forked from dunkyfool/walabot
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathmain.py
249 lines (219 loc) · 7.31 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
from model.NNWalabot3 import *
from tool.log2 import *
import time
import matplotlib.pyplot as plt
import sys
import os
import cv2
def quick_scan(X,y,X1,y1,e,lr_range=[-3.0,-3.7],reg_range=[1,0],epoch=1,sample=10):
results = {}
learning_rates = lr_range
regularization_strengths = reg_range
#best_val = -1
best_val = np.inf
best_lr, best_reg = None, None
tic = time.time()
for i in range(sample):
print '['+str(i)+']'
# random choose lr & reg within the range
lr = 10**np.random.uniform(learning_rates[0],learning_rates[1])
reg = 10**np.random.uniform(regularization_strengths[0],regularization_strengths[1])
print 'lr:\t'+str(lr)
print 'reg:\t'+str(reg)
net = modelX()
net.loss(X,y,X1,y1,e,mode='train',lr=lr,reg=reg,batch=10,epoch=epoch)
#results[(lr,reg)]=(net.X_acc_history[-1],net.X1_acc_history[-1])
#results[(lr,reg)]=(net.X_loss_history[-1],net.X1_loss_history[-1])
cmd = 'echo '+str(lr)+' '+str(reg)+' '+str(net.X_acc_history[-1])+' '+str(net.X1_acc_history[-1])+'>>qs.log'
os.system(cmd)
if best_val > net.X1_acc_history[-1]:
best_val = net.X1_acc_history[-1]
best_lr = lr
best_reg = reg
# shrink mem
#del net
toc = time.time()
print 'Total Training: computed in %fs' % (toc - tic)
print 'Best Validation Record %.5f' % (best_val)
print 'Best Validation learning rate %.10f' % (best_lr)
print 'Best Validation regularization %.10f' % (best_reg)
def quick_scan_plot():
##########################################
# Visualize the cross-validation results #
##########################################
import math
data=None
results={}
data = np.loadtxt('qs.log')
for i in range(data.shape[0]):
results[(data[i,0],data[i,1])] = (data[i,2],data[i,3])
x_scatter = [math.log10(x[0]) for x in results]
y_scatter = [math.log10(x[1]) for x in results]
print x_scatter,y_scatter
# plot training accuracy
marker_size = 100
colors = [results[x][0] for x in results]
plt.subplot(2, 1, 1)
plt.scatter(x_scatter, y_scatter, marker_size, c=colors)
plt.colorbar()
plt.xlabel('log learning rate')
plt.ylabel('log regularization strength')
plt.title('Training Accuracy')
# plot validation accuracy
colors = [results[x][1] for x in results] # default size of markers is 20
plt.subplot(2, 1, 2)
plt.scatter(x_scatter, y_scatter, marker_size, c=colors)
plt.colorbar()
plt.xlabel('log learning rate')
plt.ylabel('log regularization strength')
plt.title('Validation Accuracy')
plt.savefig("loss.png")
plt.show()
def marathon(X,y,X1,y1,X2,y2,e,lr=1e-4,reg=1e-4,epoch=20):
net = modelX()
#'''
opt = raw_input('Restart training??[y/n]')
if opt=='y':
net.loss(X,y,X1,y1,e,mode='train',lr=lr,reg=reg,batch=10,epoch=epoch)
elif opt=='n':
net.loss(X,y,X1,y1,e,mode='train',lr=lr,reg=reg,batch=10,epoch=epoch,opt=False)
#'''
#_net = modelX()
#_net.loss(X2,y2,X1,y1,mode='test')
####################################################
# Visualize training loss and train / val accuracy #
####################################################
plt.subplot(2, 1, 1)
plt.title('Training loss')
plt.plot(net.X_loss_history, 'o')
plt.xlabel('Iteration')
plt.subplot(2, 1, 2)
plt.title('Accuracy')
plt.plot(net.X_acc_history, '-o', label='train')
plt.plot(net.X1_acc_history, '-o', label='val')
plt.plot([0.5] * len(net.X1_acc_history), 'k--')
plt.xlabel('Epoch')
plt.legend(loc='lower right')
plt.gcf().set_size_inches(15, 12)
plt.savefig("graph.png")
plt.show()
pass
def review(X,y,X1,y1,e):
net = modelX()
_,_ = net.loss(X,y,X1,y1,e,mode='test')
def duel(X,y,X1,y1,e):
net = modelX()
org_output,org_output1 = net.loss(X,y,X1,y1,e,mode='test')
#for i in range(1024):
# if y[0,i]!=1 and y[0,i]!=0:
# print y[0,i]
#raw_input('pause')
for _ in range(org_output.shape[0]):
idx = int(np.random.rand()* org_output.shape[0])
#print idx
tmp = np.ceil(org_output[idx] * (org_output[idx]>0.5))
#tmp = org_output[idx]
# for i in range(1024):
# if tmp[i]!=1 and tmp[i]!=0:
# print i, tmp[i]
# raw_input('pause')
pred_pic = tmp.reshape(32,32)*255
corr_pic = y[idx].reshape(32,32)*255
plt.clf()
plt.subplot(121)
plt.imshow(pred_pic,cmap='Greys_r')
plt.subplot(122)
plt.imshow(corr_pic,cmap='Greys_r')
plt.savefig('data/compare_img/{0}.png'.format(idx))
def mask_input(X,y,X1,y1,e):
idx = 800#int(sys.argv[1])
#print idx,type(idx)
span = 3
#mask_num = 1024
#interval = X.shape[2]/mask_num
#mask = np.zeros((1,1,mask_num,40))
net = modelX()
inp = X[idx:idx+3,:,:,:].reshape((span/1,1,2048,40))
oup = y[idx:idx+3].reshape((span/1,-1))
#inp = X[idx::-40,:,:,:].reshape((span/1,1,2048,40))
#oup = y[idx::-40].reshape((span/1,-1))
#print input[:,:,mask_num*i:mask_num*(i+1),:].shape
#raw_input()
# input[:,:,mask_num*i:mask_num*(i+1),:] = mask
#print input.shape
#print output.shape
#raw_input()
pred, _ = net.loss(X,y,inp,oup,e,mode='test')
#'''
tmp = np.ceil(pred * (pred>0.5))
pred_pic = tmp.reshape(3,32,32)*255
corr_pic = oup.reshape(3,32,32)*255
plt.clf()
plt.subplot(231)
plt.imshow(pred_pic[0],cmap='Greys_r')
plt.subplot(232)
plt.imshow(pred_pic[1],cmap='Greys_r')
plt.subplot(233)
plt.imshow(pred_pic[2],cmap='Greys_r')
plt.subplot(234)
plt.imshow(corr_pic[0],cmap='Greys_r')
plt.subplot(235)
plt.imshow(corr_pic[1],cmap='Greys_r')
plt.subplot(236)
plt.imshow(corr_pic[2],cmap='Greys_r')
plt.savefig('data/mask_input/{0}.png'.format(0))
#'''
if __name__=='__main__':
os.system('cp tmp tmp.bak && echo > tmp')
e = mp.Event()
e2 = mp.Event()
prev_cpu = mp.Value('d',0.0)
prev_mem = mp.Value('d',0.0)
prev_swap = mp.Value('d',0.0)
ctr = mp.Value('d',0.0)
state = mp.Value('d',0.0)
p1 = mp.Process(target=log, args=(e,e2,prev_cpu,prev_mem,prev_swap,ctr,state))
p1.start()
start = time.time()
trainData,trainLabel,valData,valLabel,testData,testLabel = load()
#print testData.shape, testLabel.shape
print time.time()-start
# speedup test
#testData = testData[0:100,:]
#testLabel = testLabel[0:100,:]
#trainData = trainData.reshape(-1,32,32,3)
#valData = valData.reshape(-1,32,32,3)
#testData = testData.reshape(-1,32,32,3)
trainData = trainData.astype(np.float)
valData = valData.astype(np.float)
testData = testData.astype(np.float)
'''
trainData *= 6.
valData *= 6.
testData *= 6.
'''
#trainData = trainData-np.mean(trainData,axis=0)
#valData = valData-np.mean(trainData,axis=0)
#testData = testData-np.mean(trainData,axis=0)
print 'max',trainData.max(),'min',trainData.min(), 'mean',trainData.mean()
print 'max',trainLabel.max(),'min',trainLabel.min(), 'mean',trainLabel.mean()
'''
ctr0,ctr1,ctrx= 0,0,0
for i in trainLabel[640]:
if i==0: ctr0+=1
elif i==1: ctr1+=1
else: ctrx+=1
print ctr0,ctr1,ctrx,ctr0+ctr1+ctrx
'''
start = time.time()
# quick_scan_plot()
# quick_scan(trainData[0:960],trainLabel[0:960],
# valData,valLabel,e,lr_range=[-2,-3],reg_range=[-2,-3],epoch=10,sample=1)
# marathon(trainData,trainLabel,valData,valLabel,testData,testLabel,e,
# lr=0.00138100536852,reg=0.00140667975999, epoch=10)
# review(testData,testLabel,valData,valLabel,e)
# duel(testData,testLabel,valData,valLabel,e)
# mask_input(testData,testLabel,valData,valLabel,e)
e2.set()
print time.time()-start
pass