一维矩阵也就是一维数组,就是一个数。网络结构是1*2*2,
设r是0-1之间的随机数
让A=sigmoid(r),B=sigmoid(dis+r)
其中0.5<=dis<=3
收敛标准0.5到1e-5,共17个收敛标准,每个收敛标准收敛199次,取平均值统计。
先比较迭代次数
dis | 0.5 | 0.6 | 0.7 | 0.8 | 0.9 | 1 | 1.1 | 1.2 | 1.3 | 1.4 | 1.5 | 1.6 | 1.7 | 1.8 | 1.9 | 2 | 2.1 | 2.2 | 2.3 | 2.4 | 2.5 | 2.6 | 2.7 | 2.8 | 2.9 | 3 |
δ | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n | 迭代次数n |
0.5 | 419.839196 | 417.7487437 | 408.8643216 | 367.5929648 | 371.5376884 | 383.2613065 | 342.5628141 | 339.9849246 | 323.5577889 | 315.0552764 | 318.1758794 | 321.0301508 | 327.201005 | 326.9497487 | 307.2512563 | 311.9145729 | 295.6532663 | 294.718593 | 292.9949749 | 290.9045226 | 298.4120603 | 292.0603015 | 291.6331658 | 299.1306533 | 292.7085427 | 304.3567839 |
0.4 | 16955.19095 | 14775.66332 | 12474.9598 | 11153.58291 | 10406.0603 | 9348.798995 | 8698.643216 | 8266.115578 | 7758.743719 | 7292.015075 | 6952.452261 | 6627.542714 | 6757.045226 | 6415.849246 | 5978.798995 | 6087.984925 | 6005.251256 | 5917.311558 | 5697.40201 | 5540.61809 | 5534.020101 | 5463.281407 | 5372.668342 | 5431.939698 | 5376.778894 | 5310.462312 |
0.3 | 29202.47739 | 24254.14573 | 20976.20603 | 19051.22111 | 16651.19095 | 15631.55276 | 14454.8794 | 13589.45226 | 12593.22111 | 12051.42211 | 11587.0603 | 11067.82412 | 10680.10553 | 10357.55276 | 9982.969849 | 9805.994975 | 9616.135678 | 9529.874372 | 9433.452261 | 9166.477387 | 9072.447236 | 8885.773869 | 8794.959799 | 8639.231156 | 8840.487437 | 8586.366834 |
0.2 | 44825.76382 | 37251.53266 | 32396.08543 | 28633.85427 | 25624.8794 | 23813.93467 | 22113.58291 | 20480.39698 | 19426.60804 | 18268.70854 | 17760.58794 | 16908.8593 | 16387.99497 | 15944.38693 | 15408.49749 | 15312.37688 | 14971.39196 | 14585.1809 | 14421.36181 | 14048.59799 | 13691.82412 | 13827.28141 | 13446.65829 | 13384.05528 | 13303.44221 | 13093.0804 |
0.1 | 72291.31156 | 59483.47236 | 50957.67337 | 45194.40704 | 40832.59799 | 37240.12563 | 34698.70854 | 32627.55276 | 31157.1005 | 29656.14573 | 28511.07035 | 27674.14573 | 26548.24623 | 25682.58794 | 25103.52261 | 24639.55276 | 24022.86935 | 23650.22613 | 23469.58291 | 23347.55276 | 22652.28643 | 22466.11558 | 22176.52764 | 22038.33668 | 21754.77889 | 21807.54271 |
0.01 | 278803.794 | 183967.1608 | 144001.6834 | 122592.6784 | 108123.7538 | 99274.29648 | 92195.31156 | 87206.74874 | 83313.53266 | 80193.37186 | 77681.81407 | 75590.71859 | 74378.20603 | 72915.0804 | 71467.26131 | 71040.70854 | 69932.12563 | 69362.02513 | 68646.94975 | 68408.73869 | 68062.40704 | 67523.96482 | 67531.13065 | 66462.35678 | 66488.79899 | 66376.27638 |
0.001 | 1713934.035 | 674994.9899 | 413344.809 | 305566.0251 | 251532.608 | 219443.5729 | 199507.3216 | 186584.4472 | 177075.603 | 170630.1457 | 165983.995 | 162911.3618 | 161348.9397 | 159932.1558 | 159337.8342 | 158983.4221 | 159077.6332 | 159544.1256 | 159821.0503 | 160663.3819 | 161116.2261 | 161708.4271 | 162089.4221 | 162288.5879 | 162991.3116 | 163313.2915 |
1.00E-04 | 1.19E+07 | 2679698.859 | 1232490.146 | 777732.2965 | 574552.3166 | 470602.1055 | 407878.7387 | 366201.6231 | 335943.9146 | 316365.4824 | 304257.8342 | 297485.4422 | 296012.8894 | 297276.608 | 301680.1256 | 307999.794 | 314286.8995 | 322093.3116 | 329162.1658 | 338214.7487 | 344505.2915 | 352738.196 | 359406.4372 | 365799.4322 | 372157.3618 | 377038.1759 |
9.00E-05 | 1.30E+07 | 2872340.97 | 1295095.935 | 808873.7437 | 597747.0804 | 486176.6482 | 421541.0804 | 376480.799 | 346545.7035 | 324163 | 312488.0854 | 305251.593 | 303056.8492 | 304840.8191 | 309849.9347 | 315659.3417 | 323358.6884 | 332205.2412 | 339669.4724 | 348841.7638 | 357132.0352 | 366845.593 | 372995.1005 | 381125.1106 | 386435.7236 | 391558.0251 |
8.00E-05 | 1.44E+07 | 3074091.231 | 1365654.99 | 850489.3417 | 626130.2864 | 505004.7688 | 435323.1809 | 387880.5578 | 356512.1859 | 333874.7588 | 320704.1658 | 313778.5578 | 312788.1256 | 314933.4523 | 318527.9447 | 325469.3015 | 335544.6281 | 342807.402 | 353732.9598 | 362814.2462 | 371092.809 | 379741.8543 | 387441.1005 | 394750.3467 | 402623.3417 | 408783.2915 |
7.00E-05 | 1.59E+07 | 3335533.362 | 1463761.643 | 892413.8844 | 654262.0553 | 526938.5276 | 453691.6131 | 403053.2412 | 368741.6533 | 345177.5628 | 330839.3417 | 323714.0854 | 323022.4874 | 324382.799 | 329690.1558 | 336914.7789 | 345986.4975 | 356570.8492 | 366661.3719 | 377116.4372 | 386719.6935 | 396212.397 | 406620.4874 | 414203.9146 | 422376.206 | 428869.8543 |
6.00E-05 | 1.84E+07 | 3649046.266 | 1554796.658 | 949248.7286 | 690925.804 | 554851.392 | 474887.3015 | 420141.8442 | 383213.5226 | 358228.8894 | 343465.1005 | 335314.5276 | 332910.7588 | 335194.5176 | 342062.0955 | 351656.4673 | 361890.3166 | 373365.2211 | 384106.608 | 397505.3417 | 406170.3769 | 418025.9447 | 428835.1206 | 436813.4925 | 445218.2965 | 452498.0251 |
5.00E-05 | 2.10E+07 | 4046202.256 | 1709205.111 | 1021742.206 | 735146.8392 | 589089.5528 | 500490.7789 | 441564.9497 | 401032.1759 | 373825.603 | 358363.2513 | 350773.8241 | 348866.4774 | 350744.1357 | 357790.8693 | 368642.7186 | 380881.4523 | 392406.9698 | 404601.402 | 418133.8844 | 434234.1055 | 444025.593 | 455025.8342 | 466890.3367 | 476046.1457 | 486246.2864 |
4.00E-05 | 2.49E+07 | 4633258.296 | 1893380.899 | 1112750.447 | 798799.1407 | 630300.2663 | 534570.5779 | 468051.1809 | 423859.4322 | 393420.7487 | 375380.9095 | 367307.6332 | 365698.4271 | 369310.1658 | 378250.0754 | 390983.4121 | 401978.5075 | 417196.6382 | 434782.206 | 448920.8995 | 464603.0503 | 478823.0503 | 493251.5528 | 504019.6131 | 516600.1256 | 525520.1558 |
3.00E-05 | 3.15E+07 | 5484745.372 | 2156059.492 | 1243715.864 | 877070.608 | 692638.5578 | 580010.0553 | 505632.7688 | 454636.5477 | 421475.7337 | 400995.191 | 391657.6734 | 389171.6935 | 393782.3065 | 405175.9447 | 420001.9548 | 436304.2462 | 454887.4824 | 471590.1759 | 493132.9296 | 510671.7538 | 528130.7789 | 543198.6683 | 559419.4523 | 572758.4673 | 585288.1859 |
2.00E-05 | 4.28E+07 | 6910820.357 | 2589055.925 | 1453151.09 | 1012830.92 | 783340.2864 | 651728.0854 | 561421.2613 | 500413.201 | 460462.5276 | 437078.4372 | 427456.9598 | 424849.2211 | 430710.6181 | 443335.603 | 461438.2161 | 484146.4372 | 510004.4271 | 534865.7236 | 562289.3116 | 582864.0854 | 606075.4523 | 626554.6281 | 646372.9397 | 661013.9246 | 682289.1407 |
1.00E-05 | 7.45E+07 | 1.03E+07 | 3539154.497 | 1904230.698 | 1277830.608 | 974121.0905 | 791173.3317 | 667469.7839 | 587220.6382 | 536781.8844 | 507505.6131 | 494140.7789 | 491504.9598 | 497705.3015 | 516433.191 | 545009.0804 | 580720.4673 | 622064.1156 | 661299.8442 | 699378.7588 | 734980.3065 | 771159.3417 | 802233.7638 | 830242.6281 | 856295.4824 | 882793.7236 |
随着dis的减小,迭代次数增加,
按照假设1
假设1:完全相同的两个对象无法被分成两类,与之对应的分类迭代次数为无穷大
当dis=0时A=B,无法分类。因此随着dis减小A与B之间的差异也减小,A与B越来越难以分类,外在表现就是迭代次数随着dis的减小而增加。
但值得注意的是迭代次数的最低值是在dis=1.7.
当dis>1.7以后,随着dis的增加迭代次数是增加的。这个与假设1不符。这个现象可能和sigmoid的特性有关,因为输入数据越大sigmoid的输出会越接近1,导致数字的区分度下降。
再比较分类准确率
dis | 0.5 | 0.6 | 0.7 | 0.8 | 0.9 | 1 | 1.1 | 1.2 | 1.3 | 1.4 | 1.5 | 1.6 | 1.7 | 1.8 | 1.9 | 2 | 2.1 | 2.2 | 2.3 | 2.4 | 2.5 | 2.6 | 2.7 | 2.8 | 2.9 | 3 |
δ | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave |
0.5 | 0.499398896 | 0.501345567 | 0.503141332 | 0.508010523 | 0.501526652 | 0.502210753 | 0.502854613 | 0.503976338 | 0.500022636 | 0.503981368 | 0.505570898 | 0.501159451 | 0.507834468 | 0.501526652 | 0.505530656 | 0.502155422 | 0.506783165 | 0.508629232 | 0.500706737 | 0.506179546 | 0.510264033 | 0.500596073 | 0.503986398 | 0.502829463 | 0.505404902 | 0.512266035 |
0.4 | 0.75264209 | 0.802732381 | 0.852807582 | 0.892933134 | 0.908270079 | 0.915136242 | 0.927470184 | 0.936016418 | 0.94316427 | 0.951418755 | 0.960166196 | 0.965307016 | 0.97157459 | 0.978078581 | 0.979839136 | 0.985603694 | 0.988098651 | 0.990769664 | 0.993425586 | 0.995402438 | 0.995910483 | 0.997811882 | 0.998163993 | 0.998229385 | 0.998948697 | 0.999190145 |
0.3 | 0.753225588 | 0.802863165 | 0.853496713 | 0.904688608 | 0.951006283 | 0.973304963 | 0.98352121 | 0.990186166 | 0.995347106 | 0.998219325 | 0.999064391 | 0.999728372 | 0.999969819 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
0.2 | 0.751148133 | 0.800579474 | 0.850433348 | 0.895438152 | 0.94122263 | 0.979094673 | 0.997027178 | 0.999939638 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
0.1 | 0.739271936 | 0.777345185 | 0.815579399 | 0.853486653 | 0.888627321 | 0.929985262 | 0.962027354 | 0.991428615 | 0.999818914 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
0.01 | 0.68160623 | 0.703482377 | 0.736168329 | 0.770167152 | 0.799618714 | 0.836645691 | 0.871987566 | 0.910956182 | 0.947424812 | 0.987032258 | 0.999904427 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
0.001 | 0.675348716 | 0.692783236 | 0.718220734 | 0.745831259 | 0.77718422 | 0.807103586 | 0.840081287 | 0.873446311 | 0.90958798 | 0.949346331 | 0.985090618 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
1.00E-04 | 0.677058969 | 0.693718845 | 0.716384727 | 0.747435878 | 0.773527296 | 0.803813864 | 0.831635656 | 0.853622467 | 0.873954356 | 0.896519635 | 0.918390753 | 0.946690409 | 0.975367327 | 0.996046298 | 0.999894367 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
9.00E-05 | 0.677340658 | 0.692788266 | 0.715745897 | 0.746686385 | 0.774840167 | 0.802038219 | 0.829281543 | 0.852988667 | 0.874794393 | 0.892666536 | 0.914970247 | 0.943365476 | 0.970870368 | 0.995608674 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
8.00E-05 | 0.68095734 | 0.692622271 | 0.716515511 | 0.744860439 | 0.774865318 | 0.80367302 | 0.829035065 | 0.852309596 | 0.87181654 | 0.891449238 | 0.911942093 | 0.934487251 | 0.966966967 | 0.994456768 | 0.999803824 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
7.00E-05 | 0.679659559 | 0.692863718 | 0.718406849 | 0.743059643 | 0.775222459 | 0.801786712 | 0.831082339 | 0.852279415 | 0.869638483 | 0.887128334 | 0.908124205 | 0.931373585 | 0.961303012 | 0.991262619 | 0.999597588 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
6.00E-05 | 0.677390959 | 0.693970353 | 0.716440058 | 0.744981162 | 0.775212399 | 0.80336618 | 0.829578322 | 0.849814639 | 0.868275311 | 0.886952279 | 0.904381769 | 0.926725721 | 0.956569635 | 0.987032258 | 0.998405441 | 0.999969819 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
5.00E-05 | 0.682124336 | 0.694478398 | 0.717581904 | 0.746656204 | 0.772903557 | 0.805056313 | 0.830423388 | 0.850659705 | 0.864135492 | 0.880141448 | 0.898441155 | 0.920805227 | 0.949406693 | 0.981312971 | 0.996639856 | 0.999884306 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
4.00E-05 | 0.682637412 | 0.695046806 | 0.715735837 | 0.746454998 | 0.776404545 | 0.803134793 | 0.83143445 | 0.845876027 | 0.861922224 | 0.876172655 | 0.887666561 | 0.910131237 | 0.938813185 | 0.969029331 | 0.99516099 | 0.999959759 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
3.00E-05 | 0.683095156 | 0.693713815 | 0.718195583 | 0.745152188 | 0.776037344 | 0.805830957 | 0.829709106 | 0.845549067 | 0.858552019 | 0.868290401 | 0.880901002 | 0.903592034 | 0.924648266 | 0.955312096 | 0.985840111 | 0.999089542 | 1 | 1 | 1 | 1 | 1 | 0.997394379 | 1 | 1 | 1 | 1 |
2.00E-05 | 0.681872828 | 0.695268133 | 0.716998406 | 0.746173309 | 0.777717416 | 0.806328942 | 0.829216151 | 0.841957536 | 0.850886062 | 0.860689836 | 0.869019774 | 0.887993521 | 0.903682577 | 0.940302111 | 0.968118873 | 0.993948723 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
1.00E-05 | 0.686022706 | 0.69648543 | 0.720947078 | 0.747415758 | 0.777551421 | 0.807596541 | 0.828894221 | 0.837616511 | 0.843692939 | 0.848184868 | 0.85049874 | 0.858023853 | 0.869930232 | 0.885875826 | 0.925261945 | 0.969361321 | 0.993003053 | 0.999552316 | 1 | 1 | 1 | 1 | 1 | 1 | 1 | 1 |
随着dis的减小分类准确率降低,
假设2:对应不同的两个对象,迭代次数越大,二者的相对速度越大;相对速度越大分类准确率越大。
按照假设2:两个对象分类准确率是100%就意味这两个对象以光速相互远离,彼此毫无影响,则dis越大在等收敛标准下两个对象的相对速度越大。按照假设3这是由于dis越小A与B的相似程度愈大,则等效交叉程度越大,导致A与B的惯性质量越大。
假设当收敛标准同样大的时候,施加到两个对象的功相等,假设两个对象只有动能没有势能,
则有关系
考虑收敛标准δ=1e-6的数据
惯性质量 | 0.5 | 0.6 | 0.7 | 0.8 | 0.9 | 1 | 1.1 | 1.2 | 1.3 | 1.4 | 1.5 | 1.6 | 1.7 | 1.8 | 1.9 | 2 | 2.1 | 2.2 |
1.00E-06 | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave |
相对速度 | 0.686023 | 0.696485 | 0.720947 | 0.747416 | 0.777551 | 0.807597 | 0.828894 | 0.837617 | 0.843693 | 0.848185 | 0.850499 | 0.858024 | 0.86993 | 0.885876 | 0.925262 | 0.969361 | 0.993003 | 0.999552 |
考虑0.5<=dis<=2.2的数据
如果用一个对数函数去拟合惯性质量m,则可以得到表达式
0.9667552609375146 ****** 决定系数 r**2
得到的数据
惯性质量m | 0.5 | 0.6 | 0.7 | 0.8 | 0.9 | 1 | 1.1 | 1.2 | 1.3 | 1.4 | 1.5 | 1.6 | 1.7 | 1.8 | 1.9 | 2 | 2.1 | 2.2 |
1.00E-06 | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave | 平均准确率p-ave |
相对速度v | 0.686023 | 0.696485 | 0.720947 | 0.747416 | 0.777551 | 0.807597 | 0.828894 | 0.837617 | 0.843693 | 0.848185 | 0.850499 | 0.858024 | 0.86993 | 0.885876 | 0.925262 | 0.969361 | 0.993003 | 0.999552 |
v^2 | 0.470627 | 0.485092 | 0.519765 | 0.55863 | 0.604586 | 0.652212 | 0.687066 | 0.701601 | 0.711818 | 0.719418 | 0.723348 | 0.736205 | 0.756779 | 0.784776 | 0.85611 | 0.939661 | 0.986055 | 0.999105 |
换算后的质量 | 2.129649 | 1.995764 | 1.882566 | 1.78451 | 1.698018 | 1.620648 | 1.550659 | 1.486763 | 1.427985 | 1.373565 | 1.322901 | 1.275509 | 1.23099 | 1.189017 | 1.149313 | 1.111647 | 1.075819 | 1.041658 |
w | 1.002271 | 0.968129 | 0.978491 | 0.996881 | 1.026598 | 1.057006 | 1.065404 | 1.043115 | 1.016465 | 0.988167 | 0.956918 | 0.939036 | 0.931587 | 0.933112 | 0.983938 | 1.044572 | 1.060817 | 1.040725 |
由假设1,2,3得到了dis和pave之间的数学关系,至少在0.5到2.2这个区间由已知的dis可以相对精确的预测训练集的pave。如果这一数学规则普遍成立,多维的数据集也应该存在dis和pave之间的数学关系,这意味dis可以唯一的决定网络的分类准确率。