隐藏层节点数对搜索区域的影响
直线上有6个格子，向格子里扔3个石子，有4种情况
| 1 |
1 |
1 |
0 |
0 |
0 |
| 0 |
1 |
1 |
1 |
0 |
0 |
| 0 |
0 |
1 |
1 |
1 |
0 |
| 0 |
0 |
0 |
1 |
1 |
1 |
第一种3个石子的间距是0，0，3，有4种可能
| 1 |
1 |
0 |
1 |
0 |
0 |
| 0 |
1 |
1 |
0 |
1 |
0 |
| 0 |
0 |
1 |
1 |
0 |
1 |
第二种3个石子的间距是0，1，2，有3种
| 1 |
0 |
1 |
1 |
0 |
0 |
| 0 |
1 |
0 |
1 |
1 |
0 |
| 0 |
0 |
1 |
0 |
1 |
1 |
第三种，间距是2，1，0，有3种
| 1 |
0 |
1 |
0 |
1 |
0 |
| 0 |
1 |
0 |
1 |
0 |
1 |
第四种间距1，1，1，有2种
| 编号 |
数量 |
| 0 |
4 |
| 2 |
3 |
| 1 |
3 |
| 3 |
2 |
(A，B)---1*n*2---(1,0)(0,1)
现在做一个网络让A就是0，2，1，3，这4个结构，B全是0.改变隐藏层节点的数量n，观察n对迭代次数顺序的影响。
第一次尝试了26个值
| 260 |
240 |
220 |
200 |
190 |
180 |
170 |
150 |
100 |
80 |
| 70 |
60 |
50 |
30 |
20 |
16 |
14 |
11 |
10 |
8 |
| 7 |
6 |
5 |
4 |
3 |
2 |
第二次尝试了32个值
| 260 |
240 |
220 |
200 |
190 |
180 |
170 |
150 |
140 |
130 |
| 120 |
110 |
100 |
90 |
80 |
70 |
60 |
50 |
40 |
30 |
| 20 |
16 |
14 |
11 |
10 |
8 |
7 |
6 |
5 |
4 |
| 3 |
2 |
两组的顺序为
| 260 |
0 |
2 |
1 |
3 |
260 |
0 |
2 |
1 |
3 |
|
| 240 |
0 |
2 |
1 |
3 |
240 |
0 |
2 |
1 |
3 |
|
| 220 |
0 |
1 |
2 |
3 |
220 |
0 |
1 |
2 |
3 |
|
| 200 |
0 |
1 |
2 |
3 |
200 |
0 |
1 |
2 |
3 |
|
| 190 |
0 |
1 |
2 |
3 |
190 |
0 |
1 |
2 |
3 |
|
| 180 |
0 |
2 |
1 |
3 |
180 |
0 |
2 |
1 |
3 |
|
| 170 |
0 |
2 |
1 |
3 |
170 |
0 |
2 |
1 |
3 |
|
| 150 |
0 |
2 |
1 |
3 |
150 |
0 |
2 |
1 |
3 |
|
|
|
140 |
0 |
2 |
1 |
3 |
|||||
|
|
130 |
0 |
2 |
1 |
3 |
|||||
|
|
120 |
0 |
2 |
1 |
3 |
|||||
|
|
110 |
0 |
2 |
1 |
3 |
|||||
| 100 |
0 |
2 |
1 |
3 |
100 |
0 |
2 |
1 |
3 |
|
|
|
90 |
0 |
2 |
1 |
3 |
|||||
| 80 |
0 |
2 |
1 |
3 |
80 |
0 |
2 |
1 |
3 |
|
| 70 |
0 |
2 |
1 |
3 |
70 |
0 |
2 |
1 |
3 |
|
| 60 |
0 |
2 |
1 |
3 |
60 |
0 |
2 |
1 |
3 |
|
| 50 |
0 |
2 |
1 |
3 |
50 |
0 |
2 |
1 |
3 |
|
|
|
40 |
0 |
2 |
1 |
3 |
|||||
| 30 |
0 |
2 |
1 |
3 |
30 |
0 |
2 |
1 |
3 |
|
| 20 |
0 |
2 |
1 |
3 |
20 |
0 |
2 |
1 |
3 |
|
| 16 |
0 |
2 |
1 |
3 |
16 |
0 |
2 |
1 |
3 |
|
| 14 |
0 |
2 |
1 |
3 |
14 |
0 |
2 |
1 |
3 |
|
| 11 |
0 |
2 |
1 |
3 |
11 |
0 |
2 |
1 |
3 |
|
| 10 |
0 |
2 |
1 |
3 |
10 |
0 |
2 |
1 |
3 |
|
| 8 |
0 |
2 |
1 |
3 |
8 |
0 |
2 |
1 |
3 |
|
| 7 |
2 |
0 |
1 |
3 |
7 |
0 |
2 |
1 |
3 |
|
| 6 |
0 |
1 |
2 |
3 |
6 |
0 |
2 |
3 |
1 |
|
| 5 |
0 |
3 |
2 |
1 |
5 |
0 |
2 |
3 |
1 |
|
| 4 |
1 |
0 |
2 |
3 |
4 |
2 |
0 |
1 |
3 |
|
| 3 |
0 |
1 |
2 |
3 |
3 |
0 |
2 |
1 |
3 |
|
| 2 |
0 |
2 |
1 |
3 |
2 |
0 |
2 |
1 |
3 |
|
这数据可以分成3部分，
0-7：混乱区，因为两组顺序不一致
8-180：稳定区，两组顺序是一致，前后一致，和计算顺序也一致
190-240过度拉伸，两组顺序是一致的，但同组数据前后不一致。
比较两次的迭代次数
| n |
1 | 2 | 3 | 4 | 比值3/2 |
n |
1 |
2 |
3 |
4 |
’3/2 |
|||
| 2 |
231947 |
232539 |
233137 |
233785 |
1.003 |
2 |
231904 |
232544 |
233181 |
233760 |
1.0027 |
|||
| 3 |
175900 |
177918 |
178332 |
178889 |
1.002 |
3 |
176177 |
177618 |
178253 |
179648 |
1.0036 |
|||
| 4 |
132737 |
133352 |
133727 |
135830 |
1.003 |
4 |
131288 |
132569 |
133508 |
134739 |
1.0071 |
|||
| 5 |
107257 |
109164 |
109971 |
110483 |
1.007 |
5 |
108173 |
109066 |
109224 |
109327 |
1.0015 |
|||
| 6 |
93075.9 |
93457.8 |
93554.2 |
95622.1 |
1.001 |
6 |
93088 |
93585.2 |
94526.7 |
94656.5 |
1.0101 |
|||
| 7 |
82933.7 |
83513.9 |
84376.5 |
85873.9 |
1.01 |
7 |
82341 |
83141.6 |
85240 |
85270 |
1.0252 |
|||
| 8 |
75463.8 |
75814.7 |
76304.2 |
77469.9 |
1.006 |
8 |
75149 |
76202.4 |
77143 |
77518 |
1.0123 |
|||
| 10 |
64895.5 |
66143.3 |
66312.2 |
67223.3 |
1.003 |
10 |
64730 |
65703.4 |
66831.8 |
67700.1 |
1.0172 |
|||
| 11 |
61036.7 |
62047.4 |
63008.1 |
63364.5 |
1.015 |
11 |
61387 |
62373 |
62637 |
63571.1 |
1.0042 |
|||
| 14 |
53313.2 |
54007 |
54984.9 |
56009.8 |
1.018 |
14 |
53115 |
54370.1 |
55168.4 |
56115.4 |
1.0147 |
|||
| 16 |
49719.5 |
50472.6 |
51743 |
52745.7 |
1.025 |
16 |
49289 |
50537.1 |
51463.8 |
52042.1 |
1.0183 |
|||
| 20 |
44376.2 |
45244.3 |
46336.6 |
47804.1 |
1.024 |
20 |
44100 |
45379.8 |
46256.1 |
47760.6 |
1.0193 |
|||
| 30 |
36446.1 |
38139.8 |
39136.3 |
40975.6 |
1.026 |
30 |
36422 |
38293.3 |
39246.1 |
41183.4 |
1.0249 |
|||
| 40 |
31686 |
33986.3 |
35099 |
37784.7 |
1.0327 |
|||||||||
| 50 |
28500.4 |
31270.1 |
32679.8 |
35720 |
1.045 |
50 |
28497 |
31239.3 |
32560.9 |
35716.3 |
1.0423 |
|||
| 60 |
25919.9 |
29208 |
30553.3 |
34315.7 |
1.046 |
60 |
25871 |
29163.3 |
30508.8 |
34254.4 |
1.0461 |
|||
| 70 |
23640.8 |
27485.8 |
28846.5 |
33358.9 |
1.05 |
70 |
23648 |
27517.7 |
28810.2 |
33412.9 |
1.047 |
|||
| 80 |
21776.1 |
26089.3 |
27284.2 |
32635.8 |
1.046 |
80 |
21762 |
26097.2 |
27323.5 |
32631 |
1.047 |
|||
| 90 |
20068 |
24960.9 |
25955.3 |
32107.8 |
1.0398 |
|||||||||
| 100 |
18556.9 |
23829.8 |
24765.2 |
31722.1 |
1.039 |
100 |
18546 |
23831.3 |
24759.1 |
31764.4 |
1.0389 |
|||
| 110 |
17169 |
22894.8 |
23669.3 |
31427.1 |
1.0338 |
|||||||||
| 120 |
15952 |
22077.5 |
22706 |
31209.5 |
1.0285 |
|||||||||
| 130 |
14861 |
21288.8 |
21829.7 |
31010.7 |
1.0254 |
|||||||||
| 140 |
13891 |
20581.1 |
21019.1 |
30921.3 |
1.0213 |
|||||||||
| 150 |
13035.4 |
19993.1 |
20349.9 |
30854.7 |
1.018 |
150 |
13036 |
19990 |
20332.4 |
30898.3 |
1.0171 |
|||
| 170 |
11576.4 |
18895.8 |
19264.7 |
30876 |
1.02 |
170 |
11598 |
18899 |
19271.2 |
30882.2 |
1.0197 |
|||
| 180 |
10968.8 |
18433.1 |
18772.1 |
30940.4 |
1.018 |
180 |
10963 |
18430.8 |
18767.1 |
30947.3 |
1.0182 |
|||
| 190 |
10410.5 |
17624.2 |
18014 |
31035.7 |
1.022 |
190 |
10407 |
17617 |
18019.4 |
31024.6 |
1.0228 |
|||
| 200 |
9881.49 |
16461.4 |
17351.6 |
31183.6 |
1.054 |
200 |
9881.9 |
16460.8 |
17348.3 |
31158.7 |
1.0539 |
|||
| 220 |
8901.95 |
14111.4 |
14404.6 |
31529.9 |
1.021 |
220 |
8905.7 |
14108.3 |
14416.4 |
31528.4 |
1.0218 |
|||
| 240 |
7953.59 |
11797.7 |
11832.9 |
32023.5 |
1.003 |
240 |
7956.7 |
11797.2 |
11840.7 |
32034.4 |
1.0037 |
|||
| 260 |
6971.04 |
9575.6 |
9717.78 |
32675.7 |
1.015 |
260 |
6968.5 |
9579.64 |
9715.19 |
32658.2 |
1.0141 |
随着n的增加迭代次数是减小的，
特别比较中间两组数据的比值变化，
Ⅰ混乱区是从2-11，Ⅱ稳定区11-150，Ⅲ过度拉伸区>150.
产生混乱区的原因，可能是因为隐藏层数量太少搜索区域尚未能稳定形成。
| 1 |
1 |
0 |
1 |
0 |
0 |
| 1 |
0 |
1 |
1 |
0 |
0 |
按照计算中间的两组出现的概率是相同的，无论谁在前都可以。网络在11-150的区间给出了一致的判断。但当n=190-220的时候，网络的判断前后不一致，所以这组数据暗示了，至少从这4个结构出现的顺序来比较，n应该设置在Ⅱ区，因为这一区域顺序稳定，前后一致。
这也暗示了增加隐藏层节点数看起来像是增加了搜索区域的分辨率，但过多也会有问题。