-
Notifications
You must be signed in to change notification settings - Fork 0
/
Deducing.py
207 lines (145 loc) · 10.3 KB
/
Deducing.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
import numpy as np
from scipy.special import expit
"""
----------------------------------------------------------------------------------------------------------------------------------------------------------
Notes for readers
Thank you for reading this note. This code is the deducing phase for deep deducing for leveraging sets of weight matrix to perform MWM-
SGD to play Conway's Game of Life.
You may change or tune any of the following parameters or variables. However, it is recommended that you do so only if the following
note suggests so.
We hope you enjoy it.
----------------------------------------------------------------------------------------------------------------------------------------------------------
"""
"""
----------------------------------------------------------------------------------------------------------------------------------------------------------
Part A. Functions for Generating Pre-activated Inner Values of the Missing Input Data for Each Cell, etc.
----------------------------------------------------------------------------------------------------------------------------------------------------------
"""
def generate_inner(inner, initial_state):
missing_input_inner = list()
back_ground = np.zeros((initial_state.shape[0] + 2, initial_state.shape[1] + 2))
back_ground[1 : 1 + initial_state.shape[0], 1 : 1 + initial_state.shape[1]] = initial_state[0 : initial_state.shape[0], 0 : initial_state.shape[1]]
for i in range(initial_state.shape[0]):
for j in range(initial_state.shape[1]):
sequence = (np.random.random(9) - 0.5) * 0.1 - inner * ( (back_ground[1+i-1 : 1+i+2, 1+j-1 : 1+j+2].flatten()) * 2 - 1)
sequence[4] = (np.random.random(1) - 0.5) * 0.1 + 0
missing_input_inner.append( sequence )
missing_input_inner = np.array(missing_input_inner)
return missing_input_inner
"""
----------------------------------------------------------------------------------------------------------------------------------------------------------
Part B. Starting Generations
----------------------------------------------------------------------------------------------------------------------------------------------------------
"""
initial_state = np.array\
([
# Referring to the initial state of the Game of Life table. We highly recommend readers to play with this initial state to see what will happen.
[0, 1, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 1, 0, 0, 0, 0, 0, 0, 0],
[1, 1, 1, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
[0, 0, 0, 0, 0, 0, 0, 0, 0, 0],
])
# Referring to times of test. we recommend readers to change this parameter.
generation = 2000
for i in range(generation):
"""
-----------------------------------------------------------------------------------------------------------------------------------------------------
Part B-1. Loading the Trained Sets of Weight Matrix for MWM-SGD
-----------------------------------------------------------------------------------------------------------------------------------------------------
"""
# we recommend readers to try different size of sets of weight matrix by deleting some of the weight or slope list to grasp a feeling how
# size of sets of weight matrix affect overall accuracy and variability.
weight_lists = list()
slope_lists = list()
weight_list = np.load("self.Conway_1_100x100x100_30_0.000001_50m_[1]_weight_list.npy" , allow_pickle=True)
slope_list = np.load("self.Conway_1_100x100x100_30_0.000001_50m_[1]_slope_list.npy" , allow_pickle=True)
weight_lists.append(weight_list)
slope_lists.append(slope_list)
weight_list = np.load("self.Conway_1_100x100x100_30_0.000001_50m_[2]_weight_list.npy" , allow_pickle=True)
slope_list = np.load("self.Conway_1_100x100x100_30_0.000001_50m_[2]_slope_list.npy" , allow_pickle=True)
weight_lists.append(weight_list)
slope_lists.append(slope_list)
weight_list = np.load("self.Conway_1_100x100x100_30_0.000001_50m_[3]_weight_list.npy" , allow_pickle=True)
slope_list = np.load("self.Conway_1_100x100x100_30_0.000001_50m_[3]_slope_list.npy" , allow_pickle=True)
weight_lists.append(weight_list)
slope_lists.append(slope_list)
weight_list = np.load("self.Conway_1_100x100x100_30_0.000001_50m_[4]_weight_list.npy" , allow_pickle=True)
slope_list = np.load("self.Conway_1_100x100x100_30_0.000001_50m_[4]_slope_list.npy" , allow_pickle=True)
weight_lists.append(weight_list)
slope_lists.append(slope_list)
"""
-----------------------------------------------------------------------------------------------------------------------------------------------------
Part B-2. Initializing Pre-activated Inner Values of the Missing Input Data for Each Cell, etc.
-----------------------------------------------------------------------------------------------------------------------------------------------------
"""
inner = -3.50
missing_input_inner = generate_inner(inner, initial_state)
missing_input_resistor = np.ones_like(missing_input_inner) * np.array([0, 0, 0, 0, 1, 0, 0, 0, 0])
desired_output = np.ones((missing_input_inner.shape[0], missing_input_inner.shape[1]))
"""
-----------------------------------------------------------------------------------------------------------------------------------------------------
Part B-3. Importing Model
-----------------------------------------------------------------------------------------------------------------------------------------------------
"""
from Brain_for_deducing import *
network_size = np.array([9, 100, 100, 100, 9])
# Referring to deducing rate. we recommend readers to change this parameter.
beta = 0.1
# Referring to deducing epochs "between each mandatory pulse". we recommend readers to change this parameter.
epoch_of_deducing = 10000
# Referring to the rate of neurons dropped out in the hidden layers. For example, if drop_rate = 0.2, it means 20% of the neurons in the hidden layers will be dropped out on a random base. we recommend readers to change this parameter.
drop_rate = 0.1
Machine = Brain(network_size, beta, epoch_of_deducing, drop_rate)
"""
-----------------------------------------------------------------------------------------------------------------------------------------------------
Part B-4-1. Back-propagation and WMW-SGD
-----------------------------------------------------------------------------------------------------------------------------------------------------
"""
for j in range(Machine.epoch_of_deducing ):
random_index = np.random.randint(np.array(weight_lists).shape[0])
weight_list = weight_lists[random_index]
slope_list = slope_lists[random_index]
missing_input_inner = Machine.deduce_batch(missing_input_inner,
missing_input_resistor,
desired_output,
weight_list,
slope_list
)
"""
-----------------------------------------------------------------------------------------------------------------------------------------------------
Part B-4-2. Mandatory Pulse
-----------------------------------------------------------------------------------------------------------------------------------------------------
"""
sequence = missing_input_inner[:, 4].flatten()
new_state = np.zeros_like(sequence)
for j in range(sequence.shape[0]):
if sequence[j] >= 1.0:
new_state[j] = 1
elif sequence[j] <= -1.0:
new_state[j] = 0
else:
new_state[j] = initial_state.flatten()[j]
new_state = new_state.reshape((initial_state.shape[0], initial_state.shape[1]))
new_state = np.array(new_state, dtype=int)
"""
-----------------------------------------------------------------------------------------------------------------------------------------------------
Part B-5. Printing out the Next Generation of the Table Proposed by the Machine, etc.
-----------------------------------------------------------------------------------------------------------------------------------------------------
"""
print("Generation -------------------------------------------------------------")
print(i)
print("------------------------------------------------------------------------")
print(new_state)
if (1 in new_state[:, -1]) | (1 in new_state[-1, :]):
new_state = np.vstack(( new_state, np.zeros((5, new_state.shape[1])) ))
new_state = np.hstack(( new_state, np.zeros((new_state.shape[0], 5)) ))
new_state = new_state[4:-1, 4:-1]
initial_state = new_state
else:
initial_state = new_state
#initial_state = new_state