-
Notifications
You must be signed in to change notification settings - Fork 0
/
RBM.py
183 lines (148 loc) · 7.17 KB
/
RBM.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
import numpy as np
import tensorflow as tf
from tensorflow import keras
import tensorflow_probability as tfp
class RBM(keras.Model):
def __init__(self, num_v, num_h, batch_size, learning_rate, num_epoch, k=2):
super(RBM, self).__init__()
self.num_v = num_v
self.num_h = num_h
self.k = k
self.batch_size = batch_size
self.learning_rate = learning_rate
self.num_epoch = num_epoch
self.W = self.add_weight(shape=(num_v, num_h), initializer='random_normal', trainable=True)
self.a = self.add_weight(shape=(num_v,), initializer='zeros', trainable=True)
self.b = self.add_weight(shape=(num_h,), initializer='zeros', trainable=True)
def call(self, v):
# Perform Gibbs sampling and return the final visible layer
_, _, vk, _ = self._gibbs_sampling(v)
return vk
def _gibbs_sampling(self, v):
print("Shape of visible units v:", v.shape)
print("Shape of weight matrix W:", self.W.weights[0].shape)
print("Shape of visible bias a:", self.a.weights[0].shape)
print("Shape of hidden bias b:", self.b.weights[0].shape)
v0 = v
prob_h_v0 = self._prob_h_given_v(v0)
vk = v
prob_h_vk = prob_h_v0
for _ in range(self.k):
hk = self._bernoulli_sampling(prob_h_vk)
prob_v_hk = self._prob_h_given_v(vk)
vk = self._bernoulli_sampling(prob_v_hk)
prob_h_vk = self._prob_h_given_v(vk)
mask = tf.cast(tf.equal(v0, 0), dtype=tf.float32)
vk = mask * v0 + (1 - mask) * vk
prob_h_vk = prob_h_vk * mask + prob_h_v0 * (1 - mask)
print("v0 shape:", v0.shape)
print("prob_h_v0 shape:", prob_h_v0.shape)
print("prob_h_vk shape:", prob_h_vk.shape)
return v0, prob_h_v0, vk, prob_h_vk
def _prob_v_given_h(self, h):
return tf.sigmoid(tf.add(self.a(h), tf.matmul(h, tf.transpose(self.W.weights[0]))))
def _prob_h_given_v(self, v):
return tf.sigmoid(tf.add(self.b(v), tf.matmul(v, self.W.weights[0])))
def _bernoulli_sampling(self, prob):
distribution = tfp.distributions.Bernoulli(probs=prob, dtype=tf.float32)
return tf.cast(distribution.sample(), tf.float32)
def _compute_gradient(self, v0, prob_h_v0, vk, prob_h_vk):
outer_product0 = tf.matmul(tf.transpose(v0), prob_h_v0)
outer_productk = tf.matmul(tf.transpose(vk), prob_h_vk)
W_grad = tf.reduce_mean(outer_product0 - outer_productk, axis=0)
a_grad = tf.reduce_mean(v0 - vk, axis=0)
b_grad = tf.reduce_mean(prob_h_v0 - prob_h_vk, axis=0)
return W_grad, a_grad, b_grad
def _optimize(self, v):
v0, prob_h_v0, vk, prob_h_vk = self._gibbs_sampling(v)
W_grad, a_grad, b_grad = self._compute_gradient(v0, prob_h_v0, vk, prob_h_vk)
self.W.weights[0].assign_add(self.learning_rate * W_grad)
self.a.weights[0].assign_add(self.learning_rate * a_grad)
self.b.weights[0].assign_add(self.learning_rate * b_grad)
error = tf.reduce_mean(tf.square(v0 - vk))
return error
def train(self, X_train):
"""
Model training
@param X_train: training dataset
"""
model = RBM(self.num_v, self.num_h, self.batch_size, self.learning_rate, self.num_epoch, self.k)
model.compile(optimizer=keras.optimizers.Adam(learning_rate=self.learning_rate),
loss=self.compute_loss)
model.fit(X_train, X_train, batch_size=self.batch_size, epochs=self.num_epoch, verbose=1)
# Print training error
train_loss = model.evaluate(X_train, X_train, verbose=0)
print("Training Error: ", train_loss)
def compute_loss(self, v0, vk):
# Compute the loss function (e.g., mean squared error)
return tf.reduce_mean(tf.square(v0 - vk))
def predict(self, v):
prob_h_v = self.call(v)
h = tf.cast(tf.random.uniform(tf.shape(prob_h_v)) < prob_h_v, tf.float32)
prob_v_h = self.call(h)
return prob_v_h
def _gibbs_sampling(self, v):
print("Shape of visible units v:", v.shape)
print("Shape of weight matrix W:", self.W.shape)
print("Shape of visible bias a:", self.a.shape)
print("Shape of hidden bias b:", self.b.shape)
v0 = v
prob_h_v0 = self._prob_h_given_v(v0)
vk = v
prob_h_vk = prob_h_v0
for _ in range(self.k):
hk = self._bernoulli_sampling(prob_h_vk)
prob_v_hk = self._prob_h_given_v(vk)
vk = self._bernoulli_sampling(prob_v_hk)
prob_h_vk = self._prob_h_given_v(vk)
mask = tf.cast(tf.equal(v0,0), dtype=tf.float32)
vk = mask * v0 + (1 - mask) * vk
prob_h_vk = prob_h_vk * mask + prob_h_v0 * (1 - mask)
print("v0 shape:", v0.shape)
print("prob_h_v0 shape:", prob_h_v0.shape)
print("prob_h_vk shape:", prob_h_vk.shape)
return v0, prob_h_v0, vk, prob_h_vk
def _prob_v_given_h(self, h):
return tf.sigmoid(tf.add(self.a, tf.matmul(h, tf.transpose(self.W))))
def _prob_h_given_v(self, v):
return tf.sigmoid(tf.add(self.b, tf.matmul(v, self.W)))
def _bernoulli_sampling(self, prob):
distribution = tfp.distributions.Bernoulli(probs=prob, dtype=tf.float32)
return tf.cast(distribution.sample(), tf.float32)
def _compute_gradient(self, v0, prob_h_v0, vk, prob_h_vk):
outer_product0 = tf.matmul(tf.transpose(v0), prob_h_v0)
outer_productk = tf.matmul(tf.transpose(vk), prob_h_vk)
W_grad = tf.reduce_mean(outer_product0 - outer_productk, axis=0)
a_grad = tf.reduce_mean(v0 - vk, axis=0)
b_grad = tf.reduce_mean(prob_h_v0 - prob_h_vk, axis=0)
return W_grad, a_grad, b_grad
def _optimize(self, v):
v0, prob_h_v0, vk, prob_h_vk = self._gibbs_sampling(v)
W_grad, a_grad, b_grad = self._compute_gradient(v0, prob_h_v0, vk, prob_h_vk)
self.W.assign_add(self.learning_rate * W_grad)
self.a.assign_add(self.learning_rate * a_grad)
self.b.assign_add(self.learning_rate * b_grad)
error = tf.reduce_mean(tf.square(v0 - vk))
return error
def train(self, X_train):
"""
Model training
@param X_train: training dataset
"""
model = keras.Sequential()
model.add(keras.layers.Dense(self.num_h, activation='sigmoid', input_shape=(self.num_v,)))
model.add(keras.layers.Dense(self.num_v, activation='sigmoid'))
model.compile(optimizer=keras.optimizers.Adam(learning_rate=self.learning_rate),
loss=self.compute_loss)
model.fit(X_train, X_train, batch_size=self.batch_size, epochs=self.num_epoch, verbose=1)
# Print training error
train_loss = model.evaluate(X_train, X_train, verbose=0)
print("Training Error: ", train_loss)
def compute_loss(self, v0, vk):
# Compute the loss function (e.g., mean squared error)
return tf.reduce_mean(tf.square(v0 - vk))
def predict(self, v):
prob_h_v = tf.sigmoid(tf.matmul(v, self.W) + self.b)
h = self._bernoulli_sampling(prob_h_v)
prob_v_h = tf.sigmoid(tf.matmul(h, tf.transpose(self.W)) + self.a)
return prob_v_h