这是今天最好成绩，很开心，上网买条春秋穿的裤子奖励自己！
看架构：（训练cifar10，83.10版本）
//resnet18//resnet18//resnet18//使用标签平滑，所有relu都是leaky relu
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 5, 32, 32, 32, 3, 1, 1));
//第一残差开始
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 32, 32, 32));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 32, 32, 32));
//第二残差开始
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 32, 64, 32, 32, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 64, 32, 32));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 64, 32, 32));
//第三残差开始
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 64, 32, 32));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 64, 32, 32));
//第四残差开始，虚线 //xuxianResidual
layers.emplace_back(std::make_shared<xuxianResidual2>(cudnn, batch, 64, 32, 32));//out->batch, 128, 16, 16
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 128, 16, 16));
//第五残差开始，residualExt2+relu是一个完整残差块
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 128, 16, 16));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 128, 16, 16));
//第六残差开始
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 128, 16, 16));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 128, 16, 16));
//第七残差开始，虚线 //xuxianResidual2+relu是一个完整残差块
layers.emplace_back(std::make_shared<xuxianResidual2>(cudnn, batch, 128, 16, 16));//out->batch, 256, 8, 8
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 256, 8, 8));
//八残差开始
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 256, 8, 8));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 256, 8, 8));
//第九残差开始
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 256, 512, 8, 8, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 512, 8, 8));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 512, 8, 8));
//一共9个残差块
layers.emplace_back(std::make_shared<AvgPool2D>(cudnn, batch, 512, 8,8, 2, 2, 0, 2));
layers.emplace_back(std::make_shared<Linear>(cublas, batch, 512 * 16, 384));
layers.emplace_back(std::make_shared<BN>(cudnn, batch, 384, 1, 1));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 384, 1, 1));
layers.emplace_back(std::make_shared<Dropout>(cudnn, batch, 384, 1, 1));//=0.5
layers.emplace_back(std::make_shared<Linear>(cublas, batch, 384, 10));
这个架构对比resnet18（78分版本）：（感觉便复杂了！）
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 5, 64, 32, 32, 3, 1, 1));
//第一残差开始
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 64, 32, 32));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 64, 32, 32));
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 64, 32, 32));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 64, 32, 32));
//第三残差开始，虚线 //xuxianResidual
layers.emplace_back(std::make_shared<xuxianResidual2>(cudnn, batch, 64, 32, 32));//out->batch, 128, 16, 16
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 128, 16, 16));
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 128, 16, 16));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 128, 16, 16));
//第五残差开始
layers.emplace_back(std::make_shared<xuxianResidual2>(cudnn, batch, 128, 16, 16));//out->batch, 256, 8, 8
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 256, 8, 8));
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 256, 8, 8));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 256, 8, 8));
//第七残差开始
layers.emplace_back(std::make_shared<xuxianResidual2>(cudnn, batch, 256, 8, 8));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 512, 4, 4));
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 512, 4, 4));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 512, 4, 4));
//一共8个残差块
//全局平均池化
layers.emplace_back(std::make_shared<GaverPool2D>(cudnn, batch, 512, 4, 4//全局平均池化成功
, 4, 4
, 0//pading
, 1//stride
));
layers.emplace_back(std::make_shared<Linear>(cublas, batch, 512, 10));
---------------------------------以下是最好85.78版本（全部实线残差实现），可以对比着看：
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 5, 32, 32, 32, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt22>(cudnn, batch, 32, 32, 32));
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 32, 64, 32, 32, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt22>(cudnn, batch, 64, 32, 32));
//layers.emplace_back(std::make_shared<MaxPool2D>(cudnn, batch, 64, 32, 32, 2, 2, 0, 2));
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 64, 128, 32, 32, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt22>(cudnn, batch, 128, 32, 32));
layers.emplace_back(std::make_shared<MaxPool2D>(cudnn, batch, 128, 32, 32, 2, 2, 0, 2));
//没有标签平滑，residualExt22是leakyrelu后+x方式
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 128, 256, 16, 16, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt22>(cudnn, batch, 256, 16, 16));
layers.emplace_back(std::make_shared<MaxPool2D>(cudnn, batch, 256, 16, 16, 2, 2, 0, 2));
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 256, 512, 8, 8, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt22>(cudnn, batch, 512, 8, 8));
layers.emplace_back(std::make_shared<averPool2D>(cudnn, batch, 512, 8, 8, 2, 2, 0, 2));
// 一共5个残差块
layers.emplace_back(std::make_shared<Linear>(cublas, batch, 512*16, 384));
layers.emplace_back(std::make_shared<BN>(cudnn, batch, 384, 1, 1));
layers.emplace_back(std::make_shared<LeakyRL>(cudnn, batch, 384, 1, 1));
layers.emplace_back(std::make_shared<Dropout >(cudnn, batch, 384, 1, 1));//=0.5
layers.emplace_back(std::make_shared<Linear>(cublas, batch, 384, 10));
下面是虚实结合版本，训练结果：（均值方差控制不错！耗时与85.78(5个残差)版本都是49秒）
轮次：0
learn rate：0.001
时间: 41564.335938 ms
train Classification result: 10.02% ok (used 49984 images)
时间: 3580.116943 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：1
learn rate：0.001
时间: 48624.343750 ms
train Classification result: 18.97% ok (used 49984 images)
时间: 3562.688965 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：2
learn rate：0.001
时间: 48703.542969 ms
train Classification result: 42.21% ok (used 49984 images)
时间: 3417.924072 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：3
learn rate：0.001
时间: 47049.328125 ms
train Classification result: 54.80% ok (used 49984 images)
时间: 3439.458984 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：4
learn rate：0.001
时间: 47071.804688 ms
train Classification result: 63.31% ok (used 49984 images)
时间: 3435.260010 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：5
learn rate：0.001
时间: 47160.933594 ms
train Classification result: 69.32% ok (used 49984 images)
时间: 3446.052002 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：6
learn rate：0.001
时间: 47375.757812 ms
train Classification result: 73.16% ok (used 49984 images)
时间: 3442.916992 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：7
learn rate：0.001
时间: 47288.992188 ms
train Classification result: 76.01% ok (used 49984 images)
时间: 3439.748047 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：8
learn rate：0.001
rb均值: -0.9556195140,rb方差:5.076510429382
时间: 47364.398438 ms
train Classification result: 78.72% ok (used 49984 images)
时间: 3460.628906 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：9
learn rate：0.001
rb均值: -0.9417217970,rb方差:6.015316486359
时间: 47465.812500 ms
train Classification result: 80.87% ok (used 49984 images)
时间: 3472.185059 ms
Test Classification result: 9.98% ok (used 9984 images)
轮次：10
learn rate：0.001
rb均值: -1.1234992743,rb方差:5.018473625183
rb均值: -0.8293356299,rb方差:5.748104572296
rb均值: -1.2596969604,rb方差:8.631543159485
时间: 47402.671875 ms
train Classification result: 82.90% ok (used 49984 images)
时间: 3451.395020 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：11
learn rate：0.001
rb均值: 0.4612812102,rb方差:5.783927440643
rb均值: -1.0449855328,rb方差:6.379163265228
rb均值: -1.2443413734,rb方差:6.815364360809
rb均值: -0.6524645090,rb方差:7.917567253113
rb均值: -1.3998067379,rb方差:11.942907333374
rb均值: 0.2518322170,rb方差:5.853224277496
rb均值: -0.4105462134,rb方差:5.870759963989
rb均值: -0.0995458439,rb方差:5.018486499786
rb均值: -2.6136467457,rb方差:5.115705966949
时间: 49133.164062 ms
train Classification result: 84.68% ok (used 49984 images)
时间: 3661.134033 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：12
learn rate：0.001
rb均值: -0.0834091902,rb方差:5.268047332764
rb均值: 0.5993465185,rb方差:5.013388156891
rb均值: 0.1550722718,rb方差:7.692845344543
rb均值: -0.2239504755,rb方差:5.128870964050
rb均值: -0.7764792442,rb方差:6.014815807343
rb均值: -0.9981870651,rb方差:8.860441207886
rb均值: 0.3065760434,rb方差:5.472651481628
rb均值: -1.4053789377,rb方差:10.108287811279
rb均值: -0.6342469454,rb方差:11.414999008179
rb均值: -1.6827325821,rb方差:16.235258102417
rb均值: 0.6299193501,rb方差:5.404631614685
rb均值: -0.4375291765,rb方差:8.573559761047
rb均值: -0.6945425868,rb方差:7.837462425232
rb均值: -1.4719511271,rb方差:5.201077461243
rb均值: -0.1822230220,rb方差:6.499783515930
rb均值: 1.4212073088,rb方差:6.212995529175
rb均值: -2.9409224987,rb方差:6.198729991913
时间: 49196.257812 ms
train Classification result:86.21% ok (used 49984 images)
时间: 3637.355957 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：13
learn rate：0.0001
rb均值: -0.0582699999,rb方差:5.782805442810
rb均值: 0.4773569405,rb方差:5.270914077759
rb均值: 0.0506652817,rb方差:8.331352233887
rb均值: -0.2650005817,rb方差:5.282112598419
rb均值: -0.9031506777,rb方差:6.436485290527
rb均值: -0.9766959548,rb方差:9.328955650330
rb均值: 0.3824808002,rb方差:6.198109626770
rb均值: -1.4969514608,rb方差:10.321689605713
rb均值: -0.7338016629,rb方差:12.408943176270
rb均值: -1.9699673653,rb方差:17.594989776611
rb均值: 0.5892682076,rb方差:5.594651699066
rb均值: -0.4762113094,rb方差:9.016823768616
rb均值: -0.7005262375,rb方差:8.267576217651
rb均值: -1.5033551455,rb方差:5.616190433502
rb均值: 0.3212858438,rb方差:5.280098915100
rb均值: -0.0744815469,rb方差:6.679601192474
rb均值: 1.4984625578,rb方差:6.322074413300
rb均值: -2.9965698719,rb方差:6.448714733124
时间: 49454.171875 ms
train Classification result: 91.86% ok (used 49984 images)
时间: 3673.620117 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：14
learn rate：0.0001
rb均值: -0.0327393077,rb方差:5.908860683441
rb均值: 0.4739944041,rb方差:5.208246231079
rb均值: -0.0130712083,rb方差:8.170441627502
rb均值: -0.2847779393,rb方差:5.493597507477
rb均值: -0.8846755624,rb方差:6.541474819183
rb均值: -1.0205224752,rb方差:9.304904937744
rb均值: 0.3461281955,rb方差:6.222161293030
rb均值: -1.5872770548,rb方差:10.306935310364
rb均值: -0.7820466161,rb方差:12.163268089294
rb均值: -1.9483764172,rb方差:17.580869674683
rb均值: 0.5974633098,rb方差:5.690805912018
rb均值: -0.4547826350,rb方差:9.036277770996
rb均值: -0.7059786916,rb方差:8.258251190186
rb均值: -1.5130012035,rb方差:5.718024253845
rb均值: 0.3299407065,rb方差:5.296885967255
rb均值: -0.0320599526,rb方差:6.713675498962
rb均值: 1.5218615532,rb方差:6.384797573090
rb均值: -2.9931437969,rb方差:6.457657337189
时间: 49281.781250 ms
train Classification result: 93.51% ok (used 49984 images)
时间: 3653.164062 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：15
learn rate：1e-05
rb均值: -0.0430723578,rb方差:5.874685764313
rb均值: 0.4959256351,rb方差:5.238189220428
rb均值: -0.0102475062,rb方差:8.151710510254
rb均值: -0.3014647961,rb方差:5.482594966888
rb均值: -0.8870178461,rb方差:6.476900100708
rb均值: -1.0182703733,rb方差:9.370736122131
rb均值: 0.3315711617,rb方差:6.285742759705
rb均值: -1.6002076864,rb方差:10.338290214539
rb均值: -0.7817471623,rb方差:12.205394744873
rb均值: -1.9541268349,rb方差:17.726718902588
rb均值: 0.5962846875,rb方差:5.713078498840
rb均值: -0.4490083456,rb方差:9.123505592346
rb均值: -0.6929129362,rb方差:8.334821701050
rb均值: -1.5182359219,rb方差:5.740020751953
rb均值: 0.3318162560,rb方差:5.291309356689
rb均值: -0.0155738583,rb方差:6.673139095306
rb均值: 1.5235553980,rb方差:6.418627262115
rb均值: -2.9985218048,rb方差:6.504796981812
时间: 49205.410156 ms
train Classification result: 94.63% ok (used 49984 images)
时间: 3683.264893 ms
Test Classification result: 82.63% ok (used 9984 images)
轮次：16
learn rate：1e-05
rb均值: -0.0463114455,rb方差:5.888206005096
rb均值: 0.4902145267,rb方差:5.247935295105
rb均值: -0.0084887464,rb方差:8.166854858398
rb均值: -0.2978139222,rb方差:5.477181911469
rb均值: -0.8966431022,rb方差:6.478566169739
rb均值: -1.0230754614,rb方差:9.345773696899
rb均值: 0.3366082013,rb方差:6.297463417053
rb均值: -1.6075497866,rb方差:10.362123489380
rb均值: -0.7749924660,rb方差:12.268142700195
rb均值: -1.9577939510,rb方差:17.812854766846
rb均值: 0.6031007171,rb方差:5.692900657654
rb均值: -0.4484292567,rb方差:9.132855415344
rb均值: -0.7030869722,rb方差:8.313133239746
rb均值: -1.5199253559,rb方差:5.753658771515
rb均值: 0.3339170218,rb方差:5.322653293610
rb均值: -0.0105585251,rb方差:6.657339572906
rb均值: 1.5267819166,rb方差:6.445106029510
rb均值: -3.0003211498,rb方差:6.529494285583
时间: 47487.671875 ms
train Classification result: 94.77% ok (used 49984 images)
时间: 3665.245117 ms
Test Classification result: 82.50% ok (used 9984 images)
轮次：17
learn rate：1e-06
rb均值: -0.0455969311,rb方差:5.895104408264
rb均值: 0.4897707403,rb方差:5.252963542938
rb均值: -0.0094080912,rb方差:8.173747062683
rb均值: -0.2963077426,rb方差:5.476254463196
rb均值: -0.8992136717,rb方差:6.480432987213
rb均值: -1.0244648457,rb方差:9.343390464783
rb均值: 0.3376792073,rb方差:6.288130760193
rb均值: -1.6076655388,rb方差:10.376107215881
rb均值: -0.7755109072,rb方差:12.282517433167
rb均值: -1.9598026276,rb方差:17.841623306274
rb均值: 0.6033918858,rb方差:5.693520545959
rb均值: -0.4501479864,rb方差:9.131182670593
rb均值: -0.7041436434,rb方差:8.311721801758
rb均值: -1.5203974247,rb方差:5.754602909088
rb均值: 0.3343850672,rb方差:5.325381755829
rb均值: -0.0097342376,rb方差:6.655287265778
rb均值: 1.5267264843,rb方差:6.450014591217
rb均值: -3.0008618832,rb方差:6.532120227814
时间: 49183.371094 ms
train Classification result: 94.79% ok (used 49984 images)
时间: 3674.278076 ms
Test Classification result: 83.06% ok (used 9984 images)
轮次：18
learn rate：1e-06
rb均值: -0.0442814827,rb方差:5.897310256958
rb均值: 0.4891677797,rb方差:5.251501560211
rb均值: -0.0097068641,rb方差:8.174299240112
rb均值: -0.2959270775,rb方差:5.476613521576
rb均值: -0.8996802568,rb方差:6.479226589203
rb均值: -1.0245624781,rb方差:9.339140892029
rb均值: 0.3371581733,rb方差:6.288616657257
rb均值: -1.6086497307,rb方差:10.371869087219
rb均值: -0.7766163349,rb方差:12.278637886047
rb均值: -1.9597586393,rb方差:17.837800979614
rb均值: 0.6033881903,rb方差:5.692530632019
rb均值: -0.4505719841,rb方差:9.129829406738
rb均值: -0.7050693631,rb方差:8.312390327454
rb均值: -1.5203868151,rb方差:5.754531383514
rb均值: 0.3345035613,rb方差:5.325196266174
rb均值: -0.0090401666,rb方差:6.656464576721
rb均值: 1.5267961025,rb方差:6.443959236145
rb均值: -3.0004746914,rb方差:6.536443233490
时间: 49547.503906 ms
train Classification result: 94.72% ok (used 49984 images)
时间: 3390.149902 ms
Test Classification result: 83.10% ok (used 9984 images)
下面是main函数：看看lr的控制！（与85.78版本一致）
cifar10Dataset Datasets = cifar10Dataset{ "c:\\data_batch_1.bin", "c:\\data_batch_1.bin" };
cifar10DatasetTEST DatasetsTest = cifar10DatasetTEST{ "c:\\test_batch_1.bin", "c:\\test_batch_1.bin" };
int dev = 0;
cudaSetDevice(dev);
// 创建流
cudaStream_t stream;
checkCuda(cudaStreamCreate(&stream));
cublasHandle_t cublas;
cudnnHandle_t cudnn;
cublasCreate(&cublas);//句柄是与当前上下文绑定的，不能一个句柄执行在不同GPU
cudnnCreate(&cudnn);
int batch_size = 10; batch_size = 32; batch_size = 64;// batch_size = 128;
float lr = 0.018f; lr = 0.001f;
//lr = 0.024f;
float bijiao = 0;
float* chengji;
chengji0 = 0;
int 起作用 = 0;
LeNet LeNet_net(cublas, cudnn, batch_size);
//for (int i = 0; i <25; i++)
for (int i = 0; i < 80; i++)//训练成绩达到100四十次，test成绩下降到52
{//训练四十次，test成绩64.95
std::cout << "轮次：" << i << std::endl;
std::cout << "learn rate：" << lr << std::endl;
bool hehe= train(781, batch_size, lr, LeNet_net, chengji, stream, Datasets);
float testscore= test2(156, batch_size, LeNet_net, stream, DatasetsTest);//10000/64=156
if (hehe)
{
lr *= 0.1;
}
if (lr <= 0.001)lr = 0.001;
if (chengji0 >= 85)
// if (chengji0 >= 88)
{
起作用++;
lr = 0.0001;
if (起作用 >=3)
{
lr = 0.00001;//这个78.32分，创纪录了
if(起作用 >= 5)//为什么需要这句话拖延一下时间，给gpu
{
////学习cos调整lr，83.5分左右
//lr = 0.0000001 + 0.5 * (0.00001 - 0.0000001) * (1 + ::cos(3.14159 * (i-25 + 1) / 90.0));//if(i>=4)
//std::cout << "调整后learn rate：" << lr << std::endl;
// lr = lr - i * 0.00001 / 60;
//这句话注释了，就是不行！202608031122
//i = 100;//执行5次
lr = 0.000001;
if (起作用 >= 7)
{
/*if (起作用 == 7)jilui = i;
lr = 0.0000001 + 0.5 * (0.000001 - 0.0000001) * (1 + ::cos(3.14159 * (i - jilui+1 ) / 6.0));
if (lr == 0.0000001)i = 100;*/
lr = 0.0000001;
if (起作用 >= 9)//上84分
{
i = 100;
}
}
}
}
}
}
下面是9月10教师节的突破：（85.78版本），写在一起好对比着看：（均值方差比83.10版本好）
11:04 2026/9/10
轮次：0
learn rate：0.001
时间: 45032.265625 ms
train Classification result: 10.00% ok (used 49984 images)
时间: 3710.165039 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：1
learn rate：0.001
时间: 49449.000000 ms
train Classification result: 10.19% ok (used 49984 images)
时间: 3701.998047 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：2
learn rate：0.001
时间: 49574.160156 ms
train Classification result: 9.77% ok (used 49984 images)
时间: 3692.930908 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：3
learn rate：0.001
时间: 49683.609375 ms
train Classification result: 10.32% ok (used 49984 images)
时间: 3719.961914 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：4
learn rate：0.001
时间: 49716.093750 ms
train Classification result: 25.26% ok (used 49984 images)
时间: 3740.791016 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：5
learn rate：0.001
时间: 49806.023438 ms
train Classification result: 53.84% ok (used 49984 images)
时间: 3722.489014 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：6
learn rate：0.001
时间: 49811.125000 ms
train Classification result: 69.04% ok (used 49984 images)
时间: 3698.967041 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：7
learn rate：0.001
时间: 49900.593750 ms
train Classification result: 76.02% ok (used 49984 images)
时间: 3724.368896 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：8
learn rate：0.001
时间: 49891.566406 ms
train Classification result: 80.41% ok (used 49984 images)
时间: 3734.409912 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：9
learn rate：0.001
时间: 49846.656250 ms
train Classification result: 83.16% ok (used 49984 images)
时间: 3745.007080 ms
Test Classification result: 9.98% ok (used 9984 images)
轮次：10
learn rate：0.001
时间: 49817.023438 ms
train Classification result: 85.76% ok (used 49984 images)
时间: 3718.322998 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：11
learn rate：0.0001
时间: 49884.902344 ms
train Classification result: 91.39% ok (used 49984 images)
时间: 3779.871094 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：12
learn rate：0.0001
时间: 49688.574219 ms
train Classification result: 93.25% ok (used 49984 images)
时间: 3739.181885 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：13
learn rate：1e-05
时间: 49828.031250 ms
train Classification result: 94.41% ok (used 49984 images)
时间: 3808.123047 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：14
learn rate：1e-05
时间: 49817.921875 ms
train Classification result: 94.78% ok (used 49984 images)
时间: 3787.574951 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：15
learn rate：1e-06
时间: 49733.281250 ms
train Classification result: 94.62% ok (used 49984 images)
时间: 3791.906982 ms
Test Classification result: 85.72% ok (used 9984 images)
轮次：16
learn rate：1e-06
时间: 49838.285156 ms
train Classification result: 94.87% ok (used 49984 images)
时间: 3774.705078 ms
Test Classification result: 85.71% ok (used 9984 images)
轮次：17
learn rate：1e-07
时间: 49747.105469 ms
train Classification result: 94.72% ok (used 49984 images)
时间: 3749.082031 ms
Test Classification result: 85.74% ok (used 9984 images)
轮次：18
learn rate：1e-07
时间: 49777.609375 ms
train Classification result: 94.87% ok (used 49984 images)
时间: 3773.662109 ms
Test Classification result: 85.78% ok (used 9984 images)
近期突破还是挺大的！有点蒙圈！感觉成绩上升没什么空间了！
正好桔子成熟了，要收获，就这样，先放下！
也要学会偷的人生半日闲！
看了一部好电影：a star is born（一个明星的诞生2018）
看完才知道是ladygaga，以前只知道有这么个名字！
shallow（浅滩）和亚利桑那的天空（[Always Remember Us This Way）真是伟大的作品！
终于把作品和人画上等号了！
以前很喜欢一首歌加利福尼亚！现在又多了一个亚利桑那！
美国的地名原来也有极其热爱他们的人！