(0,2)---m*n*2---(1,0)(0,1)
f2[0] | f2[1] | 迭代次数n | 平均准确率p-ave | 1-0 | 0-1 | δ | 耗时ms/次 | 耗时ms/199次 | 耗时 min/199 | 最大准确率p-max | 迭代次数标准差 | pave标准差 |
8.31E-06 | 0.999992 | 8358.603 | 0.981403 | 0.989063 | 0.974129 | 1.00E-05 | 274.9598 | 54730 | 0.912167 | 0.985586 | 1544.237 | 0.001542 |
分类mnist的0和2,当收敛误差是1E-5时,收敛199次统计平均值,(1,0)位的分类准确率是0.989,(0,1)位的分类准确率是0.974。就意味着有1.0937%的0被错误的分成了2;而有2.5871%的2被错误的分成了0.错误图片的净流向是从 2到0,数值是1.493%。这次将所有的错误图片的净流向数据都统计出来。
1-0 | 0-1 | |||||
0 | 8 | 0.995708 | 0.974199 | 0.004292 | 0.025801 | -0.02151 |
0 | 2 | 0.989063 | 0.974129 | 0.010937 | 0.025871 | -0.01493 |
0 | 6 | 0.991873 | 0.977408 | 0.008127 | 0.022592 | -0.01446 |
0 | 7 | 0.995795 | 0.990223 | 0.004205 | 0.009777 | -0.00557 |
0 | 1 | 0.997959 | 1 | 0.002041 | 0 | 0.002041 |
0 | 4 | 0.987514 | 0.997871 | 0.012486 | 0.002129 | 0.010357 |
0 | 5 | 0.972675 | 0.987674 | 0.027325 | 0.012326 | 0.014999 |
0 | 3 | 0.978356 | 0.994711 | 0.021644 | 0.005289 | 0.016355 |
0 | 9 | 0.976669 | 0.99357 | 0.023331 | 0.00643 | 0.016901 |
1 | 7 | 0.999048 | 0.956397 | 0.000952 | 0.043603 | -0.04265 |
1 | 2 | 0.995112 | 0.952553 | 0.004888 | 0.047447 | -0.04256 |
1 | 4 | 0.998645 | 0.984229 | 0.001355 | 0.015771 | -0.01442 |
1 | 9 | 0.989994 | 0.984815 | 0.010006 | 0.015185 | -0.00518 |
1 | 0 | 1 | 0.997959 | 0 | 0.002041 | -0.00204 |
1 | 6 | 0.994249 | 0.992788 | 0.005751 | 0.007212 | -0.00146 |
1 | 8 | 0.982737 | 0.986741 | 0.017263 | 0.013259 | 0.004003 |
1 | 5 | 0.9816 | 0.991995 | 0.0184 | 0.008005 | 0.010395 |
1 | 3 | 0.981511 | 0.992721 | 0.018489 | 0.007279 | 0.01121 |
2 | 6 | 0.96403 | 0.967793 | 0.03597 | 0.032207 | 0.003763 |
2 | 7 | 0.96627 | 0.974073 | 0.03373 | 0.025927 | 0.007802 |
2 | 0 | 0.974129 | 0.989063 | 0.025871 | 0.010937 | 0.014933 |
2 | 8 | 0.950202 | 0.967208 | 0.049798 | 0.032792 | 0.017006 |
2 | 9 | 0.961479 | 0.983505 | 0.038521 | 0.016495 | 0.022026 |
2 | 4 | 0.952364 | 0.989571 | 0.047636 | 0.010429 | 0.037208 |
2 | 3 | 0.942138 | 0.982745 | 0.057862 | 0.017255 | 0.040607 |
2 | 1 | 0.952553 | 0.995112 | 0.047447 | 0.004888 | 0.042559 |
2 | 5 | 0.939825 | 0.984192 | 0.060175 | 0.015808 | 0.044367 |
3 | 8 | 0.976481 | 0.90831 | 0.023519 | 0.09169 | -0.06817 |
3 | 2 | 0.982745 | 0.942138 | 0.017255 | 0.057862 | -0.04061 |
3 | 6 | 0.998358 | 0.965952 | 0.001642 | 0.034048 | -0.03241 |
3 | 0 | 0.994711 | 0.978356 | 0.005289 | 0.021644 | -0.01636 |
3 | 1 | 0.992721 | 0.981511 | 0.007279 | 0.018489 | -0.01121 |
3 | 7 | 0.976556 | 0.979758 | 0.023444 | 0.020242 | 0.003202 |
3 | 9 | 0.967461 | 0.978381 | 0.032539 | 0.021619 | 0.010919 |
3 | 4 | 0.982646 | 0.99365 | 0.017354 | 0.00635 | 0.011004 |
3 | 5 | 0.916046 | 0.957067 | 0.083954 | 0.042933 | 0.041021 |
4 | 2 | 0.989571 | 0.952364 | 0.010429 | 0.047636 | -0.03721 |
4 | 6 | 0.990461 | 0.970022 | 0.009539 | 0.029978 | -0.02044 |
4 | 8 | 0.985774 | 0.974524 | 0.014226 | 0.025476 | -0.01125 |
4 | 3 | 0.99365 | 0.982646 | 0.00635 | 0.017354 | -0.011 |
4 | 0 | 0.997871 | 0.987514 | 0.002129 | 0.012486 | -0.01036 |
4 | 5 | 0.980427 | 0.972734 | 0.019573 | 0.027266 | -0.00769 |
4 | 9 | 0.93322 | 0.9343 | 0.06678 | 0.0657 | 0.00108 |
4 | 1 | 0.984229 | 0.998645 | 0.015771 | 0.001355 | 0.014417 |
4 | 7 | 0.954027 | 0.972054 | 0.045973 | 0.027946 | 0.018027 |
5 | 2 | 0.984192 | 0.939825 | 0.015808 | 0.060175 | -0.04437 |
5 | 3 | 0.957067 | 0.916046 | 0.042933 | 0.083954 | -0.04102 |
5 | 8 | 0.954391 | 0.923024 | 0.045609 | 0.076976 | -0.03137 |
5 | 9 | 0.980902 | 0.95456 | 0.019098 | 0.04544 | -0.02634 |
5 | 6 | 0.974626 | 0.949151 | 0.025374 | 0.050849 | -0.02548 |
5 | 7 | 0.994079 | 0.973823 | 0.005921 | 0.026177 | -0.02026 |
5 | 0 | 0.987674 | 0.972675 | 0.012326 | 0.027325 | -0.015 |
5 | 1 | 0.991995 | 0.9816 | 0.008005 | 0.0184 | -0.0104 |
5 | 4 | 0.972734 | 0.980427 | 0.027266 | 0.019573 | 0.007693 |
6 | 8 | 0.989409 | 0.968766 | 0.010591 | 0.031234 | -0.02064 |
6 | 2 | 0.967793 | 0.96403 | 0.032207 | 0.03597 | -0.00376 |
6 | 7 | 0.994293 | 0.993675 | 0.005707 | 0.006325 | -0.00062 |
6 | 1 | 0.992788 | 0.994249 | 0.007212 | 0.005751 | 0.001461 |
6 | 9 | 0.989661 | 0.997933 | 0.010339 | 0.002067 | 0.008272 |
6 | 0 | 0.977408 | 0.991873 | 0.022592 | 0.008127 | 0.014465 |
6 | 4 | 0.970022 | 0.990461 | 0.029978 | 0.009539 | 0.020439 |
6 | 5 | 0.949151 | 0.974626 | 0.050849 | 0.025374 | 0.025476 |
6 | 3 | 0.965952 | 0.998358 | 0.034048 | 0.001642 | 0.032406 |
7 | 4 | 0.972054 | 0.954027 | 0.027946 | 0.045973 | -0.01803 |
7 | 2 | 0.974073 | 0.96627 | 0.025927 | 0.03373 | -0.0078 |
7 | 8 | 0.975016 | 0.970056 | 0.024984 | 0.029944 | -0.00496 |
7 | 3 | 0.979758 | 0.976556 | 0.020242 | 0.023444 | -0.0032 |
7 | 6 | 0.993675 | 0.994293 | 0.006325 | 0.005707 | 0.000618 |
7 | 0 | 0.990223 | 0.995795 | 0.009777 | 0.004205 | 0.005572 |
7 | 9 | 0.949524 | 0.956129 | 0.050476 | 0.043871 | 0.006605 |
7 | 5 | 0.973823 | 0.994079 | 0.026177 | 0.005921 | 0.020256 |
7 | 1 | 0.956397 | 0.999048 | 0.043603 | 0.000952 | 0.042651 |
8 | 2 | 0.967208 | 0.950202 | 0.032792 | 0.049798 | -0.01701 |
8 | 1 | 0.986741 | 0.982737 | 0.013259 | 0.017263 | -0.004 |
8 | 7 | 0.970056 | 0.975016 | 0.029944 | 0.024984 | 0.004961 |
8 | 4 | 0.974524 | 0.985774 | 0.025476 | 0.014226 | 0.011251 |
8 | 6 | 0.968766 | 0.989409 | 0.031234 | 0.010591 | 0.020644 |
8 | 0 | 0.974199 | 0.995708 | 0.025801 | 0.004292 | 0.02151 |
8 | 5 | 0.923024 | 0.954391 | 0.076976 | 0.045609 | 0.031367 |
8 | 9 | 0.935205 | 0.977902 | 0.064795 | 0.022098 | 0.042698 |
8 | 3 | 0.90831 | 0.976481 | 0.09169 | 0.023519 | 0.068172 |
9 | 8 | 0.977902 | 0.935205 | 0.022098 | 0.064795 | -0.0427 |
9 | 2 | 0.983505 | 0.961479 | 0.016495 | 0.038521 | -0.02203 |
9 | 0 | 0.99357 | 0.976669 | 0.00643 | 0.023331 | -0.0169 |
9 | 3 | 0.978381 | 0.967461 | 0.021619 | 0.032539 | -0.01092 |
9 | 6 | 0.997933 | 0.989661 | 0.002067 | 0.010339 | -0.00827 |
9 | 7 | 0.956129 | 0.949524 | 0.043871 | 0.050476 | -0.0066 |
9 | 4 | 0.9343 | 0.93322 | 0.0657 | 0.06678 | -0.00108 |
9 | 1 | 0.984815 | 0.989994 | 0.015185 | 0.010006 | 0.005179 |
9 | 5 | 0.95456 | 0.980902 | 0.04544 | 0.019098 | 0.026343 |
在所有这些组合中
8 | 3 | 0.90831 | 0.976481 | 0.09169 | 0.023519 | 0.068172 |
6 | 7 | 0.994293 | 0.993675 | 0.005707 | 0.006325 | -0.00062 |
分类8和3的净流向数值最大,而分类6和7的数值最小,二者相差了110倍。在8和3分类的行为中,错误图片的净流向是从8到3.这个意味着我们通常觉得8和3长的很像,但事实上是8长的更像3.
(A,B)---m*n*2---(1,0)(0,1)
作用力和反作用力总是相等的,但A和B之间的相似性确并不一定等于B与A之间的相似性。
(2 , x)---m*n*2---(1,0)(0,1)
当2与其他对象x分类时,错误图像都是由2流向x.如果把错误图像流向比作电流,表明2可以被所有其他图像氧化,是一种强还原剂。
(5 , x)---m*n*2---(1,0)(0,1)
当分类原点是5时,除了4以外流向都是由x到5,表明5可以氧化除了4以外的所有形态,是强氧化剂。
有了氧化性和还原性是否可能由此估计错误图片流向?
观察(8,9),(9,1),( 1,8)这三组分类
8 | 9 | 0.935205 | 0.977902 | 0.064795 | 0.022098 | 0.042698 |
9 | 1 | 0.984815 | 0.989994 | 0.015185 | 0.010006 | 0.005179 |
1 | 8 | 0.982737 | 0.986741 | 0.017263 | 0.013259 | 0.004003 |
流向分别是8→9,9→1,1→8.形成了一个循环,9可以氧化8,1可以氧化9,而8确可以氧化1。这个现象在某种角度上验证了形态数轴的非递进现象,因为形态没有内在的递进规律,因此基于形态的现象可以显得矛盾重重。