-
Notifications
You must be signed in to change notification settings - Fork 0
/
Final_MLP.py
208 lines (167 loc) · 9.16 KB
/
Final_MLP.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
import numpy as np
import gzip
from sklearn.metrics import confusion_matrix
import matplotlib.pyplot as plt
global train_images, train_labels, train_labels_final, weights2, deltaO, deltaH, \
weights3,finding_accuracy_train,finding_accuracy_test,forward_phase_input_to_hidden, \
output_weight_update , input_weight_change, accuracy_train, accuracy_test,\
train_images_quarter, train_images_half,train_labels_quarter, train_labels_half, \
train_labels_quarter_final, train_labels_half_final
main_size = 60000.00
main_size_quarter = 15000.00
main_size_half = 10000.00
Learning_rate = 0.1
Momentum = 0.5
Hidden_layer = 100
forward_phase_input_to_hidden = np.zeros(Hidden_layer+1)
forward_phase_input_to_hidden[0] = 1
output_weight_update = np.zeros((Hidden_layer+1,10))
input_weight_change = np.zeros((785,Hidden_layer))
forward_phase_input_to_hidden_test = np.zeros(Hidden_layer+1)
forward_phase_input_to_hidden_test[0] = 1
accuracy_train = np.zeros((50))
accuracy_test = np.zeros((50))
class Perceptron:
def __init__(self):
f = gzip.open("train-images.gz", 'rb')
self.train_images = np.frombuffer(f.read(), np.uint8, offset=16)
self.train_images = self.train_images.reshape(-1, 1, 28, 28)
self.train_images = self.train_images.reshape(self.train_images.shape[0],784)
self.train_images = self.train_images / np.float32(255)
self.train_images = np.append(self.train_images,np.ones((60000,1)),axis = 1)
#for experiment 3
#self.train_images_quarter = self.train_images[:15000,:]
self.train_images_half = self.train_images[:10000,:]
f = gzip.open("train-labels.gz", 'rb')
self.train_labels = np.frombuffer(f.read(), np.uint8, offset=8)
self.train_labels = self.train_labels.reshape(60000,1)
#for experiment 3
#self.train_labels_quarter = self.train_labels[:15000,:]
self.train_labels_half = self.train_labels[:10000,:]
#self.train_labels_final = np.full((60000, 10), 0.1, dtype=float)
#self.train_labels_quarter_final = np.full((15000, 10), 0.1, dtype=float)
self.train_labels_half_final = np.full((10000, 10), 0.1, dtype=float)
for row in range(len(self.train_labels_half_final)):
index = self.train_labels_half[row].astype(int)
self.train_labels_half_final[row][index[0]] += 0.8
#self.train_labels_quarter_final = self.train_labels_final[:15000,:]
#self.train_labels_quarter_final = self.train_labels_final[:10000,:]
self.weights2 = np.random.uniform(-0.05, 0.05, size=(785,Hidden_layer))
self.deltaO = np.zeros((Hidden_layer+1,10))
self.deltaH = np.zeros((Hidden_layer,785))
self.weights3 = np.random.uniform(-0.05, 0.05, size=(Hidden_layer+1,10))
accuracy = 0
self.finding_accuracy_train = np.zeros((10000,1))
#exeriment 3
#self.finding_accuracy_train = np.zeros((15000,1))
#self.finding_accuracy_train = np.zeros((10000,1))
self.finding_accuracy_test = np.zeros((10000,1))
f = gzip.open("test-images.gz", 'rb')
self.test_images = np.frombuffer(f.read(), np.uint8, offset=16)
self.test_images = self.test_images.reshape(-1, 1, 28, 28)
self.test_images = self.test_images.reshape(self.test_images.shape[0],784)
self.test_images = np.append(self.test_images,np.ones((10000,1)),axis = 1)
f = gzip.open("test-labels.gz", 'rb')
self.test_labels = np.frombuffer(f.read(), np.uint8, offset=8)
self.test_labels = self.test_labels.reshape(10000,1)
self.test_labels_final = np.full((10000, 10), 0.1, dtype=float)
for row in range(len(self.test_labels)):
index = self.test_labels[row].astype(int)
self.test_labels_final[row][index[0]] += 0.8
def sigmoid(self,input):
return (1/ (1 + np.exp(-input)))
def matrix_mult_input_to_hidden(self,row):
#return np.dot(self.train_images[row][:],self.weights2)
#experiment 3
#return np.dot(self.train_images_quarter[row][:],self.weights2)
return np.dot(self.train_images_half[row][:],self.weights2)
def matrix_mult_input_to_hidden_test(self,row,weights2):
return np.dot(self.test_images[row][:],weights2)
def matrix_mult_hidden_to_output(self,forward_phase_input_to_hidden):
return np.dot(forward_phase_input_to_hidden,self.weights3)
def matrix_mult_hidden_to_output_test(self,forward_phase_input_to_hidden,weights_3):
return np.dot(forward_phase_input_to_hidden,weights_3)
def output_error(self,row,output_sigmoid):
#return output_sigmoid * (1 - output_sigmoid) * (self.train_labels_final[row][:] - output_sigmoid)
#experiment 3
#return output_sigmoid * (1 - output_sigmoid) * (self.train_labels_quarter_final[row][:] - output_sigmoid)
return output_sigmoid * (1 - output_sigmoid) * (self.train_labels_half_final[row][:] - output_sigmoid)
def hidden_error(self,forward_phase_input_to_hidden,output_error,):
x = forward_phase_input_to_hidden[1:] * ( 1 - forward_phase_input_to_hidden[1:] )
y = np.dot(self.weights3[1:,:],np.transpose(output_error))
hidden_error = x * y
return hidden_error
def hidden_to_output_weight_update(self,forward_phase_input_to_hidden,output_error):
step_1 = Learning_rate * np.outer(forward_phase_input_to_hidden,output_error)
step_2 = Momentum * self.deltaO
newdeltaO = step_1 + step_2
self.weights3 += newdeltaO
self.deltaO = newdeltaO
return self.weights3
def input_to_hidden_weight_update(self,row,hidden_error):
#step_1 = Learning_rate * np.outer(hidden_error,self.train_images[row][:])
#experiment 3
#step_1 = Learning_rate * np.outer(hidden_error,self.train_images_quarter[row][:])
step_1 = Learning_rate * np.outer(hidden_error,self.train_images_half[row][:])
step_2 = Momentum * self.deltaH
newdeltaH = step_1 + step_2
self.weights2 += np.transpose(newdeltaH)
self.deltaH = newdeltaH
return self.weights2
def main():
for i in range(50):
global output_weight_update , input_weight_change
class_object = Perceptron()
for row in range(len(class_object.train_images_half)):
input_hidden_mat_mul = class_object.matrix_mult_input_to_hidden(row)
input_hidden_mat_mul_sigmoid = class_object.sigmoid(input_hidden_mat_mul)
forward_phase_input_to_hidden[1:] = input_hidden_mat_mul_sigmoid
hidden_output_mat_mul = class_object.matrix_mult_hidden_to_output(forward_phase_input_to_hidden)
output_sigmoid = class_object.sigmoid(hidden_output_mat_mul)
output_error = class_object.output_error(row,output_sigmoid)
hidden_error = class_object.hidden_error(forward_phase_input_to_hidden,output_error)
# updating weights
output_weight_update = class_object.hidden_to_output_weight_update(forward_phase_input_to_hidden,output_error)
input_weight_change = class_object.input_to_hidden_weight_update(row,hidden_error)
#useful for confusion_matrix
index = np.argmax(output_sigmoid,axis=0)
class_object.finding_accuracy_train[row] = index
print "entering test section :" , i
for row in range(len(class_object.test_images)):
#forward phase
test_input_hidden = class_object.matrix_mult_input_to_hidden_test(row,input_weight_change)
test_input_hidden_sigmoid = class_object.sigmoid(test_input_hidden)
forward_phase_input_to_hidden_test[1:] = test_input_hidden_sigmoid
test_hidden_output_mat_mul = class_object.matrix_mult_hidden_to_output_test(forward_phase_input_to_hidden_test,output_weight_update)
test_output_sigmoid = class_object.sigmoid(test_hidden_output_mat_mul)
index = np.argmax(test_output_sigmoid,axis=0)
class_object.finding_accuracy_test[row] = index
#cfm_train = confusion_matrix(class_object.train_labels,class_object.finding_accuracy_train)
#experiment 3
#cfm_train = confusion_matrix(class_object.train_labels_quarter,class_object.finding_accuracy_train)
cfm_train = confusion_matrix(class_object.train_labels_half,class_object.finding_accuracy_train)
diagonal_sum_train = sum(np.diag(cfm_train))
#accuracy_train[i] = (diagonal_sum_train/main_size)*100
#experiment 3
accuracy_train[i] = (diagonal_sum_train/main_size_half)*100
#accuracy_train[i] = (diagonal_sum_train/main_size_half)*100
#print cfm_train
#print class_object.accuracy_train
cfm_test = confusion_matrix(class_object.test_labels,class_object.finding_accuracy_test)
diagonal_sum_test = sum(np.diag(cfm_test))
accuracy_test[i] = (diagonal_sum_test/10000.00)*100
#print cfm_test
#print class_object.accuracy_test
print cfm_test
print accuracy_test
print cfm_train
print accuracy_train
plt.plot(accuracy_train)
plt.plot(accuracy_test)
plt.ylabel("Accuracy in %")
plt.xlabel("Epoch")
image= "100_half.png"
plt.title("For 100 hidden units _ 0.5")
plt.savefig(image)
plt.show()
main()