def initialDBScanSciLearn(self):
db = DBSCAN(eps=8, min_samples=self.minPts,
algorithm='brute').fit(self.buffer)
clusters = db.labels_
self.buffer['clusters'] = clusters
clusterNumber = np.unique(clusters)
for clusterId in clusterNumber:
if (clusterId != -1):
cl = self.buffer[self.buffer['clusters'] == clusterId]
cl = cl.drop('clusters', axis=1)
sample = Sample(cl.iloc[0].tolist())
mc = MicroCluster(sample, self.currentTimestamp, self.lamb)
for sampleNumber in range(len(cl[1:])):
sample = Sample(cl.iloc[sampleNumber].tolist())
mc.insertSample(sample, self.currentTimestamp)
self.pMicroCluster.insert(mc)
def initWithoutDBScan(self):
sample = Sample(self.buffer.iloc[0].values, 0)
sample.set_timestamp(1)
mc = MicroCluster(1, self.lamb, self.pMicroCluster.N + 1)
maxEpsilon = 0
for sampleNumber in range(0, len(self.buffer)):
sample = Sample(self.buffer.iloc[sampleNumber].values,
sampleNumber)
sample.set_timestamp(sampleNumber + 1)
mc.insert_sample(sample)
if mc.radius > maxEpsilon:
maxEpsilon = mc.radius
# print 'New max: {}'.format(mc.radius)
self.pMicroCluster.insert(mc)
if isinstance(self.epsilon, str):
if self.epsilon == 'auto':
self.epsilon = self.pMicroCluster.clusters[
0].radius * self.radiusFactor
self.epsilon = maxEpsilon
def initWithoutDBScan(self):
sample = Sample(self.buffer.iloc[0].values, 0)
sample.setTimestamp(1)
mc = MicroCluster(1, self.lamb)
for sampleNumber in range(0, len(self.buffer)):
sample = Sample(self.buffer.iloc[sampleNumber].values,
sampleNumber)
sample.setTimestamp(sampleNumber + 1)
mc.insertSample(sample, self.currentTimestamp)
self.pMicroCluster.insert(mc)
if isinstance(self.epsilon, str):
if self.epsilon == 'auto':
self.epsilon = self.pMicroCluster.clusters[
0].radius * self.radiusFactor
def initWithoutDBScan(self):
sample = Sample(self.buffer[0], 0)
sample.setTimestamp(1)
mc = MicroCluster(1, self.lamb, self.pMicroCluster.N + 1)
maxEpsilon = 0
for sampleNumber in range(0, len(self.buffer)):
sample = Sample(self.buffer[sampleNumber], sampleNumber)
sample.setTimestamp(sampleNumber + 1)
mc.insertSample(sample, self.currentTimestamp)
if mc.radius > maxEpsilon:
maxEpsilon = mc.radius
self.pMicroCluster.insert(mc)
if isinstance(self.epsilon, str):
if self.epsilon == 'auto':
self.epsilon = self.pMicroCluster.clusters[
0].radius * self.radiusFactor
self.epsilon = maxEpsilon
def runOnNewSample(self, sample):
if simulation:
self.currentTimestamp += 1
sample.setTimestamp(self.currentTimestamp)
else:
self.currentTimestamp = time.time()
### INITIALIZATION PHASE ###
if not self.inizialized:
self.buffer.append(sample)
if (len(self.buffer) >= self.numberInitialSamples):
self.resetLearningImpl()
self.initialDBScanSciLearn()
self.inizialized = True
### MERGING PHASE ###
else:
merged = False
TrueOutlier = True
returnOutlier = True
if len(self.pMicroCluster.clusters) != 0:
closestMicroCluster = self.nearestCluster(
sample, self.currentTimestamp, kind='cluster')
backupClosestCluster = copy.deepcopy(closestMicroCluster)
backupClosestCluster.insertSample(sample,
self.currentTimestamp)
if (backupClosestCluster.radius <= self.epsilon):
closestMicroCluster.insertSample(sample,
self.currentTimestamp)
merged = True
TrueOutlier = False
returnOutlier = False
self.updateAll(closestMicroCluster)
if not merged and len(self.oMicroCluster.clusters) != 0:
closestMicroCluster = self.nearestCluster(
sample, self.currentTimestamp, kind='outlier')
backupClosestCluster = copy.deepcopy(closestMicroCluster)
backupClosestCluster.insertSample(sample,
self.currentTimestamp)
if (backupClosestCluster.radius <= self.epsilon):
closestMicroCluster.insertSample(sample,
self.currentTimestamp)
merged = True
if (closestMicroCluster.weight > self.beta * self.mu):
self.oMicroCluster.clusters.pop(
self.oMicroCluster.clusters.index(
closestMicroCluster))
self.pMicroCluster.insert(closestMicroCluster)
self.updateAll(closestMicroCluster)
if not merged:
newOutlierMicroCluster = MicroCluster(1, self.lamb)
newOutlierMicroCluster.insertSample(sample,
self.currentTimestamp)
for clusterTest in self.pMicroCluster.clusters:
if np.linalg.norm(
clusterTest.center -
newOutlierMicroCluster.center) < 2 * self.epsilon:
TrueOutlier = False
if TrueOutlier:
self.oMicroCluster.insert(newOutlierMicroCluster)
self.updateAll(newOutlierMicroCluster)
else:
self.pMicroCluster.insert(newOutlierMicroCluster)
self.updateAll(newOutlierMicroCluster)
returnOutlier = False
if self.currentTimestamp % self.tp == 0:
for cluster in self.pMicroCluster.clusters:
if cluster.weight < self.beta * self.mu:
self.pMicroCluster.clusters.pop(
self.pMicroCluster.clusters.index(cluster))
for cluster in self.oMicroCluster.clusters:
creationTimestamp = cluster.creationTimeStamp
xs1 = math.pow(
2, -self.lamb * (self.currentTimestamp -
creationTimestamp + self.tp)) - 1
xs2 = math.pow(2, -self.lamb * self.tp) - 1
xsi = xs1 / xs2
if cluster.weight < xsi:
self.oMicroCluster.clusters.pop(
self.oMicroCluster.clusters.index(cluster))
if self.exportVariables:
record = {
'pMicroClusters': self.pMicroCluster.clusters,
'oMicroClusters': self.oMicroCluster.clusters,
'result': returnOutlier
}
return record
else:
return returnOutlier