-
Notifications
You must be signed in to change notification settings - Fork 0
/
train.py
166 lines (131 loc) · 4.57 KB
/
train.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
import numpy as np
import scipy.io as sio
import matplotlib.pyplot as plt
from matplotlib import gridspec
import pickle
import time
import random
from CNN_layer import *
from remaining_time import *
#Hyperparameters
Output_number = 10
learningrate = 0.01 #learning rate
Image_size = 28
Image_channel = 1
kernl_shape=5
kernl1_number = 8
kernl2_number =8
batch_sze = 20
epoch_number = 2 # number of iterations
mu = 0.98
#PICKLE_FILE = 'output.pickle'
PICKLE_FILE = 'trained.pickle'
## Data extracting
m =10000
X = extract_data('t10k-images-idx3-ubyte.gz', m, Image_size)
y_dash = extract_labels('t10k-labels-idx1-ubyte.gz', m).reshape(m,1)
image=X[99]
plt.imshow(image.reshape((28,28)))
plt.show()
# print(X.shape)
X-= int(np.mean(X))
X/= int(np.std(X))
print(X)
image=X[99]
plt.imshow(image.reshape((28,28)))
plt.show()
test_data = np.hstack((X,y_dash))
# print(test_data)
# print('loaded00000000000000000000000000000000000000')
m =50000
X = extract_data('train-images-idx3-ubyte.gz', m, Image_size)
y_dash = extract_labels('train-labels-idx1-ubyte.gz', m).reshape(m,1)
# print (np.mean(X), np.std(X))
X-= int(np.mean(X)) #mean is set to 0
X/= int(np.std(X)) #data is set is normalized to 11
print(np.std(X) , np.mean(X))
train_data = np.hstack((X,y_dash))
np.random.shuffle(train_data)
digi_img = train_data.shape[0]
print(digi_img, "yahoo")
## Initializing all the parameters
kernel1 = {}
kernel2 = {}
biasconv1 = {}
biasconv2 = {}
for i in range(0,kernl1_number):
kernel1[i] = init_kernel(kernl_shape, Image_channel, distribution='normal')
biasconv1[i] = 0
# v1[i] = 0
for i in range(0,kernl2_number):
kernel2[i] = init_kernel(kernl_shape, kernl1_number, distribution='normal')
biasconv2[i] = 0
# v2[i] = 0
w1 = Image_size-kernl_shape+1
w2 = w1-kernl_shape+1
theta3 = initialize_theta(Output_number, int((w2/2))*int((w2/2))*kernl2_number)
biasfc = np.zeros((Output_number,1))
cost = []
accuracy = []
# pickle_in = open(PICKLE_FILE, 'rb')
# out = pickle.load(pickle_in)
# [kernel1, kernel2, biasconv1, biasconv2, theta3, biasfc, cost, accuracy] = out
xrange = range
print("Learning Rate:"+str(learningrate)+", Batch Size:"+str(batch_sze))
## Training start here
for epoch in range(0,epoch_number):
np.random.shuffle(train_data)
#xrange gives 20 if number of train images is less than 20 or the number of train images if higher
batches = [train_data[k:k + batch_sze] for k in xrange(0, digi_img, batch_sze)]
x=0
for batch in batches:
#time is started for that batch
stime = time.time()
# learningrate = learningrate/(1+epoch/10.0)
out = Momentum_Grad_des(batch, learningrate, Image_size, Image_channel, mu, kernel1, kernel2, biasconv1, biasconv2, theta3, biasfc, cost, accuracy)
[kernel1, kernel2, biasconv1, biasconv2, theta3, biasfc, cost, accuracy] = out
print(kernel1[0].shape, kernel2[0].shape, "yesss")
epoch_accuracy = round(np.sum(accuracy[int(epoch*digi_img/batch_sze):])/(x+1),2)
per = float(x+1)/len(batches)*100
print("Epoch:"+str(round(per,2))+"% Of "+str(epoch+1)+"/"+str(epoch_number)+", Cost:"+str(cost[-1])+", B.Acc:"+str(accuracy[-1]*100)+", E.Acc:"+str(epoch_accuracy))
ftime = time.time()
deltime = ftime-stime
remtime = (len(batches)-x-1)*deltime+deltime*len(batches)*(epoch_number-epoch-1)
printTime(remtime)
x+=1
## saving the trained model parameters
with open(PICKLE_FILE, 'wb') as file:
pickle.dump(out, file)
## Opening the saved model parameter
pickle_in = open(PICKLE_FILE, 'rb')
out = pickle.load(pickle_in)
[kernel1, kernel2, biasconv1, biasconv2, theta3, biasfc, cost, accuracy] = out
## Plotting the cost and accuracy over different background
gs = gridspec.GridSpec(2, 1, height_ratios=[1, 1])
ax0 = plt.subplot(gs[0])
line0, = ax0.plot(cost, color='b')
ax1 = plt.subplot(gs[1], sharex = ax0)
line1, = ax1.plot(accuracy, color='r', linestyle='--')
plt.setp(ax0.get_xticklabels(), visible=False)
ax0.legend((line0, line1), ('Loss', 'Accuracy'), loc='upper right')
# remove vertical gap between subplots
plt.subplots_adjust(hspace=.0)
plt.show(block=False)
## Computing Test accuracy
X = test_data[:,0:-1]
X = X.reshape(len(test_data), Image_channel, Image_size, Image_size)
y = test_data[:,-1]
print('dhjadgfhdsbfk')
print(y.shape)
corr = 0
print("Computing accuracy over test set:")
for i in range(0,len(test_data)):
image = X[i]
digit, prob = predict(image, kernel1, kernel2, biasconv1, biasconv2, theta3, biasfc)
print (digit, y[i], prob)
if digit==y[i]:
corr+=1
if (i+1)%int(0.01*len(test_data))==0:
print(str(float(i+1)/len(test_data)*100)+"% Completed")
test_accuracy = float(corr)/len(test_data)*100
print("Test Set Accuracy:"+str(test_accuracy))