from em.gmm_penality import *

from sklearn import mixture

def getDataList():

fileNameList = []

dataList = ["amix1-est", "amix2-est", "golub-est"]

for file in dataList:

filePath = "./data/%s.dat" % (file)

fileNameList.append(filePath)

return fileNameList

def checkResult():

fileNameList = getDataList()

for fileName in fileNameList:

X = np.loadtxt(fileName)

searchK = 3

penalty = [0, 1, 2]

epoch = 2

maxLogLikelihood = float('-inf')

maxResult = None

maxK = 0

alpha = None

bestP = 0

for i in range(2, searchK):

k = i

for p in penalty:

for j in range(epoch):

model1 = GMMPenality(k, penalty = p)

model1.fit(X)

if model1.loglike > maxLogLikelihood:

maxLogLikelihood = model1.loglike

maxResult, _ = model1.predict(X)

maxK = k

alpha = model1.alpha

bestP = p

alphaSorted = sorted(model1.alpha.tolist(), reverse=True)

print("fileName = %s, k = %s, penalty = %s alpha = %s, loglike = %s" % (fileName.split("/")[-1], k, p, [round(p[0], 5) for p in alphaSorted], round(model1.loglike, 5)))

alpha, maxResult = changeLabel(alpha.reshape(1, -1).tolist()[0], maxResult)

print("myself GMM alpha = %s, loglikelihood = %s, bestP = %s"%

([round(a, 5) for a in alpha], round(maxLogLikelihood, 5), bestP))

# maxK = 3

model2 = mixture.BayesianGaussianMixture(n_components=maxK,covariance_type='full')

result2 = model2.fit_predict(X)

sklearnAlpha, result2 = changeLabel(model2.weights_.tolist(), result2)

result = np.sum(maxResult==result2)

percent = np.mean(maxResult==result2)

print("sklearn GMM alpha = %s, loglikelihood = %s"%

([round(a, 5) for a in sklearnAlpha], round(model2.lower_bound_, 5)))

print("succ = %s/%s"%(result, len(result2)))

print("succ = %s"%(percent))

print(maxResult[:20])

print(result2[:20])

suffix = ["tst", "val"]

for suf in suffix:

newfileName = fileName.replace("-est", "-%s"%suf)

newX = np.loadtxt(newfileName)

maxResult, loglike = model1.predict(newX)

print("fileName = %s, loglike = %s"%(newfileName.split("/")[-1],loglike))

def changeLabel(alpha, predict):

alphaSorted = sorted(alpha, reverse=True)

labelOld = []

for i in predict:

if i not in labelOld:

labelOld.append(i)

if len(labelOld) == len(alpha):

break

labelNew = sorted(labelOld)

for i, old in enumerate(labelOld):

predict[predict == old] = labelNew[i] + 100

return alphaSorted, predict - 100

if __name__ == "__main__":

checkResult()