-
Notifications
You must be signed in to change notification settings - Fork 1
/
HopfieldNetwork.py
134 lines (106 loc) · 3.89 KB
/
HopfieldNetwork.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
import numpy as np
import matplotlib.pyplot as plt
import matplotlib.cm as cm
class HopefieldNetwork(object):
def __init__(self, size):
self.num_neurons = size
self.W = np.ones((self.num_neurons, self.num_neurons), dtype = 'float16')
def resetNetwork(self):
self.W = np.ones((self.num_neurons, self.num_neurons), dtype = 'float16')
def trainHebb(self, data, showWeights = False):
self.type = self.trainHebb
if showWeights:
self.initialize_for_showing_weights()
X = np.array(data).T
self.W = (np.matmul(X,X.T) - len(data)*np.eye(self.num_neurons, dtype='float16'))/self.num_neurons
if showWeights:
self.show_weights(f'Final', block=True)
self.save_weights()
def trainOja(self, data, u = 0.0001, iter = 1000, showWeights = False):
self.type = self.trainOja
self.trainHebb(data)
if showWeights:
self.initialize_for_showing_weights()
self.show_weights(f'Iteration#{0}')
for it in range(iter):
Wprev = self.W.copy()
for x in data:
x = (x+1)/2
y = np.matmul(x, self.W)
for i in range(self.num_neurons):
for j in range(self.num_neurons):
self.W[i,j] += u*(y[i]*x[j] - self.W[i,j]*(y[i]**2))
if(showWeights):
self.show_weights(f'Iteration#{it+1}')
print(f'Iteration #{it+1}/{iter}', "\t", np.linalg.norm(Wprev - self.W))
if np.linalg.norm(Wprev - self.W) < 1e-14:
break
self.W -= np.diag(np.diag(self.W))
if showWeights:
self.show_weights(f'Final', block=True)
self.save_weights()
def _async(self, x, W):
xsync, _ = self._sync(x, W)
change = np.multiply(xsync, x)
changed = np.argwhere(change < 0)
if changed.size ==0:
return np.array(x), False
else:
toChange = np.random.choice(changed.reshape(-1))
x[toChange] = xsync[toChange]
return np.array(x), True
def _sync(self, x, W):
return np.sign(np.matmul(W, x)), True
def forward(self, data, iter = 20, asyn = False, print = None):
s = data
if asyn:
f = self._async
else:
f = self._sync
e = []
e.append(0)
e.append(self.energy(s))
for i in range(iter):
if self.type == self.trainOja:
s = (s+1)/2
s, cont = f(s, self.W)
e.append(self.energy(s))
if print != None:
# print(s, f'{i+1}')
print(s, f'{i+1} E: {e[-1]}')
self.plotEnergy(e)
# if not cont:
# return s
# if f == self._sync and e[-1] == e[-2]:
# return s
return s
def energy(self, s):
return -0.5*np.matmul(np.matmul(s, self.W), s)
def plotEnergy(self, energy):
plt.figure(5000)
# print(energy[-3], energy[-2], energy[-1])
plt.ion()
plt.clf()
plt.title('Energy value')
plt.ylabel('Energy')
plt.xlabel('Iteration')
plt.plot(energy)
def initialize_for_showing_weights(self):
plt.ion()
plt.figure(figsize=(6, 5))
plt.tight_layout()
def show_weights(self, iteration_name, block = False):
plt.figure(1)
plt.clf()
plt.title(f'Network Weights - {iteration_name}')
colors = plt.imshow(self.W, cmap=cm.coolwarm)
plt.colorbar(colors)
plt.show(block = block)
plt.pause(0.1)
def save_weights(self):
plt.figure(1)
plt.clf()
plt.title(f'Network Weights - Final')
colors = plt.imshow(self.W, cmap=cm.coolwarm)
plt.colorbar(colors)
plt.savefig('weights.png')