def clusterYogaTest():
print('YOGA')
traionSize = 300
valSize = 1000
testSize = 2000
seed = 0
distParams={'window': 3, 'psi': 0}
quantizationDistParams={'window': 2, 'psi': 0}
kmeansWindowSize=2
#generalClusterParams = {'dists_merger':clustering.singleLinkageUpdater, 'min_clusters':10}
generalClusterParams = {'dists_merger':clustering.singleLinkageUpdater, 'min_clusters':2}
pqClusterParams = {'k':199800,'quantizer_usage':clustering.QuantizerUsage.TOP_K_ONLY_AT_INITIALISATION}
quantizationDistParams={'window': 2, 'psi': 0}
qParams=[]
qParams.append(q.ProductQuantiserParameters(107,100,distParams=distParams,
subsetType=q.SubsetSelectionType.NO_OVERLAP,
kmeansWindowSize=kmeansWindowSize,
distanceCalculation=q.DISTANCECALCULATION.SYMMETRIC,
computeDistanceCorrection = False,
quantizationDistParams=quantizationDistParams,
#km_init="random",
max_iters=2))
XTrain,YTrain,XTest,YTest=loadYogaDataSetForClustering(traionSize, testSize+valSize)
XVal, YVal, XTest, YTest = take2RandDataParts(XTest,YTest,valSize, testSize, seed)
#print(YVal[1500:1700], np.unique(YVal))
distanceAndClusterTests(XTrain,XTest, qParams, distParams, generalClusterParams, pqClusterParams)
def clusterECG5000Test():
print('ECG')
traionSize = 500
valSize = 1500
testSize = 3000
seed = 0
distParams={'window': 2, 'psi': 0}
quantizationDistParams={'window': 1, 'psi': 0}
nwDistParams={'window': 2, 'psi': 0}
kmeansWindowSize=2
#generalClusterParams = {'dists_merger':clustering.singleLinkageUpdater, 'min_clusters':10}
generalClusterParams = {'dists_merger':clustering.singleLinkageUpdater, 'min_clusters':20}
pqClusterParams = {'k':120000,'quantizer_usage':clustering.QuantizerUsage.TOP_K_ONLY_AT_INITIALISATION}
qParams=[]
qParams.append(q.ProductQuantiserParameters(70,200,distParams=distParams,
subsetType=q.SubsetSelectionType.DOUBLE_OVERLAP,
kmeansWindowSize=kmeansWindowSize,
distanceCalculation=q.DISTANCECALCULATION.SYMMETRIC,
computeDistanceCorrection = True,
quantizationDistParams=quantizationDistParams,
#km_init="random",
max_iters=2))
XTrain,YTrain,XTest,YTest=loadEcg500DataSetForClustering(traionSize, testSize+valSize)
XVal, YVal, XTest, YTest = take2RandDataParts(XTest,YTest,valSize, testSize, seed)
#print(YVal[1500:1700], np.unique(YVal))
rclust = createRealClustering(YVal,5)
rclust = None
#print(YVal)
distanceAndClusterTests(XTrain,XTest, qParams, distParams, generalClusterParams, pqClusterParams,groundTruth=rclust)
def clusterElectricalDevicesTest():
print('ELD')
traionSize = 1000
valSize = 2000
testSize = 5000
seed = 0
distParams={'window': 15, 'psi': 0}
quantizationDistParams={'window':4, 'psi': 0}
nwDistParams={'window': 2, 'psi': 0}
kmeansWindowSize=2
#generalClusterParams = {'dists_merger':clustering.singleLinkageUpdater, 'min_clusters':10}
generalClusterParams = {'dists_merger':None, 'min_clusters':7}
pqClusterParams = {'k':199800,'quantizer_usage':clustering.QuantizerUsage.ONLY_APPROXIMATES}
qParams=[]
qParams.append(q.ProductQuantiserParameters(48,200,distParams=distParams,
subsetType=q.SubsetSelectionType.NO_OVERLAP,
kmeansWindowSize=kmeansWindowSize,
distanceCalculation=q.DISTANCECALCULATION.ASYMMETRIC,
computeDistanceCorrection = True,
quantizationDistParams=quantizationDistParams,
#km_init="random",
max_iters=2))
#qParams.append(q.ProductQuantiserParameters(12,2,computeDistanceCorrection=False))
qNWParams=[]
qNWParams.append(q.ProductQuantiserParameters(20,40,
quantizerType=q.QuantizerType.PQNeedlemanWunsch,nwDistParams=nwDistParams, distParams=distParams, subsetType=q.SubsetSelectionType.DOUBLE_OVERLAP))
XTrain,YTrain,XTest,YTest=loadElectricDevicesDataSetForClustering(traionSize, testSize+valSize)
XVal, YVal, XTest, YTest = take2RandDataParts(XTest,YTest,valSize, testSize, seed)
#import matplotlib as pl
#pl.pyplot.plot(XVal[3])
#pl.pyplot.plot(XVal[4])
#pl.pyplot.plot(XVal[5])
#pl.pyplot.show()
#print(YVal[1500:1700], np.unique(YVal))
rclust =createRealClustering(YVal,7)
rclust =None
distanceAndClusterTests(XTrain,XTest, qParams, distParams, generalClusterParams, pqClusterParams, groundTruth=rclust)
def clusterStarLight30AsymTest():
print('STARLIGHT30ASYMMETRIC')
print(50)
traionSize = 500
valSize = 2000
testSize = 5000
seed = 0
distParams={'window': 30, 'psi': 0}
quantizationDistParams={'window': 4, 'psi': 0}
nwDistParams={'window': 2, 'psi': 0}
kmeansWindowSize=2
#generalClusterParams = {'dists_merger':clustering.singleLinkageUpdater, 'min_clusters':10}
generalClusterParams = {'dists_merger':clustering.singleLinkageUpdater, 'min_clusters':10}
pqClusterParams = {'k':199800,'quantizer_usage':clustering.QuantizerUsage.TOP_K_ONLY_AT_INITIALISATION}
qParams=[]
qParams.append(q.ProductQuantiserParameters(128,100,distParams=distParams,
subsetType=q.SubsetSelectionType.NO_OVERLAP,
kmeansWindowSize=kmeansWindowSize,
distanceCalculation=q.DISTANCECALCULATION.ASYMMETRIC,
computeDistanceCorrection = True,
quantizationDistParams=quantizationDistParams,
km_init="random",
max_iters=0))
#qParams.append(q.ProductQuantiserParameters(32,2,computeDistanceCorrection=False))
qNWParams=[]
qNWParams.append(q.ProductQuantiserParameters(20,40,
quantizerType=q.QuantizerType.PQNeedlemanWunsch,nwDistParams=nwDistParams, distParams=distParams, subsetType=q.SubsetSelectionType.DOUBLE_OVERLAP))
qNWPileParams=[]
qNWPileParams.append(q.ProductQuantiserParameters(20,20,
quantizerType=q.QuantizerType.VQNeedlemanWunsch,nwDistParams=nwDistParams, distParams=distParams, subsetType=q.SubsetSelectionType.DOUBLE_OVERLAP))
XTrain,YTrain,XTest,YTest=loadStarLightDataSetForClustering(traionSize, testSize+valSize)
XVal, YVal, XTest, YTest = take2RandDataParts(XTest,YTest,valSize, testSize, seed)
#print(YVal[1500:1700], np.unique(YVal))
distanceAndClusterTests(XTrain,XTest, qParams, distParams, generalClusterParams, pqClusterParams)
def clusterStarLight17Test():
print('STARLIGHT17')
traionSize = 500
valSize = 2000
testSize = 5000
seed = 0
distParams={'window': 17, 'psi': 0}
quantizationDistParams={'window': 4, 'psi': 0}
nwDistParams={'window': 2, 'psi': 0}
kmeansWindowSize=2
#generalClusterParams = {'dists_merger':clustering.singleLinkageUpdater, 'min_clusters':10}
generalClusterParams = {'dists_merger':clustering.singleLinkageUpdater, 'min_clusters':3}
pqClusterParams = {'k':199800,'quantizer_usage':clustering.QuantizerUsage.TOP_K_ONLY_AT_INITIALISATION}
print()
qParams=[]
qParams.append(q.ProductQuantiserParameters(128,100,distParams=distParams,
subsetType=q.SubsetSelectionType.NO_OVERLAP,
kmeansWindowSize=kmeansWindowSize,
distanceCalculation=q.DISTANCECALCULATION.SYMMETRIC,
computeDistanceCorrection = True,
#km_init="random",
quantizationDistParams=quantizationDistParams,
max_iters=2))
#qParams.append(q.ProductQuantiserParameters(9,4,computeDistanceCorrection=False))
XTrain,YTrain,XTest,YTest=loadStarLightDataSetForClustering(traionSize, testSize+valSize)
XVal, YVal, XTest, YTest = take2RandDataParts(XTest,YTest,valSize, testSize, seed)
#print(YVal[1500:1700], np.unique(YVal))
#rclust = createRealClustering(YVal,3)
rclust = None
distanceAndClusterTests(XTrain,XTest, qParams, distParams, generalClusterParams, pqClusterParams, groundTruth=rclust)