-
Notifications
You must be signed in to change notification settings - Fork 0
/
sin_expect.py
167 lines (125 loc) · 3.97 KB
/
sin_expect.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
# coding: utf-8
# In[1]:
import numpy as np
from matplotlib import pyplot as plt
'''
sin波
'''
f = []
T = 1000
x = np.sin(np.arange(0, 2 * T + 1))
noise = 0.05 * np.random.uniform(low=-1.0, high=1.0, size=len(x))
f = [i + j for (i, j) in zip(x, noise)]
length_of_sequences = 2 * T
maxlen = 25 # ひとつの時系列データの長さ
li = list(f)
group_by = 1
Xl = [li[i:i + group_by] for i in range(0, len(li), group_by)]
Xl = np.array_split(Xl,250)
print(Xl)
input_layer = 1
hidden_layer = 10
output_layer = 1
input_hidden_weight = np.random.randn(hidden_layer,input_layer)
hidden_hidden_weight = np.random.randn(hidden_layer,hidden_layer)
hidden_output_weight = np.random.randn(output_layer,hidden_layer)
#print(str(input_hidden_weight)+"\n"+str(hidden_hidden_weight)+"\n"+str(hidden_output_weight))
# In[2]:
def sigmoid(arr):
return 1.0 / (1.0 + np.exp(-arr))
def tanh(arr):
return np.tanh(arr)
def forward(x,z,p,q):
r = hidden_hidden_weight.dot(z)
z = input_hidden_weight.dot(x) + r
p.append(z)
z = sigmoid(z)
#print(z)
y = hidden_output_weight.dot(z)
q.append(y)
#y = sigmoid(y)
return z, y, p, q
def forward_seq(X):
z = np.zeros(hidden_layer)
zs, ys = ([], [])
p,q = ([], [])
for x in X:
z,y,p,q = forward(x,z,p,q)
zs.append(z)
ys.append(y)
return zs, ys, p, q
# In[3]:
epsilon = 0.05
epoch = 100
loss = np.zeros(epoch)
f2 = open("loss.csv","w")
### train start ###
for epo in range(epoch):
for X in np.random.permutation(Xl):
zs,ys,p,q = forward_seq(X)
#print("zs = \n" + str(zs) + "\n")
#print("ys = \n" + str(ys) + "\n")
hidden_delta = np.zeros(hidden_layer)
hidden_dEdw = np.zeros(input_hidden_weight.shape)
recurr_dEdw = np.zeros(hidden_hidden_weight.shape)
output_dEdw = np.zeros(hidden_output_weight.shape)
tau = X.shape[0]
for t in range(tau - 1)[::-1]:
#fx = sigmoid(q[t])
#print("t = " + str(t))
#output_delta = (1 / (np.cosh(q[t]) ** 2)) * (ys[t] - X[t+1, :])
output_delta = (ys[t] - X[t+1,:])
output_dEdw += output_delta.reshape(-1, 1) * zs[t]
#print(output_dEdw)
#print()
fx = sigmoid(p[t])
hidden_delta = (1 - fx) * fx * ((output_delta.dot(hidden_output_weight)) + hidden_delta.dot(hidden_hidden_weight))
#hidden_delta = (1 / (np.cosh(p[t]) ** 2)) * ((output_delta.dot(hidden_output_weight)) + hidden_delta.dot(hidden_hidden_weight))
hidden_dEdw += hidden_delta.reshape(-1, 1) * X[t,:]
#print(hidden_dEdw)
#print()
zs_prev = zs[t - 1] if t > 0 else np.zeros(hidden_layer)
recurr_dEdw += hidden_delta.reshape(-1, 1) * zs_prev
#print(recurr_dEdw)
#print()
loss[epo] += 0.5 * (ys[t] - X[t + 1]).dot((ys[t] - X[t + 1]).reshape((-1, 1))) / (tau - 1)
#print(loss)
#print()
hidden_output_weight -= epsilon * output_dEdw
input_hidden_weight -= epsilon * hidden_dEdw
hidden_hidden_weight -= epsilon * recurr_dEdw
f2.write(str(loss[epo]) + "\n")
print("epoch: {" + str(epo) + "}:\t" + str(loss[epo]))
f2.close()
# In[4]:
plt.plot(np.arange(loss.size), loss)
plt.ylabel("E")
plt.xlabel("epoch")
plt.savefig("loss_sin.png")
plt.show()
# In[5]:
zs,ys,ps,qs = forward_seq(Xl[0][:2])
print(Xl[0][:2])
print(ys)
#print("ys = \n" + str(y_after) + "\n\nzs = \n" + str(z_after))
#print()
y, z = ys[-1], zs[-1]
p,q =([],[])
for t in range(100):
z, y, p, q = forward(y, z, p, q)
zs.append(z)
ys.append(y)
print(ys)
# In[6]:
y_input = np.array(ys)[:100]
x = np.sin(np.arange(0, 100))
x = np.array(x)[:]
plt.plot(x, linestyle='dotted',color="Blue")
plt.ylim([-1.5, 1.5])
plt.plot(y_input,linestyle="solid")
plt.plot(y_input[:len(Xl[0][:2])],linestyle="dashed",color="Red")
plt.xlabel("t")
plt.ylabel("sin(t)")
plt.savefig("sin.png")
plt.show()
# In[ ]: