質問編集履歴
2
コードを追加
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -39,3 +39,59 @@
|
|
|
39
39
|
|
|
40
40
|
|
|
41
41
|
np.tensordot で対応できると良いのですが...
|
|
42
|
+
|
|
43
|
+
|
|
44
|
+
|
|
45
|
+
### 実装
|
|
46
|
+
|
|
47
|
+
たとえば，n=3 だあれば，こんな感じかと思います．
|
|
48
|
+
|
|
49
|
+
絶対にもっと良い実装方法があるとは思うのですが...
|
|
50
|
+
|
|
51
|
+
|
|
52
|
+
|
|
53
|
+
```Python
|
|
54
|
+
|
|
55
|
+
l1 = 3
|
|
56
|
+
|
|
57
|
+
l2 = 3
|
|
58
|
+
|
|
59
|
+
l3 = 3
|
|
60
|
+
|
|
61
|
+
l4 = 3
|
|
62
|
+
|
|
63
|
+
J = 2
|
|
64
|
+
|
|
65
|
+
X = np.random.uniform(0,1,(l1,l2,l3,l4))
|
|
66
|
+
|
|
67
|
+
M = np.random.uniform(0,1,(J,l3))
|
|
68
|
+
|
|
69
|
+
|
|
70
|
+
|
|
71
|
+
def mode_3_prodcut(X,M,mode=3):
|
|
72
|
+
|
|
73
|
+
A = np.zeros((l1,l2,J,l4))
|
|
74
|
+
|
|
75
|
+
for i1 in range(l1):
|
|
76
|
+
|
|
77
|
+
for i2 in range(l2):
|
|
78
|
+
|
|
79
|
+
for j in range(J):
|
|
80
|
+
|
|
81
|
+
for i4 in range(l4):
|
|
82
|
+
|
|
83
|
+
term = 0
|
|
84
|
+
|
|
85
|
+
for i3 in range(l3):
|
|
86
|
+
|
|
87
|
+
term += X[i1,i2,i3,i4] * M[j,i3]
|
|
88
|
+
|
|
89
|
+
|
|
90
|
+
|
|
91
|
+
A[i1,i2,j,i4] = term
|
|
92
|
+
|
|
93
|
+
|
|
94
|
+
|
|
95
|
+
return A
|
|
96
|
+
|
|
97
|
+
```
|
1
参照に追加
test
CHANGED
|
File without changes
|
test
CHANGED
|
@@ -35,3 +35,7 @@
|
|
|
35
35
|
モードn積の定義やRでの実装は以下を参考にしています．
|
|
36
36
|
|
|
37
37
|
https://www.alexejgossmann.com/tensor_decomposition_tucker/
|
|
38
|
+
|
|
39
|
+
|
|
40
|
+
|
|
41
|
+
np.tensordot で対応できると良いのですが...
|