aiops/AnomalyDetection/utils/utils.py [112:206]:
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -
def insertNormality(datas,quickRatio,quickNumRatio,quickTimeRange,labels,exeTime,edges):
    exeTime=exeTime[1:]
    edges=edges[:-1]
    #exeTime = np.array([5, 15, 25, 35, 55, 90, 130, 170, 210, 255, 305, 355, 405])
    #edges = [0,10, 20, 30, 40, 70, 110, 150, 190, 230, 280, 330, 380]
    for i, data in enumerate(datas):
        length, interval = data.shape
        changeNum = math.ceil(length * quickRatio)
        positions = [random.randint(0, length - 1) for i in range(changeNum)]
        labels[i, positions] = -1.
        for position in positions:
            quickNum = np.random.random(interval - 1)
            quickNum[0] = 0
            quickNum /= quickNum.sum()
            quickNum *= quickNumRatio
            quickTime = np.random.randint(quickTimeRange[0], quickTimeRange[1], size=interval - 1)
            presTime = np.maximum(exeTime - quickTime,0)
            slowDur = random.randint(2, 5)
            for k in range(min(slowDur, length - position)):
                labels[i, position + k] = -1.
                for j in range(interval - 1):
                    transfer = min(datas[i, position + k, j], quickNum[j])
                    datas[i, position + k, j] = max(datas[i, position + k, j] - quickNum[j], 0)
                    target = len(edges)-1
                    for edge in reversed(edges):
                        if presTime[j] >= edge:
                            break
                        target -= 1
                    datas[i, position + k, target] += transfer
    return datas,labels

def insertAnomaly(datas,anomalyRatio,slowNumRatio,slowTimeRange,labels,exeTime,edges):
    #labels=np.zeros((datas.shape[0],datas.shape[1]))
    exeTime=exeTime[:-1]
    edges=edges[1:]
    #exeTime=np.array([5,15,25,35,55,90,130,170,210,255,305,355,405])
    #edges=[10,20,30,40,70,110,150,190,230,280,330,380,430]
    for i,data in enumerate(datas):
        length,interval=data.shape
        anomalyNum=math.ceil(length*anomalyRatio)
        positions=[random.randint(0,length-1) for i in range(anomalyNum)]
        for position in positions:
            slowNum=np.random.random(interval-1)
            slowNum[-1]=0
            slowNum/=slowNum.sum()
            slowNum*=slowNumRatio
            slowTime=np.random.randint(slowTimeRange[0],slowTimeRange[1],size=interval-1)
            presTime=exeTime+slowTime
            slowDur = random.randint(2, 5)
            for k in range(min(slowDur,length-position)):
                flag = True
                for j in range(interval - 1):
                    transfer=min(datas[i,position+k,j],slowNum[j])
                    datas[i,position + k, j] = max(datas[i,position + k, j] - slowNum[j], 0)
                    target = 0
                    if transfer>0:
                        flag=False
                    for edge in edges:
                        if presTime[j] < edge:
                            break
                        target += 1
                    datas[i,position + k, target] += transfer
                if not flag:
                    labels[i,position+k] = 1.
    return datas,labels

def insertAnomaly2(datas,anomalyRatio,slowNumRatio,slowTimeRange,exeTime,edges):
    labels=np.zeros(datas.shape[0])
    exeTime = exeTime[:-1]
    edges=edges[1:]
    #exeTime=np.array([5,15,25,35,55,90,130,170,210,255,305,355,405])
    #edges=[10,20,30,40,70,110,150,190,230,280,330,380,430]
    length,interval=datas.shape
    anomalyNum=math.ceil(length*anomalyRatio)
    positions=[random.randint(0,length-1) for i in range(anomalyNum)]
    labels[positions]=1.
    for position in positions:
        slowNum=np.random.random(interval-1)
        slowNum[-1]=0
        slowNum/=slowNum.sum()
        slowNum*=slowNumRatio
        slowTime=np.random.randint(slowTimeRange[0],slowTimeRange[1],size=interval-1)
        presTime=exeTime+slowTime
        slowDur=random.randint(2,8)
        for k in range(min(slowDur,length-position)):
            labels[position+k]=1.
            for j in range(interval-1):
                datas[position+k,j]=max(datas[position+k,j]-slowNum[j],0)
                target=0
                for edge in edges:
                    if presTime[j]<edge:
                        break
                    target+=1
                datas[position+k,target]+=slowNum[j]
    return datas,labels
- - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - - -