"""

李偉的第一個gmm模型用來做聚類

"""

print(__doc__)

from scipy.stats import multivariate_normal as MN

import numpy as np

from sklearn import preprocessing

from sklearn.datasets.samples_generator import make_blobs

from sklearn import metrics

from sklearn import cluster, datasets, mixture

from pprint import pprint

## INIT PROCESS

ITER = 50

CLUSTER_NUM = 2

centers = [[2, 2], [-2, -2], [2, -2],[-2,2]]

"""

DATA, LABEL = make_blobs(n_samples=1000, centers=centers, cluster_std=0.9)

"""

DATA, LABEL = datasets.make_circles(

n_samples=1000,

factor=.5,

noise=.05

)

FEAT_NUM = DATA.shape[1]

INS_NUM = DATA.shape[0]

MEAN = DATA.mean(axis=0)

MIN,MAX = DATA.min(axis=0),DATA.max(axis=0)

WEIGHT = np.random.random_sample((1,CLUSTER_NUM))

#WEIGHT = [0.3,0.7]

##DISTRI

DISTRI = [

MN(mean=np.random.uniform(MIN,MAX))

for i in range(CLUSTER_NUM)

]

for i in range(ITER):

dis_all = np.array([

[

DISTRI[c].pdf(d)

for c in range(CLUSTER_NUM)

]

for d in DATA

],dtype=np.float64)

dis_all = dis_all * WEIGHT

dis_all = dis_all.transpose()/dis_all.sum(axis=1)

dis_all = dis_all.transpose()

WEIGHT = dis_all.sum(axis=0,keepdims=1)/INS_NUM

print(WEIGHT)

label = np.argmax(dis_all,axis=1)

loss = metrics.adjusted_rand_score(LABEL, label)

print("iter %s loss %.3f" % (i,loss))

mu_mo = np.matmul(DATA.transpose(),dis_all)

mu_no = dis_all.sum(axis=0,keepdims=1)

mu_mat = mu_mo/mu_no

cov_mat = []

for c in range(CLUSTER_NUM):

#tmp = np.eye(FEAT_NUM,FEAT_NUM)

tmp = np.zeros([FEAT_NUM,FEAT_NUM])

for d in range(INS_NUM):

D = DATA[d]-mu_mat.transpose()[c]

D = np.expand_dims(D,axis=0)

tmp += (

np.matmul(

D.transpose(),

D

) * dis_all[d][c]

)

cov_mat.append(tmp/mu_no[0][c])

DISTRI = [

MN(mean=mu_mat.transpose()[i],cov=cov_mat[i])

for i in range(CLUSTER_NUM)

]

pprint(dis_all)

pprint(cov_mat)

#print([LABEL,np.argmax(dis_all,axis=1)])

label = np.argmax(dis_all,axis=1)

import matplotlib.pyplot as plt

from itertools import cycle

colors = cycle('bgrcmykbgrcmykbgrcmykbgrcmyk')

for k, col in zip(range(CLUSTER_NUM), colors):

my_members = label == k

plt.plot(DATA[my_members, 0], DATA[my_members, 1], col + '.')

plt.title('Estimated number of clusters: %d' % CLUSTER_NUM)

plt.show()