haar', or 'DWT-db4'
'lmbda': 1e-4, # sparisty regularization parameter
'trainingPeriod': 12, # training time periods
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': False, # whether to save log into file
'debugMode': False, #whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('Compressive Sensing: [Luo et al., MobiCom\'2009].')
# load the dataset
dataMatrix = dataloader.load(para)
dataMatrix = dataMatrix[:,0:24]
# evaluate compressive monitoring algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'topK_U': 60, # the parameter of TopK similar users, the default value is
# topK_U = 60 as in the reference paper
'topK_S': 300, # the parameter of TopK similar users, the default value is
# topK_S = 300 as in the reference paper
'dimension': 10, # dimenisionality of the latent factors
'etaInit': 0.01, # inital learning rate. We use line search
# to find the best eta at each iteration
'lambda': 30, # L2 regularization parameter
'alpha': 15, # parameter of user and service regularization
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('EMF: [Lo et al., SCC 2012]')
# load the dataset
dataMatrix = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' +
utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'outPath': 'result/',
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE', 'NPRE'], # delete where appropriate
'density': np.arange(0.05, 0.31, 0.05), # matrix density
'rounds': 20, # how many runs are performed at each matrix density
'dimension': 10, # dimenisionality of the latent factors
'etaInit': 0.001, # inital learning rate. We use line search
# to find the best eta at each iteration
'lambda': 200, # regularization parameter
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('PMF: Probabilistic Matrix Factorization')
# load the dataset
dataTensor = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataTensor, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'dataPath': '../../../data/',
'dataName': 'dataset#2',
'dataType': 'rt', # set the dataType as 'rt' or 'tp'
'outPath': 'result/',
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE',
'NPRE'], # delete where appropriate
'density': np.arange(0.05, 0.31, 0.05), # matrix density
'rounds': 20, # how many runs are performed at each matrix density
'topK': 10, # the parameter of TopK similar users or services
'lambda': 0.8, # the combination coefficient of UPCC and IPCC
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('Approach: [UPCC, IPCC, UIPCC][TSC 2011]')
# load the dataset
dataTensor = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataTensor, para)
logger.info('All done. Elaspsed time: ' +
utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
# parameter config area
para = {
'dataPath': '../../../data/',
'dataName': 'dataset#2',
'dataType': 'tp', # set the dataType as 'rt' or 'tp'
'outPath': 'result/',
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE',
'NPRE'], # delete where appropriate
'density': np.arange(0.05, 0.31, 0.05), # matrix density
'rounds': 20, # how many runs are performed at each matrix density
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('Baseline approach: Average')
# load the dataset
dataTensor = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataTensor, para)
logger.info('All done. Elaspsed time: ' +
utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'topK_U': 60, # the parameter of TopK similar users, the default value is
# topK_U = 60 as in the reference paper
'topK_S': 300, # the parameter of TopK similar users, the default value is
# topK_S = 300 as in the reference paper
'dimension': 10, # dimenisionality of the latent factors
'etaInit': 0.01, # inital learning rate. We use line search
# to find the best eta at each iteration
'lambda': 30, # L2 regularization parameter
'alpha': 15, # parameter of user and service regularization
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('EMF: [Lo et al., SCC 2012]')
# load the dataset
dataMatrix = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'rounds': 20, # how many runs are performed at each matrix density
'dimension': 10, # dimenisionality of the latent factors
'eta': 0.0001, # learning rate
'alpha': 0.6, # the combination coefficient
'lambda': 5, # regularization parameter
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('LN-LFM: Latent Neighbor and Latent Factor Model')
# load the dataset
dataMatrix = dataloader.load(para)
# load the service location information
wsInfoList = dataloader.loadServInfo(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataMatrix, wsInfoList, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'rounds': 20, # how many runs are performed at each matrix density
'dimension': 10, # dimenisionality of the latent factors
'eta': 0.0001, # learning rate
'alpha': 0.6, # the combination coefficient
'lambda': 5, # regularization parameter
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('LN-LFM: Latent Neighbor and Latent Factor Model')
# load the dataset
dataMatrix = dataloader.load(para)
# load the service location information
wsInfoList = dataloader.loadServInfo(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataMatrix, wsInfoList, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'dataName': 'dataset#2',
'dataType': 'rt', # set the dataType as 'rt' or 'tp'
'outPath': 'result/',
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE', 'NPRE'], # delete where appropriate
'density': np.arange(0.05, 0.31, 0.05), # matrix density
'rounds': 20, # how many runs are performed at each matrix density
'dimension': 10, # dimenisionality of the latent factors
'lambda': 35, # regularization parameter
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('NTF: Non-negative Tensor Factorization [WWW 2014]')
# load the dataset
dataTensor = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataTensor, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'topK': 10, # the parameter of TopK similar users or services, the default
# value is topK = 10 as in the reference paper
'dimension': 10, # dimenisionality of the latent factors
'etaInit': 0.01, # inital learning rate. We use line search
# to find the best eta at each iteration
'lambda': 30, # L2 regularization parameter
'alpha': 0.4, # the parameter of combination, 0.4 as in the reference paper
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('NIMF: Neighbourhood Integrated Matrix Factorization')
# load the dataset
dataMatrix = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'outPath': 'result/',
'metrics': ['NDCG', 'Precision'], # delete where appropriate
'metric_parameter': [1, 5, 10, 50, 100],
'density': [0.01, 0.1, 0.3], # matrix density
'rounds': 10, # how many runs are performed at each matrix density
'dimension': 10, # dimenisionality of the latent factors
'etaInit': 0.01, # inital learning rate. We use line search
# to find the best eta at each iteration
'lambda': 0.1, # regularization parameter
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('PMF: Probabilistic Matrix Factorization')
# load the dataset
dataMatrix = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' +
utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
def load(para):
if para['dataName'] == 'dataset#1':
datafile = para['dataPath'] + para['dataName'] + '/' + para[
'dataType'] + 'Matrix.txt'
logger.info('Loading data: %s' % os.path.abspath(datafile))
dataMatrix = np.loadtxt(datafile)
logger.info('Data size: %d users * %d services'\
%(dataMatrix.shape[0], dataMatrix.shape[1]))
elif para['dataName'] == 'dataset#2':
datafile = para['dataPath'] + para['dataName'] + '/' + para[
'dataType'] + 'data.txt'
logger.info('Loading data: %s' % os.path.abspath(datafile))
dataMatrix = -1 * np.ones((142, 4500, 64))
fid = open(datafile, 'r')
for line in fid:
data = line.split(' ')
rt = float(data[3])
if rt > 0:
dataMatrix[int(data[0]), int(data[1]), int(data[2])] = rt
fid.close()
logger.info('Data size: %d users * %d services * %d timeslices'\
%(dataMatrix.shape[0], dataMatrix.shape[1], dataMatrix.shape[2]))
dataMatrix = preprocess(dataMatrix, para)
logger.info('Loading data done.')
logger.info('----------------------------------------------')
return dataMatrix
'outPath': 'result/',
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE', 'NPRE'], # delete where appropriate
'density': np.arange(0.05, 0.31, 0.05), # matrix density
'rounds': 20, # how many runs are performed at each matrix density
'dimension': 10, # dimenisionality of the latent factors
'eta': 0.8, # learning rate
'lambda': 0.0002, # regularization parameter
'maxIter': 50, # the max iterations
'convergeThreshold': 5e-2, # stopping criteria for convergence
'beta': 0.3, # the controlling weight of exponential moving average
'saveTimeInfo': True, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('AMF: Adaptive Matrix Factorization [TPDS]')
# load the dataset
dataTensor = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataTensor, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
# parameter config area
para = {'dataPath': '../data/', # data path
'dataName': 'google-cluster-data', # set the dataset name
'dataType': 'cpu', # data type: cpu or memory
'dataSample': 'day-sample', # choose 'day-sample', 'week-sample', or 'all-data'
'outPath': 'result/', # output path for results
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE', 'NNPRE', 'SNR'], # evaluation metrics
'samplingRate': np.arange(0.05, 0.95, 0.05), # sampling rate
'rounds': 20, # how many runs to perform at each sampling rate
'trainingPeriod': 12, # training time periods
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, #whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('Baseline approach.')
# load the dataset
dataMatrix = dataloader.load(para)
dataMatrix = dataMatrix[:,0:24]
# evaluate compressive monitoring algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
para = {'dataPath': '../../../data/',
'dataName': 'dataset#2',
'dataType': 'rt', # set the dataType as 'rt' or 'tp'
'outPath': 'result/',
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE', 'NPRE'], # delete where appropriate
'density': np.arange(0.05, 0.31, 0.05), # matrix density
'rounds': 20, # how many runs are performed at each matrix density
'topK': 10, # the parameter of TopK similar users or services
'lambda': 0.8, # the combination coefficient of UPCC and IPCC
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('Approach: [UPCC, IPCC, UIPCC][TSC 2011]')
# load the dataset
dataTensor = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataTensor, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
# parameter config area
para = {'dataPath': '../data/', # data path
'dataName': 'Orangelab_sense_temperature', # set the dataset name
'outPath': 'result/', # output path for results
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE', 'NNPRE', 'SNR'], # evaluation metrics
'samplingRate': np.arange(0.05, 0.96, 0.05), # sampling rate
'rounds': 1, # how many runs to perform at each sampling rate
'lmbda': 1e-5, # sparisty regularization parameter
'trainingPeriod': 33, # training time periods
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': False, # whether to save log into file
'debugMode': False, #whether to record the debug info
'parallelMode': False # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('CS-PCA: [Quer et al., TWC\'2012]')
# load the dataset
dataMatrix = dataloader.load(para)
# evaluate compressive monitoring algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'dataName': 'dataset#2',
'dataType': 'tp', # set the dataType as 'rt' or 'tp'
'outPath': 'result/',
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE',
'NPRE'], # delete where appropriate
'density': np.arange(0.05, 0.31, 0.05), # matrix density
'rounds': 20, # how many runs are performed at each matrix density
'dimension': 10, # dimenisionality of the latent factors
'lambda': 6000, # regularization parameter
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('NTF: Non-negative Tensor Factorization [WWW 2014]')
# load the dataset
dataTensor = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataTensor, para)
logger.info('All done. Elaspsed time: ' +
utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
'outPath': 'result/',
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE', 'NPRE'], # delete where appropriate
'density': np.arange(0.05, 0.31, 0.05), # matrix density
'rounds': 20, # how many runs are performed at each matrix density
'dimension': 10, # dimenisionality of the latent factors
'etaInit': 0.001, # inital learning rate. We use line search
# to find the best eta at each iteration
'lambda': 500, # regularization parameter
'maxIter': 300, # the max iterations
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': True, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('BiasedMF: Biased Matrix Factorization')
# load the dataset
dataMatrix = dataloader.load(para)
# evaluate QoS prediction algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')
para = {'dataPath': '../data/', # data path
'dataName': 'google-cluster-data', # set the dataset name
'dataType': 'cpu', # data type: cpu or memory
'dataSample': 'day-sample', # choose 'day-sample', 'week-sample', or 'all'
'outPath': 'result/', # output path for results
'metrics': ['MAE', 'NMAE', 'RMSE', 'MRE', 'NNPRE', 'SNR'], # delete where appropriate
'samplingRate': np.arange(0.05, 0.06, 0.05), # sampling rate
'monitorSelection': 'topW-Update', # monitor selection algorithm
# select from 'random', 'topW', 'topW-Update', 'batch-selection'
'trainingPeriod': 12, # training time periods
'saveTimeInfo': False, # whether to keep track of the running time
'saveLog': False, # whether to save log into file
'debugMode': False, # whether to record the debug info
'parallelMode': True # whether to leverage multiprocessing for speedup
}
startTime = time.time() # start timing
utils.setConfig(para) # set configuration
logger.info('==============================================')
logger.info('JGD: [Silvestri et al., ICDCS\'2015].')
# load the dataset
dataMatrix = dataloader.load(para)
dataMatrix = dataMatrix[:,0:24]
# evaluate compressive monitoring algorithm
evaluator.execute(dataMatrix, para)
logger.info('All done. Elaspsed time: ' + utils.formatElapsedTime(time.time() - startTime)) # end timing
logger.info('==============================================')