abhishek-Kumar009

mutest, sigma2test = estimateGaussian(Xtest)
ptest = multivariateGaussian(Xtest, mutest, sigma2test)
pvaltest = multivariateGaussian(Xvaltest, mutest, sigma2test)

F1test, epsilontest = selectThreshHold(yvaltest, pvaltest)
print('\nBest epsilon and F1 are\n',epsilontest, F1test)

abhishek-Kumar009

outliersTest = ptest < epsilontest
listOfOl = findIndices(outliersTest)

print('\n\n Outliers are:\n',listOfOl)
print('\n\nNumber of outliers are: ',len(listOfOl))

abhishek-Kumar009

mutest, sigma2test = estimateGaussian(Xtest)

ptest = multivariateGaussian(Xtest, mutest, sigma2test)

pvaltest = multivariateGaussian(Xvaltest, mutest, sigma2test)


F1test, epsilontest = selectThreshHold(yvaltest, pvaltest)
print('\nBest epsilon and F1 are\n',epsilontest, F1test)

abhishek-Kumar009

newDataset = sio.loadmat('anomalyDataTest.mat')

Xtest = newDataset['X']
Xvaltest = newDataset['Xval']
yvaltest = newDataset['yval']

abhishek-Kumar009

plt.scatter(X[:, 0], X[:, 1], marker = "x")
plt.xlabel('Latency(ms)')
plt.ylabel('Throughput(mb/s)')
plt.scatter(X[listOfOutliers,0], X[listOfOutliers, 1], facecolors = 'none', edgecolors = 'r')
plt.show()

abhishek-Kumar009

count_outliers = len(listOfOutliers)
print('\n\nNumber of outliers:', count_outliers)
print('\n',listOfOutliers)

abhishek-Kumar009

def findIndices(binVec):
    l = []
    for i in range(len(binVec)):
        if binVec[i] == 1:
            l.append(i)
    return l

abhishek-Kumar009

def selectThreshHold(yval, pval):
    
    F1 = 0
    bestF1 = 0
    bestEpsilon = 0
    
    stepsize = (np.max(pval) - np.min(pval))/1000
        
    epsVec = np.arange(np.min(pval), np.max(pval), stepsize)
    noe = len(epsVec)

abhishek-Kumar009

def multivariateGaussian(X, mu, sigma2):
     n = np.size(sigma2, 1)
     m = np.size(sigma2, 0)
     #print(m,n)
     
     if n == 1 or m == 1:
        # print('Yes!')
         sigma2 = np.diag(sigma2[0, :])
     #print(sigma2)
     X = X - mu

abhishek-Kumar009

def estimateGaussian(X):
    n = np.size(X, 1)
    m = np.size(X, 0)
    mu = np.zeros((n, 1))
    sigma2 = np.zeros((n, 1))
    
    mu = np.reshape((1/m)*np.sum(X, 0), (1, n))
    sigma2 = np.reshape((1/m)*np.sum(np.power((X - mu),2), 0),(1, n))
    
    return mu, sigma2