-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
255 lines (218 loc) · 8.44 KB
/
main.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
#coding=utf-8
import scipy.io as io
from sklearn.decomposition import PCA
import numpy as np
import cPickle as pk
from scipy.spatial.distance import euclidean
from DLA2 import DLA
from utils import *
import random
from updata_parameter import minbatchSGD,reduceDimension
from sklearn.model_selection import KFold
from sklearn.svm import LinearSVC,SVC
#from sklearn.metrics import confusion_matrix
from sklearn import metrics
from sklearn.model_selection import ShuffleSplit
from MFA import MFA
import matplotlib.pyplot as plt
'''
读取数据并分类
'''
def compute_count(data,range_):
count = 0
for i in range(len(data)):
if data[i] > range_[i][0] and data[i] < range_[i][1]:
count += 1
return count
def compute_range(data):
#获得最大值
max_ = np.max(data,axis=0)
min_ = np.min(data,axis=0)
return zip(min_,max_)
def batches(n_split,n):
batch = []
class_1 = np.arange(n/2)
class_2 = np.arange(n/2,n)
step = (n/2)/n_split
for s in range(n_split):
if s + step < n/2:
test_index = np.array([class_1[s:s+step],class_2[s:s+step]]).flatten()
#train_index = np.array([class_1[s+step:],class_2[s+step:]]).flatten()
train_index = np.array([np.delete(class_1,class_1[s:s+step]),np.delete(class_2,class_1[s:s+step])]).flatten()
#每次留10个做validation
#valid_index = np.array([train_index[0][:10],train_index[1][:10]]).flatten()
#train_index = np.array([train_index[0][10:],train_index[1][10:]]).flatten()
else:
test_index = np.array([class_1[s:], class_2[s:]]).flatten()
train_index = np.array([np.delete(class_1,class_1[s:]),np.delete(class_2,class_1[s:])]).flatten()
#valid_index = np.array([train_index[0][:10], train_index[1][:10]]).flatten()
#train_index = np.array([train_index[0][10:], train_index[1][10:]]).flatten()
batch.append([train_index,test_index])
return batch
#r1是320*10 r2是320*10 m是472*320
def compute(r1,r2,m):
#先把m分成两个236*320
return np.r_[np.dot(m[0:len(m)/2,:],r1),np.dot(m[len(m)/2:,:],r2)]
def remove_nan(m):
length = len(m)
return np.reshape(m[~np.isnan(m)],[length,-1])
def standardization(m):
return (m-np.mean(m,axis=0))/np.std(m,axis=0,ddof=1)
def Cross_Validation(cv,x,label,clf):
trues = []
pres = []
for train_index,test_index in cv.split(x):
trainX = x[train_index]
trainY = label[train_index]
test_x = x[test_index]
testY = label[test_index]
clf.fit(trainX,trainY)
pres.extend(clf.predict(test_x))
trues.extend(testY)
print metrics.classification_report(trues,pres)
def preprocessing(data,n):
#data是一个320*472*72
pre_matrix = np.zeros([n,472,72])
for i in range(np.shape(data)[-1]):
m = data[:, :, i].T
m = standardization(m)
m = remove_nan(m)
#m = 472*n_compoents
m = PCA(n_components=n).fit_transform(m)
pre_matrix[:,:,i] = m.T
return pre_matrix
#data是一个320*472*72的矩阵,做10折交叉验证
data = io.loadmat('./matlab/happy_angry_gabor_result.mat')
matrix = data['r']
#y = np.tile([-1,-1,-1,-1,1,1,1,1],59)
y = np.sort(np.tile([-1,1],236))
#ten_fold = ShuffleSplit(n_splits=10)
n_componts = 300
true = []
Pre = []
x = preprocessing(matrix,n_componts)
#先把数据转换成[472,n_compoent*72]
x = np.reshape(x,[472,-1])
batch_x = batches(n = len(x))
for train_index,valid_index,test_index in batch_x:
trainX = x[train_index]
trainY = y[train_index]
validX = x[valid_index]
validY = y[valid_index]
testX = x[test_index]
testY = y[test_index]
result = np.zeros([len(trainX),10,72],dtype=np.float)
trainX = np.reshape(trainX,[n_componts,len(trainX),72])
#trainX = preprocessing(trainX,n_componts)
projection = {}
#f = False
#matrix 是一个320*472*72维的矩阵, 其中320*236*72是happy, 320*236*72是angry
for i in range(np.shape(trainX)[-1]):
#print i
#取出一个 320*472的矩阵，并且转置472*320
m = trainX[:,:,i].T
#np.savetxt('m.txt',m)
#做standazition
#m = standardization(m)
#np.savetxt('m_.txt',m)
#print i
#m = remove_nan(m)
#m = PCA(n_components=n_componts).fit_transform(m)
r1,r2 = DLA(m,trainY,0.1,5,3,10)
projection[i] = [r1,r2]
r = compute(r1,r2,m)
#r = MFA(m,y,5,5,20)
#变成427*9的矩阵,保存在result中
#result[:,:,i] = np.dot(m,r)
result[:,:,i] = r
# if i == 7:
# f = True
#result = result.astype(dtype=np.float)
try:
#np.savetxt('mfaresult.txt',result)
pk.dump(projection,open('projection','wb'))
except:
#np.savez('mfaresult',result)
pass
#利用svm分类 有472个样本,每个样本的维度是9*72
#result = 472,9,72 转变成 472,9*72
data_matrix = np.reshape(result,[-1,np.shape(result)[1] * np.shape(result)[-1]])
#计算两类的中心点
mean_1 = np.mean(data_matrix[0:len(data_matrix)/2,:],axis=0)
mean_2 = np.mean(data_matrix[len(data_matrix)/2:,:],axis=0)
#np.savetxt('mean_1',mean_1)
#np.savetxt('mean_2',mean_2)
#range_1 = compute_range(data_matrix[0:len(data_matrix)/2])
#range_2 = compute_range(data_matrix[len(data_matrix)/2:])
#训练参数min-batch SGD 每次用一个样本来调参
#alpha = random.random()
#belta = random.random()
#mean_1 = np.loadtxt(open('mean_1'))
#mean_2 = np.loadtxt(open('mean_2'))
#projection = pk.load('projection')
#canshu = np.array([alpha,belta])
valid_1,valid_2 = reduceDimension(projection,validX)
#svm = LinearSVC()
svm = SVC()
features = np.r_['0,2',np.array(map(lambda a:euclidean(a,mean_1),valid_1)),np.array(map(lambda a:euclidean(a,mean_2),valid_2))].T
svm.fit(features,validY)
#belta_ = minbatchSGD(100, valid_1, valid_2, validY, mean_1, mean_2, canshu, 0.9,0.999,0.001)
#qda = QDA(data_matrix,trainY)
#建立分类器
#svm = LinearSVC()
#svm.fit(data_matrix,trainY)
test_matrix_1 = np.zeros([len(testX), 10, 72])
test_matrix_2 = np.zeros([len(testX), 10, 72])
#对test做变换 test是len(test),320*72--> len(test),320,72
testX = np.reshape(testX,[len(testX),n_componts,72])
for i in range(np.shape(testX)[-1]):
m = testX[:,:,i]
#m = standardization(m)
#m = remove_nan(m)
#m = PCA(n_components=n_componts).fit_transform(m)
r1,r2 = projection[i]
r_1 = np.dot(m,r1)
r_2 = np.dot(m,r2)
test_matrix_1[:,:,i] = r_1
test_matrix_2[:,:,i] = r_2
test_1= np.reshape(test_matrix_1,[-1,np.shape(test_matrix_1)[1]*np.shape(test_matrix_1)[-1]])
test_2= np.reshape(test_matrix_2, [-1, np.shape(test_matrix_2)[1] * np.shape(test_matrix_2)[-1]])
test_features = np.r_['0,2',np.array(map(lambda a:euclidean(a,mean_1),test_1)),np.array(map(lambda a:euclidean(a,mean_2),test_2))].T
test_predict = svm.predict(test_features)
#p_1 = svm.predict(test_1)
#p_2 = svm.predict(test_2)
# for i in range(len(p_1)):
# if p_1[i] == p_2[i]:test_predict.append(p_1[i])
# else:
# if euclidean(test_1[i], mean_1) > euclidean(test_2[i], mean_2):
# test_predict.append(1)
# elif euclidean(test_1[i], mean_1) > euclidean(test_2[i], mean_2):
# test_predict.append(-1)
#计算类标
# for i in range(len(test_1)):
# #计算距离
# if belta_[0]*euclidean(test_1[i],mean_1) > belta_[1]*euclidean(test_2[i],mean_2):
# #if compute_count(test_1[i],range_1) < compute_count(test_2[i],range_2):
# test_predict.append(1)
# #elif compute_count(test_1[i],range_1) > compute_count(test_2[i],range_2):
# elif belta_[0]*euclidean(test_1[i],mean_1) < belta_[1]*euclidean(test_2[i],mean_2):
# test_predict.append(-1)
# else:
# if np.random.uniform(0,1) > 0.5:
# test_predict.append(1)
# else:
# test_predict.append(-1)
true.extend(testY)
Pre.extend(test_predict)
cv = ShuffleSplit()
cv.split()
#C_range = np.logspace(-5, 15, 100,base=2)
#10折交叉验证
#ten_fold = KFold(n_splits=10)
#for c in C_range:
#svm = LinearSVC()
#print "C : %s"%c
#Cross_Validation(ten_fold,data_matrix,y,svm)
#svm.fit(data_matrix,trainY)
#svm.predict(test_)
print metrics.classification_report(true,Pre)