神经网络训练集最少可以是多少个?
(mnist 0 ,2)---81*30*2---(1,0)(0,1)
用81*30*2的网络分类mnist的0和2。让训练集的数量n分别等于5000,4500,4000,3500,3000,2500,2000,1500,1000,500,400,300,200,100,50,40,30,20,10,5,4,3,2,共22个值。看看训练集的大小对分类结果到底有什么影响。
让收敛标准δ等于0.5到1e-5的25个值,每个值收敛199次,取平均值。因此共收敛了25*199*22次,首先比较迭代次数
5000 |
4500 |
4000 |
3500 |
3000 |
2500 |
2000 |
1500 |
1000 |
500 |
400 |
300 |
200 |
100 |
50 |
40 |
30 |
20 |
10 |
5 |
4 |
3 |
2 |
|
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 |
0.5 |
8.5025126 |
8.6834171 |
9.5778894 |
9.4371859 |
7.8341709 |
9.6934673 |
9.8140704 |
8.8341709 |
9.7537688 |
10.095477 |
10.065327 |
9.4974874 |
9.0150754 |
8.4572864 |
10.482412 |
9.0251256 |
9.3417085 |
8.321608 |
8.6030151 |
9.5778894 |
8.6633166 |
8.1758794 |
7.9447236 |
0.4 |
211.40201 |
210.37688 |
212.00503 |
210.46231 |
212.63317 |
213.46231 |
211.74874 |
213.21106 |
213.30653 |
210.75377 |
211.64322 |
211.88442 |
212.9397 |
213.55276 |
227.33166 |
235.95477 |
211.8794 |
195.67337 |
174.76382 |
171.79397 |
150.35176 |
129.59296 |
118.9799 |
0.3 |
271.01508 |
271.75377 |
272.50251 |
269.72362 |
269.22613 |
268.26131 |
267.67839 |
268.72864 |
268.68844 |
270.77889 |
269.22111 |
267.52764 |
268.20603 |
281.98492 |
295.76382 |
294.37186 |
272.61809 |
249.99497 |
225.26131 |
217.75377 |
195.09045 |
167.88442 |
157.98492 |
0.2 |
325.13568 |
325.26131 |
325.52261 |
321.67839 |
324.00503 |
325.17085 |
325.0804 |
325.42211 |
325.43216 |
325.38693 |
325.78392 |
325.87437 |
328.44724 |
322.22111 |
349.41206 |
340.33166 |
324.15578 |
298.72362 |
271.43216 |
261.93467 |
233.53266 |
204.02513 |
195.34673 |
0.1 |
410.8593 |
410.97487 |
409.00503 |
410.27638 |
411.20101 |
411.34673 |
409.16583 |
411.29146 |
409.62814 |
410.60302 |
410.75879 |
410.36181 |
408.59296 |
397.97487 |
418.71357 |
420.41206 |
396.0804 |
366.27136 |
335.72362 |
330.03518 |
296.15578 |
262.67839 |
262.76382 |
0.01 |
688.45226 |
695.28643 |
693.63317 |
689.74372 |
688.12563 |
688.92462 |
686.01508 |
692.93467 |
688.88945 |
687.72362 |
685.58291 |
745.15578 |
651.65829 |
720.39196 |
790.16583 |
785.43719 |
799.26633 |
691.66834 |
670.41709 |
785.43216 |
724.11558 |
688.52764 |
890.45729 |
0.001 |
1435.1357 |
1435.3719 |
1435.201 |
1440.8643 |
1441.4623 |
1434.4824 |
1440.0101 |
1432.7688 |
1441.0553 |
1372.9447 |
1359.8392 |
1413.9849 |
1443.7588 |
1736.0101 |
1960.397 |
1815.5025 |
2019.0402 |
1686.5829 |
1910.7889 |
3384.2462 |
3165.5729 |
3392.8693 |
5421.1558 |
9.00E-04 |
1459.3266 |
1450.7739 |
1468.0905 |
1458.8442 |
1457.2563 |
1470.3518 |
1455.9296 |
1445.6884 |
1454.2211 |
1430.8844 |
1389.608 |
1450.3116 |
1468.3719 |
1820.4874 |
2014.6432 |
1913.0905 |
2106.3719 |
1744.2513 |
2038.8693 |
3644.6784 |
3416.3367 |
3675.3417 |
5947.9045 |
8.00E-04 |
1474.7739 |
1495.1457 |
1480.4523 |
1472.9146 |
1512.7688 |
1498.4221 |
1479.7387 |
1487.8392 |
1492.2312 |
1494.603 |
1444.0503 |
1502.7688 |
1522.6834 |
1928.2764 |
2115.6281 |
1992.995 |
2198.5779 |
1875.2965 |
2170.5678 |
3979.9095 |
3711.9045 |
4048.3266 |
6568.804 |
7.00E-04 |
1557.2563 |
1570.2211 |
1553.206 |
1561.4271 |
1545.005 |
1577.6583 |
1556 |
1545.8794 |
1556.5226 |
1577.5779 |
1537.0553 |
1560.7186 |
1608.9246 |
2056.5477 |
2240.6834 |
2128.3668 |
2295.397 |
1992.3518 |
2360.2462 |
4397.1508 |
4146.3769 |
4542.598 |
7411.608 |
6.00E-04 |
1742.1206 |
1725.6281 |
1732.7035 |
1713.0151 |
1718.4724 |
1696.7337 |
1694.7035 |
1762.8744 |
1697.3668 |
1656.5729 |
1646.2814 |
1646.2513 |
1741.7889 |
2170.3015 |
2367.4724 |
2263.5126 |
2439.3668 |
2148.1206 |
2561.6834 |
5037.7337 |
4643.9849 |
5188.4472 |
8466.9548 |
5.00E-04 |
1961.4472 |
2003.9598 |
1919.0955 |
2015.8894 |
1967.5678 |
1989.8693 |
1975.5276 |
1974.1307 |
2019.005 |
1741.9397 |
1780.8643 |
1751.2462 |
1858.5427 |
2375.7035 |
2552.4422 |
2439.3116 |
2615.4673 |
2381.7085 |
2918.7889 |
5764.6281 |
5407.3417 |
5969.9347 |
9980.6784 |
4.00E-04 |
2182.2111 |
2193.1256 |
2188.1206 |
2177.2362 |
2185.8995 |
2202.2814 |
2203.3769 |
2175.4874 |
2203.2764 |
1914.0201 |
1870.3819 |
1922.8945 |
1983.4271 |
2638.9598 |
2779.7789 |
2657.8442 |
2848.3417 |
2701.5578 |
3299.7739 |
6930.2412 |
6531.9648 |
7238.2764 |
12146.844 |
3.00E-04 |
2349.397 |
2365.7085 |
2347.1055 |
2334.2814 |
2349.1558 |
2342.8844 |
2349.809 |
2358.3518 |
2296.6533 |
2164.8643 |
1993.0452 |
2099.2563 |
2178.7538 |
2920.5578 |
3172.4121 |
2973.4271 |
3210.1106 |
3099.1005 |
3957.5427 |
8742.0452 |
8331.7839 |
9349.6533 |
15832.533 |
2.00E-04 |
2731.1457 |
2733.6784 |
2744.7035 |
2708.7035 |
2710.7337 |
2732.3216 |
2728.5126 |
2718.4724 |
2498.7638 |
2442.8342 |
2314.2211 |
2451.2462 |
2522.1407 |
3393.6784 |
3674.1608 |
3478.8442 |
3757.196 |
3898.1055 |
5219.3518 |
12351.271 |
11827.01 |
13288.427 |
22821.206 |
1.00E-04 |
3228.6935 |
3250.1809 |
3276.6533 |
3237.2261 |
3259.5779 |
3205.6181 |
3128.9447 |
3242.2814 |
3391.1055 |
3194.1508 |
2912.3216 |
3138.4523 |
3102.8543 |
4219.8392 |
4922.4472 |
4632.4171 |
5125.1809 |
5944.3266 |
8577.8442 |
22394.99 |
21686.98 |
24946.296 |
43436.141 |
9.00E-05 |
3358.2915 |
3326.8744 |
3366.3317 |
3375.3367 |
3367.4573 |
3343.4372 |
3330.9347 |
3476.9849 |
3506.593 |
3296.1709 |
3048.7638 |
3277.6683 |
3177.5377 |
4403.1256 |
5123.799 |
4878.5327 |
5354.3367 |
6259.9497 |
9137.0201 |
24672.975 |
23955.583 |
27614.417 |
47741.548 |
8.00E-05 |
3521.9698 |
3551.1759 |
3578.4121 |
3543.4774 |
3531.3065 |
3544.1206 |
3512.6834 |
3647.005 |
3581.0452 |
3399.598 |
3149.5176 |
3389.196 |
3287.4673 |
4554.3618 |
5447.5678 |
5158.5025 |
5615.3467 |
6859.7638 |
10056.377 |
27109.884 |
26445.668 |
30678.457 |
53467.683 |
7.00E-05 |
3729.4673 |
3740.1106 |
3751.6884 |
3720.4422 |
3728.593 |
3650.9246 |
3736.804 |
3816.392 |
3720.9548 |
3598.4623 |
3305.4975 |
3528.3216 |
3436.3116 |
4761.8894 |
5758.6985 |
5453.1558 |
6047.2513 |
7309.8191 |
10892.879 |
30726.698 |
29946.859 |
34728.467 |
60645.221 |
6.00E-05 |
3914.7538 |
3968.6432 |
3917.2563 |
3929.3668 |
3919.6884 |
3932.6834 |
3901.6884 |
4179.6482 |
3994.2211 |
3802.9146 |
3488.3317 |
3693.6985 |
3584.4925 |
4960.7337 |
6122.407 |
5816.1859 |
6527.1558 |
8279.3467 |
12253.643 |
35125.136 |
34565.025 |
40030.739 |
69679.583 |
5.00E-05 |
4198.8342 |
4177.9698 |
4215.9196 |
4147.9598 |
4163.2261 |
4167.0452 |
4220.8643 |
4471.206 |
4295.4271 |
4119.5578 |
3688.5226 |
3944.8744 |
3733.9196 |
5270.7437 |
6528.407 |
6223.6985 |
7145.8492 |
9307.0653 |
14100.136 |
41277.849 |
41131.583 |
48023.04 |
82621.377 |
4.00E-05 |
4530.1407 |
4611.2362 |
4602.402 |
4561.7186 |
4567.1357 |
4505.3467 |
4659.2965 |
5006.3719 |
4719.8291 |
4415.7789 |
3869.608 |
4218.4221 |
3981.8191 |
5719.8995 |
7258.2312 |
6933.598 |
8020.7739 |
10533.543 |
16602.588 |
49964.261 |
49754.824 |
57922.136 |
101854.19 |
3.00E-05 |
5094.4824 |
5060.1106 |
5054.5025 |
5015.5578 |
5106.3216 |
4995.9698 |
5403.407 |
5719.2462 |
5226.7236 |
4860.6332 |
4238.7035 |
4659.7889 |
4302.7136 |
6179.5779 |
8160.0302 |
7968.0101 |
9089.201 |
13021.302 |
20963.653 |
64935.221 |
63982.342 |
76814.055 |
133678.95 |
2.00E-05 |
6474.4322 |
6472.8643 |
6431.8894 |
6435.7889 |
6231.4271 |
5973.5678 |
6456.4623 |
7055.2261 |
6338.1206 |
5585.0653 |
4811.0653 |
5317.5176 |
4768.8844 |
7156.6432 |
10168.864 |
9704.9799 |
11167.347 |
17203.809 |
28678.688 |
94762.523 |
94579.935 |
112784.95 |
196678.68 |
1.00E-05 |
8638.3719 |
8853.1457 |
8095.7387 |
8804.1809 |
8577.8995 |
8696.2312 |
9096.5427 |
11118.04 |
8461.6583 |
7061.0151 |
5876.6633 |
6564.2613 |
5674.8141 |
9151.5678 |
14010.166 |
13193.216 |
16357.859 |
27801.749 |
49539.965 |
175852.4 |
180663.83 |
215775.98 |
381101.72 |
迭代次数的最大值出现在训练集n的数量等于2的时候,而迭代次数最小的值出现在约n=200的位置。
训练集n=5000到n=200的图。也就是迭代次数随着训练集n的数量的减小,是先减小后增大的。
再比较分类准确率的平均值pave
5000 |
4500 |
4000 |
3500 |
3000 |
2500 |
2000 |
1500 |
1000 |
500 |
400 |
300 |
200 |
100 |
50 |
40 |
30 |
20 |
10 |
5 |
4 |
3 |
2 |
|
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 |
0.5 |
0.5192888 |
0.5285348 |
0.5225431 |
0.5110218 |
0.5205626 |
0.5269464 |
0.5150129 |
0.529359 |
0.5204577 |
0.5326708 |
0.5165015 |
0.5238219 |
0.5275857 |
0.5167088 |
0.5261022 |
0.5116312 |
0.5162068 |
0.5160669 |
0.5263744 |
0.5172308 |
0.5183722 |
0.5139565 |
0.5109294 |
0.4 |
0.874694 |
0.8783655 |
0.873665 |
0.8743843 |
0.8749063 |
0.8618365 |
0.8783905 |
0.8645414 |
0.8673237 |
0.8849141 |
0.8682703 |
0.8716045 |
0.8712724 |
0.8914178 |
0.8951492 |
0.9167932 |
0.8807532 |
0.8607401 |
0.7406915 |
0.7705426 |
0.7312931 |
0.7919943 |
0.7101337 |
0.3 |
0.9513447 |
0.9512398 |
0.9501034 |
0.9513872 |
0.9521639 |
0.9489121 |
0.9501334 |
0.9506204 |
0.9490269 |
0.9503881 |
0.951075 |
0.9491044 |
0.9506654 |
0.9516344 |
0.9533503 |
0.9506504 |
0.9350855 |
0.9299554 |
0.8990229 |
0.8930487 |
0.8706655 |
0.8753434 |
0.778902 |
0.2 |
0.9366539 |
0.9373757 |
0.9373757 |
0.9366739 |
0.9378428 |
0.9383048 |
0.9406626 |
0.9376505 |
0.9378028 |
0.9374906 |
0.9389592 |
0.9386545 |
0.9394862 |
0.9592096 |
0.9579233 |
0.9530705 |
0.9394862 |
0.9416191 |
0.9253549 |
0.9018777 |
0.8931437 |
0.8569063 |
0.7845065 |
0.1 |
0.9586551 |
0.958218 |
0.9571516 |
0.9574188 |
0.9584103 |
0.9579733 |
0.9575936 |
0.958283 |
0.9574338 |
0.9584378 |
0.9579533 |
0.9580931 |
0.9514521 |
0.9609429 |
0.9647092 |
0.9584628 |
0.9415817 |
0.9471388 |
0.9354301 |
0.9012708 |
0.8983586 |
0.8439838 |
0.7785049 |
0.01 |
0.9359122 |
0.939656 |
0.9367139 |
0.940765 |
0.9369062 |
0.9365665 |
0.9397459 |
0.9374557 |
0.9365316 |
0.9390866 |
0.9408449 |
0.9758359 |
0.9614924 |
0.9696145 |
0.9737954 |
0.9642122 |
0.9532853 |
0.9521115 |
0.942281 |
0.901508 |
0.8995574 |
0.835025 |
0.7750132 |
0.001 |
0.9765053 |
0.9765153 |
0.9765228 |
0.9766127 |
0.9766027 |
0.9764928 |
0.9764329 |
0.97676 |
0.9766751 |
0.9783485 |
0.9784759 |
0.9770273 |
0.9771896 |
0.9752515 |
0.9702663 |
0.9666149 |
0.9532953 |
0.9535351 |
0.9444789 |
0.9020001 |
0.8999171 |
0.8353922 |
0.7743139 |
9.00E-04 |
0.9767725 |
0.9766526 |
0.9767101 |
0.9766576 |
0.9766526 |
0.9767476 |
0.9767625 |
0.9766626 |
0.9766177 |
0.9779064 |
0.9785058 |
0.9782486 |
0.9769998 |
0.9751841 |
0.9700241 |
0.9670295 |
0.9531554 |
0.9534751 |
0.9445263 |
0.9022573 |
0.8997497 |
0.8354771 |
0.7748459 |
8.00E-04 |
0.9767476 |
0.9767001 |
0.976775 |
0.9765028 |
0.976795 |
0.976835 |
0.9766826 |
0.9766402 |
0.976765 |
0.9780913 |
0.9785009 |
0.9784459 |
0.9772021 |
0.9749343 |
0.9700241 |
0.9673142 |
0.9529581 |
0.9534976 |
0.9445388 |
0.9028792 |
0.9001643 |
0.8352523 |
0.7750182 |
7.00E-04 |
0.9769798 |
0.9767675 |
0.9768799 |
0.9769948 |
0.9769349 |
0.9769574 |
0.9769623 |
0.9768749 |
0.9769798 |
0.9780962 |
0.9782186 |
0.9779814 |
0.9770997 |
0.974732 |
0.9698193 |
0.967529 |
0.9525435 |
0.9534052 |
0.9446986 |
0.9023272 |
0.8996623 |
0.8352823 |
0.7756651 |
6.00E-04 |
0.9775418 |
0.9777166 |
0.9776792 |
0.9775992 |
0.9778065 |
0.9777666 |
0.9774694 |
0.9779089 |
0.9781712 |
0.978366 |
0.9783035 |
0.9781287 |
0.9770098 |
0.9748369 |
0.9696719 |
0.9676489 |
0.9518942 |
0.9535476 |
0.944806 |
0.9026494 |
0.8998571 |
0.8348927 |
0.7756102 |
5.00E-04 |
0.977809 |
0.9776092 |
0.9779913 |
0.9779264 |
0.9775768 |
0.9777641 |
0.978361 |
0.9778215 |
0.9798296 |
0.9785158 |
0.9786632 |
0.9786157 |
0.9776692 |
0.9749343 |
0.9697294 |
0.9676089 |
0.9514296 |
0.9534676 |
0.9448235 |
0.9024446 |
0.899977 |
0.8349476 |
0.7769264 |
4.00E-04 |
0.9756261 |
0.9753639 |
0.9753964 |
0.9754963 |
0.9752915 |
0.9761381 |
0.9755962 |
0.9753864 |
0.9798271 |
0.9784484 |
0.9788455 |
0.9786457 |
0.977814 |
0.9746321 |
0.9700515 |
0.967564 |
0.9507403 |
0.9533952 |
0.9450233 |
0.9023422 |
0.8998796 |
0.8352049 |
0.7751231 |
3.00E-04 |
0.9738104 |
0.9743149 |
0.9742075 |
0.9739877 |
0.9735232 |
0.9740427 |
0.9738853 |
0.9743049 |
0.9783435 |
0.9784534 |
0.9778815 |
0.9771821 |
0.9771472 |
0.9739053 |
0.9698393 |
0.9676988 |
0.9500734 |
0.953048 |
0.9454304 |
0.9028867 |
0.8997697 |
0.8353048 |
0.7744263 |
2.00E-04 |
0.9798296 |
0.9802467 |
0.9797996 |
0.9794424 |
0.979405 |
0.9797821 |
0.9798346 |
0.9793375 |
0.9789804 |
0.9770772 |
0.9783535 |
0.9765502 |
0.9760482 |
0.9735407 |
0.969567 |
0.9675415 |
0.950051 |
0.9528108 |
0.9456702 |
0.9028492 |
0.9001868 |
0.8352298 |
0.7748584 |
1.00E-04 |
0.9809535 |
0.9808111 |
0.9808885 |
0.9809984 |
0.980916 |
0.980936 |
0.9812082 |
0.980931 |
0.9789229 |
0.9767476 |
0.9756911 |
0.9735207 |
0.9745971 |
0.9730636 |
0.9686055 |
0.9668746 |
0.9501933 |
0.9524936 |
0.9461622 |
0.9037184 |
0.9007887 |
0.8344831 |
0.7744013 |
9.00E-05 |
0.9807462 |
0.9807712 |
0.9805089 |
0.9807412 |
0.9807162 |
0.9807362 |
0.9806962 |
0.9807262 |
0.9787706 |
0.9763929 |
0.9753914 |
0.9730062 |
0.9745647 |
0.9728214 |
0.968523 |
0.9666998 |
0.9500085 |
0.9524511 |
0.9463021 |
0.9032114 |
0.9007013 |
0.8351 |
0.774219 |
8.00E-05 |
0.9807237 |
0.9804465 |
0.9805688 |
0.9805214 |
0.9805639 |
0.9806613 |
0.9803016 |
0.9805564 |
0.978858 |
0.9759034 |
0.9755687 |
0.9725616 |
0.9744598 |
0.9727839 |
0.9684631 |
0.9665924 |
0.9499635 |
0.9523812 |
0.946402 |
0.9032838 |
0.9003392 |
0.8343682 |
0.7755552 |
7.00E-05 |
0.9807437 |
0.9806563 |
0.9806663 |
0.9808161 |
0.9807761 |
0.9805788 |
0.980409 |
0.9804764 |
0.9791228 |
0.9757285 |
0.9762181 |
0.9722145 |
0.974275 |
0.9725766 |
0.9682958 |
0.9664475 |
0.9499186 |
0.9525435 |
0.9464669 |
0.9039956 |
0.9007188 |
0.8348402 |
0.7744538 |
6.00E-05 |
0.9812232 |
0.981408 |
0.9812232 |
0.9812931 |
0.981443 |
0.9812632 |
0.9807212 |
0.9807487 |
0.9786632 |
0.9754288 |
0.9751416 |
0.9716525 |
0.9741426 |
0.9725591 |
0.9682333 |
0.9662927 |
0.9498187 |
0.9521989 |
0.9464744 |
0.9037434 |
0.9003292 |
0.834558 |
0.7767016 |
5.00E-05 |
0.9823172 |
0.9823022 |
0.9821948 |
0.982005 |
0.9821173 |
0.9822322 |
0.9808486 |
0.9808686 |
0.9782336 |
0.9750217 |
0.9737405 |
0.9714677 |
0.9739178 |
0.9725966 |
0.9683257 |
0.9662502 |
0.9495739 |
0.9520041 |
0.9466567 |
0.9037783 |
0.9004166 |
0.8348727 |
0.7755677 |
4.00E-05 |
0.9830939 |
0.9831538 |
0.9829091 |
0.9828991 |
0.9828192 |
0.9824245 |
0.9808186 |
0.9809035 |
0.978316 |
0.9746671 |
0.9732984 |
0.9707709 |
0.9736006 |
0.972714 |
0.9681709 |
0.9661428 |
0.9495864 |
0.9520615 |
0.9467292 |
0.9035935 |
0.9006064 |
0.8346179 |
0.7746211 |
3.00E-05 |
0.9828616 |
0.9830764 |
0.9830215 |
0.9826293 |
0.9826418 |
0.9817527 |
0.9814755 |
0.9812307 |
0.9791802 |
0.9739628 |
0.9727489 |
0.970064 |
0.9731585 |
0.9729662 |
0.9680785 |
0.9660454 |
0.9494016 |
0.9520465 |
0.946944 |
0.9039282 |
0.9007463 |
0.8347403 |
0.7752405 |
2.00E-05 |
0.9834086 |
0.983551 |
0.9835035 |
0.9803541 |
0.9821248 |
0.980921 |
0.9812682 |
0.9813881 |
0.9789155 |
0.9731835 |
0.9722045 |
0.9694247 |
0.9724667 |
0.9732185 |
0.9678936 |
0.965968 |
0.9495315 |
0.9517643 |
0.9470539 |
0.9039756 |
0.9007338 |
0.8342483 |
0.774761 |
1.00E-05 |
0.9823072 |
0.9822722 |
0.9812332 |
0.979892 |
0.9826968 |
0.9822322 |
0.9803466 |
0.9804814 |
0.9778015 |
0.970616 |
0.9718323 |
0.9689027 |
0.9713728 |
0.9728813 |
0.9678262 |
0.9659805 |
0.9493292 |
0.9515645 |
0.9473311 |
0.9045501 |
0.9011159 |
0.835025 |
0.7762346 |
随着训练集n的减小分类准确率是减小的,
训练集数量n=5000到n=500的图像,
5000 |
500 |
||
5.00E-05 |
0.982317 |
0.975022 |
0.992573 |
4.00E-05 |
0.983094 |
0.974667 |
0.991428 |
3.00E-05 |
0.982862 |
0.973963 |
0.990946 |
2.00E-05 |
0.983409 |
0.973184 |
0.989602 |
1.00E-05 |
0.982307 |
0.970616 |
0.988098 |
0.99053 |
比较n=500和n=5000的数据,虽然将训练集的数量减小到原来的1/10,但分类准确率只下降了约1%。再比较n=5000和n=2500的数据
5000 |
2500 |
2500/5000 |
|
5.00E-05 |
0.982317 |
0.982232 |
0.999914 |
4.00E-05 |
0.983094 |
0.982425 |
0.999319 |
3.00E-05 |
0.982862 |
0.981753 |
0.998872 |
2.00E-05 |
0.983409 |
0.980921 |
0.99747 |
1.00E-05 |
0.982307 |
0.982232 |
0.999924 |
0.9991 |
训练集数量下降到一半,分类准确率下降约1‰,也就表明对这个网络完全可以将训练集的数量减到一半,分类差异不大。
Mnist的数据集的图片是从1开始编号,因此训练集的数量n=2意味着可以用1张图片实现分类。尽管分类准确率损失比较大。
因此对这个网络来说,从实用角度训练集数量的最小值可以是原来的50%,甚至是10%。但如果仅让网络保持基本的分类能力,训练集数量的最小值是1个。
神经网络训练集最少可以是多少个?相关推荐
- 神经网络训练集与Fock矩阵
如果将神经网络的训练集理解成哈密顿量,但是哈密顿算符并不是矩阵形式,比如拉普拉斯算符 而神经网络的训练集都是矩阵,如何才能将数学解析式变成矩阵的形式? 一种可能的办法是Fock矩阵方法,比如假设波函数 ...
- 神经网络训练集两张图片之间的相互作用
在<神经网络分类训练集的图片到底是如何相互影响的?>中得到了一个经验关系,如果(两张图片)混合后的迭代次数变小了,分类准确率可能变大:如果二者混合后迭代次数变大,分类准确率可能会变小.但上 ...
- 深度神经网络(训练集,验证集,测试集), 提升模型效果,交叉验证
随笔 神经网络训练确实是loss越小,效果越好,但不是绝对的. 因为损失小只是对训练集而言,所以要划出部分数据不做训练,只做验证,即为验证集,验证集对网络训练是没影响的, 所以要将数据集分成:训练集, ...
- 神经网络训练数据集下载,神经网络训练集数量
200组数据可以训练神经网络吗 谷歌人工智能写作项目:神经网络伪原创 BP神经网络的训练集需要大样本吗?一般样本个数为多少? BP神经网络的训练集需要大样本吗?一般样本个数为多少? BP神经网络样本数 ...
- 神经网络训练集的图片到底是如何相互作用的?
本次实验分别让0的第1张图片到第9张图片和2的第1张图片分类.然后再让0的第1,2张,3,4张-8,9张图片和2的第1张图片分类,看看两组数据有什么关系. 第一部分数据,1张图片和1张图片分类 (0[ ...
- 人工构造迭代次数高度简并的神经网络训练集
(x)---81*30*2---(1,0) 做一个结构是81*30*2的网络,人工构造一个训练集x,并让这个训练集向(1,0)收敛,不分类.让收敛标准为1e-4.使得迭代次数高度简并,有明显的谱线特征 ...
- 训练集数量对神经网络光谱的影响
(mnist 0 ,2)---81*30*2---(1,0)(0,1) 分类mnist的0和2.让训练集的数量n分别等于5000,4500,4000,3500,3000,2500,2000,1500, ...
- 神经网络一般训练次数,神经网络训练时间
为什么要批量训练神经网络 谷歌人工智能写作项目:小发猫 神经网络中学习率.批处理样本数量.迭代次数有什么意义和影响? 学习率是指每次训练过程中(迭代)变量改变(更新)的比率,例如x(t+1)=x(t) ...
- 一个训练集未知的神经网络
如有一个神经网络已经训练好了,则可以通过改变测试集与训练集相似程度的办法改变这个网络的分类准确率. (0,1)---m*n*k---(1,0)(0,1) 比如一个分类0和1的网络,如果测试集的图片写的 ...
最新文章
- 在Ubuntu 16.04.3 LTS上玩转quic-go项目
- dev gridcontrol summaryitem如何加条件_如何一次清洗1000根核磁管
- mysql after 不起作用_我的MySQL触发器不起作用,语法简单,不复杂
- 【机器学习基础】使用python实现BP算法
- 人工智能AI实战100讲(四)-基于深度学习的图像增强综述
- nginx部署,nuxt静态部署,路由跳转失效的问题
- Python学习之==常用模块
- java 找到一行 更换单词_Java实现对一行英文进行单词提取功能示例
- 高中数学40分怎么办_高中数学新教材都有哪些变化?
- rac san+oracle_Oracle RAC安装部署之规划(一)
- 重置Winsock2
- neo4j中实现关键路径算法
- Lorenz系统、简单的Rossler系统和Chua电路系统的混沌吸引子——MATLAB实现
- matlab研究微分方程,基于MATLAB的微分方程求解电路仿真实验研究与实现
- Google地图查地址、经纬度
- PPT文件如何解决不能编辑的问题
- 容器启动失败 ERROR: for log Cannot start service log: OCI runtime create failed: container_linux.go:346
- Linux c 中alarm,signal,pause的使用说明
- c语言程序 蟠桃记,蟠桃记
- Sparse coding:
热门文章
- DF以某一列的元素筛选其中属于某个集合的元素的所有行(2个版本函数)
- 河南工程学院计算机科学与技术在哪个校区,河南工学院新生在哪个校区_有几个校区及地址介绍...
- 创建 Npm+webpack
- Laravel 系列入门教程(一)【最适合中国人的 Laravel 教程】
- IDApython 命令
- python3 第三十四章 - 聊聊File对象
- 突袭HTML5之SVG 2D入门1 - SVG综述
- UA PHYS515 电磁理论I 麦克斯韦方程组基础4 介质中的麦克斯韦方程
- UA MATH571B 2K析因设计 SAS实践 分数2k析因设计
- UA MATH564 概率论I 求离散型随机变量的分布1