均值方差不稳，上不了86，稳，才能看到86点零几分，这种情况训练中很常见，这几天就是这个情况！
不过中秋这几天，也突破了真动量，所以均值方差稳不稳，都能训练cifar10测试成绩稳在86分以上！86.41分也出现了！真是过节出奇迹啊！
上一篇，已经说过卷积核类的更新函数，这一篇顺便也把线性层类的更新贴出来：网络再大再复杂，实际动量起作用就在con和linear的更新中，bn层也有动量，但cudnn已经考虑好了，这座山过年已经翻过来了！
void update(float lr) override {//线性层类的更新函数
//const float alpha = -lr;
int 替代 = out_features * in_features;
float beta = -lr; float momentum_ = 0.9f;
// v = momentum * v
cublasSscal(handle, 替代, &momentum_, 动量wwTest, 1);//动量不参与权重衰减
//cublasSscal(handleV, b_size, &momentum_, 动量vbTest, 1);
// v = v - lr * grad
cublasSaxpy(handle, 替代, &beta, grad_weight, 1, 动量wwTest, 1);
//cublasSaxpy(handleV, b_size, &beta, _grad_bias, 1, 动量vbTest, 1);
//开始引入权重衰减03170839
axpy_kernelV << <(替代 + 255) / 256, 256 >> > (替代, 0.0003f * batch, weight, 0, 1, grad_weight, 0, 1);
//sgd_update << <(替代 + 255) / 256, 256 >> > (weight, grad_weight, lr, 替代);
cublasSaxpy(handle, 替代, &beta, grad_weight, 1, weight, 1);
// param = param + v
float alpha = 1.0f;
cublasSaxpy(handle, 替代, &alpha, 动量wwTest, 1, weight, 1);
cublasSaxpy(handle, out_features, &beta, grad_bias, 1, bias, 1);
cudaDeviceSynchronize();//需要这句话吗？先试一试
cudaMemcpy(www, weight, sizeof(float) * in_features * out_features, cudaMemcpyDeviceToHost);//暂时未获取bias
}
好，我们看看这个86.41版本的诞生：（32秒一轮的快速版本，基板时稳在85.40分）
这个版本的残差块residualExt2：使用1*1降维，3*3升维！
如下：（传统残差块！relu这个实际是leaky relu）
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, _c, _c/2, _h, _w, 1, 1));
layers.emplace_back(std::make_shared<BN>(cudnn, batch, _c/2, _h, _w));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, _c/2, _h, _w));
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, _c/2, _c, _h, _w, 3, 1, 1));
layers.emplace_back(std::make_shared<BN>(cudnn, batch, _c, _h, _w));
整体架构如下：（residualExt2+relu构成一个完整残差块，relu实际是leaky relu）
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 5, 32, 32, 32, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 32, 32, 32));
layers.emplace_back(std::make_shared<ReLU>(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<residualExt2>(cudnn, batch, 64, 32, 32));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 64, 32, 32));
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 64, 128, 32, 32, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 128, 32, 32));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 128, 32, 32));
layers.emplace_back(std::make_shared<MaxPool2D>(cudnn, batch, 128, 32, 32, 2, 2, 0, 2));
layers.emplace_back(std::make_shared<Conv2D>(cudnn, batch, 128, 256, 16,16, 3, 1, 1));
layers.emplace_back(std::make_shared<residualExt2>(cudnn, batch, 256, 16, 16));
layers.emplace_back(std::make_shared<ReLU>(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<residualExt2>(cudnn, batch, 512, 8, 8));
layers.emplace_back(std::make_shared<ReLU>(cudnn, batch, 512, 8, 8));
layers.emplace_back(std::make_shared<AvgPool2D>(cudnn, batch, 512, 8, 8, 2, 2, 0, 2));
layers.emplace_back(std::make_shared<Dropout>(cudnn, batch, 512, 4,4));
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));
layers.emplace_back(std::make_shared<Linear>(cublas, batch, 384, 10));
lr更新如下：
int main() {
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;
lr = 0.0006f;
float bijiao = 0;
float* chengji;
chengji0 = 0;
int 起作用 = 0;
LeNetLeNet_net(cublas, cudnn, batch_size);//网络架构如上
for (int i = 0; i <100; i++)
{
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);
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)
{
lr = 0.000001;
if (起作用 >= 7)
{
lr = 0.0000001;
if (起作用 >= 9)//上84分
{
i = 100;
}
}
}
}
}
}
cublasDestroy(cublas);
cudnnDestroy(cudnn);
cudaStreamDestroy(stream);
system("pause");
return 0;
}
训练结果如下：
7:06 2026/9/27
lr=0.0005，跑出86.01
lr=0.0008，跑出86.09
lr=0.0006，跑出86.41
lr=0.0007，跑出86.35
轮次：0
learn rate：0.0007
时间: 30118.865234 ms
train Classification result: 9.79% ok (used 49984 images)
时间: 2475.320068 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：1
learn rate：0.0007
时间: 32678.519531 ms
train Classification result: 9.91% ok (used 49984 images)
时间: 2491.038086 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：2
learn rate：0.0007
时间: 32779.464844 ms
train Classification result: 10.02% ok (used 49984 images)
时间: 2506.808105 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：3
learn rate：0.0007
时间: 32914.789062 ms
train Classification result: 9.99% ok (used 49984 images)
时间: 2477.909912 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：4
learn rate：0.0007
时间: 32960.882812 ms
train Classification result: 9.87% ok (used 49984 images)
时间: 2473.050049 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：5
learn rate：0.0007
时间: 33182.941406 ms
train Classification result: 9.95% ok (used 49984 images)
时间: 2551.118896 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：6
learn rate：0.0007
时间: 34139.160156 ms
train Classification result: 9.99% ok (used 49984 images)
时间: 2459.322021 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：7
learn rate：0.0007
时间: 32983.617188 ms
train Classification result: 9.87% ok (used 49984 images)
时间: 2501.979004 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：8
learn rate：0.0007
时间: 33036.433594 ms
train Classification result: 10.10% ok (used 49984 images)
时间: 2493.275879 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：9
learn rate：0.0007
时间: 33009.109375 ms
train Classification result: 10.04% ok (used 49984 images)
时间: 2493.427002 ms
Test Classification result: 9.98% ok (used 9984 images)
轮次：10
learn rate：0.0007
时间: 33072.421875 ms
train Classification result: 9.95% ok (used 49984 images)
时间: 2431.406006 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：11
learn rate：0.0007
时间: 33075.484375 ms
train Classification result: 10.16% ok (used 49984 images)
时间: 2466.470947 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：12
learn rate：0.0007
时间: 33170.304688 ms
train Classification result: 9.95% ok (used 49984 images)
时间: 2494.767090 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：13
learn rate：0.0007
时间: 33049.863281 ms
train Classification result: 9.86% ok (used 49984 images)
时间: 2496.751953 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：14
learn rate：0.0007
时间: 33067.027344 ms
train Classification result: 10.18% ok (used 49984 images)
时间: 2486.579102 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：15
learn rate：0.0007
时间: 33105.812500 ms
train Classification result: 9.83% ok (used 49984 images)
时间: 2503.092041 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：16
learn rate：0.0007
时间: 33125.148438 ms
train Classification result: 9.96% ok (used 49984 images)
时间: 2502.031006 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：17
learn rate：0.0007
时间: 33191.933594 ms
train Classification result: 10.15% ok (used 49984 images)
时间: 2472.618896 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：18
learn rate：0.0007
时间: 33144.460938 ms
train Classification result: 10.06% ok (used 49984 images)
时间: 2476.176025 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：19
learn rate：0.0007
时间: 33125.914062 ms
train Classification result: 9.97% ok (used 49984 images)
时间: 2482.031006 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：20
learn rate：0.0007
时间: 33127.308594 ms
train Classification result: 10.33% ok (used 49984 images)
时间: 2507.875000 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：21
learn rate：0.0007
时间: 33139.003906 ms
train Classification result: 9.94% ok (used 49984 images)
时间: 2489.593994 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：22
learn rate：0.0007
时间: 33151.375000 ms
train Classification result: 10.07% ok (used 49984 images)
时间: 2558.768066 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：23
learn rate：0.0007
时间: 33098.851562 ms
train Classification result: 9.98% ok (used 49984 images)
时间: 2507.572021 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：24
learn rate：0.0007
时间: 33152.570312 ms
train Classification result: 9.92% ok (used 49984 images)
时间: 2481.826904 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：25
learn rate：0.0007
时间: 33137.902344 ms
train Classification result: 9.94% ok (used 49984 images)
时间: 2517.896973 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：26
learn rate：0.0007
时间: 33149.925781 ms
train Classification result: 10.03% ok (used 49984 images)
时间: 2491.683105 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：27
learn rate：0.0007
时间: 33144.742188 ms
train Classification result: 9.87% ok (used 49984 images)
时间: 2463.596924 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：28
learn rate：0.0007
时间: 33165.746094 ms
train Classification result: 9.74% ok (used 49984 images)
时间: 2465.920898 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：29
learn rate：0.0007
时间: 33222.480469 ms
train Classification result: 10.12% ok (used 49984 images)
时间: 2510.575928 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：30
learn rate：0.0007
时间: 33109.519531 ms
train Classification result: 9.99% ok (used 49984 images)
时间: 2468.898926 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：31
learn rate：0.0007
时间: 33159.582031 ms
train Classification result: 9.89% ok (used 49984 images)
时间: 2459.472900 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：32
learn rate：0.0007
时间: 33180.921875 ms
train Classification result: 9.85% ok (used 49984 images)
时间: 2480.693115 ms
Test Classification result: 9.98% ok (used 9984 images)
轮次：33
learn rate：0.0007
时间: 33107.066406 ms
train Classification result: 10.00% ok (used 49984 images)
时间: 2476.749023 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：34
learn rate：0.0007
时间: 33109.671875 ms
train Classification result: 9.87% ok (used 49984 images)
时间: 2474.357910 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：35
learn rate：0.0007
时间: 36370.738281 ms
train Classification result: 9.96% ok (used 49984 images)
时间: 2776.132080 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：36
learn rate：0.0007
时间: 34939.367188 ms
train Classification result: 9.99% ok (used 49984 images)
时间: 2482.914062 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：37
learn rate：0.0007
时间: 33674.019531 ms
train Classification result: 10.09% ok (used 49984 images)
时间: 2482.111084 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：38
learn rate：0.0007
时间: 33917.351562 ms
train Classification result: 10.00% ok (used 49984 images)
时间: 2593.073975 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：39
learn rate：0.0007
时间: 33877.144531 ms
train Classification result: 10.69% ok (used 49984 images)
时间: 2554.333984 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：40
learn rate：0.0007
时间: 34331.558594 ms
train Classification result: 11.65% ok (used 49984 images)
时间: 2553.225098 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：41
learn rate：0.0007
时间: 34152.843750 ms
train Classification result: 13.05% ok (used 49984 images)
时间: 2565.306885 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：42
learn rate：0.0007
时间: 34039.902344 ms
train Classification result: 17.03% ok (used 49984 images)
时间: 2782.355957 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：43
learn rate：0.0007
时间: 36334.937500 ms
train Classification result: 22.47% ok (used 49984 images)
时间: 2573.000977 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：44
learn rate：0.0007
时间: 34174.597656 ms
train Classification result: 32.06% ok (used 49984 images)
时间: 2547.205078 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：45
learn rate：0.0007
rb均值: -23.5637588501,rb方差:5.650913238525
rb均值: -22.2938308716,rb方差:11.421125411987
时间: 33776.687500 ms
train Classification result: 50.18% ok (used 49984 images)
时间: 2536.897949 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：46
learn rate：0.0007
rb均值: -29.8893795013,rb方差:14.865495681763
rb均值: -24.9163322449,rb方差:6.741480827332
rb均值: -25.7984523773,rb方差:7.273802280426
rb均值: -28.7765979767,rb方差:21.535430908203
时间: 34097.625000 ms
train Classification result: 60.17% ok (used 49984 images)
时间: 3067.528076 ms
Test Classification result: 9.99% ok (used 9984 images)
轮次：47
learn rate：0.0007
rb均值: -28.4401302338,rb方差:8.205205917358
rb均值: -36.3363189697,rb方差:25.126985549927
rb均值: -29.7907657623,rb方差:13.039494514465
rb均值: -31.8580303192,rb方差:15.465918540955
rb均值: -34.8623313904,rb方差:26.949035644531
时间: 33213.886719 ms
train Classification result: 67.00% ok (used 49984 images)
时间: 2534.572998 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：48
learn rate：0.0007
rb均值: -32.2388343811,rb方差:13.702013015747
rb均值: -42.9547309875,rb方差:31.358217239380
rb均值: -35.6422309875,rb方差:21.198385238647
rb均值: 28.2214260101,rb方差:6.130487918854
rb均值: -36.5089645386,rb方差:20.905700683594
rb均值: -41.4497108459,rb方差:31.305229187012
rb均值: 24.3464069366,rb方差:6.018348693848
时间: 33310.234375 ms
train Classification result: 71.76% ok (used 49984 images)
时间: 2506.730957 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：49
learn rate：0.0007
rb均值: -37.7567291260,rb方差:14.598229408264
rb均值: -48.2283325195,rb方差:37.702198028564
rb均值: -41.0778045654,rb方差:24.085765838623
rb均值: 32.3106040955,rb方差:5.997502326965
rb均值: -41.5045509338,rb方差:22.886947631836
rb均值: -47.2194519043,rb方差:33.096996307373
rb均值: 27.8427906036,rb方差:9.015285491943
时间: 33223.968750 ms
train Classification result: 74.90% ok (used 49984 images)
时间: 2529.448975 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：50
learn rate：0.0007
rb均值: -42.7248497009,rb方差:17.249694824219
rb均值: -54.4523315430,rb方差:43.226730346680
rb均值: -47.2404518127,rb方差:26.379734039307
rb均值: 35.8800811768,rb方差:9.286828041077
rb均值: -46.1789283752,rb方差:23.251369476318
rb均值: -53.7077178955,rb方差:38.246959686279
rb均值: -30.5931129456,rb方差:5.232138156891
rb均值: 31.6021823883,rb方差:12.065314292908
时间: 33146.964844 ms
train Classification result: 77.53% ok (used 49984 images)
时间: 2507.123047 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：51
learn rate：0.0007
rb均值: -47.8649101257,rb方差:24.584005355835
rb均值: -60.0479927063,rb方差:44.461090087891
rb均值: -52.4909400940,rb方差:36.528347015381
rb均值: 40.3048706055,rb方差:10.338654518127
rb均值: -51.3604545593,rb方差:30.102478027344
rb均值: -60.0524101257,rb方差:47.107074737549
rb均值: -34.1152725220,rb方差:6.761221885681
rb均值: 35.3935432434,rb方差:10.655485153198
时间: 33357.527344 ms
train Classification result: 79.97% ok (used 49984 images)
时间: 3025.229980 ms
Test Classification result: 9.98% ok (used 9984 images)
轮次：52
learn rate：0.0007
rb均值: -52.7560806274,rb方差:22.790546417236
rb均值: -65.6790771484,rb方差:55.569534301758
rb均值: -58.1712379456,rb方差:33.461746215820
rb均值: 44.1593132019,rb方差:13.123476028442
rb均值: -56.4192619324,rb方差:26.631986618042
rb均值: -35.0920143127,rb方差:5.854050159454
rb均值: -66.3105392456,rb方差:43.772258758545
rb均值: -38.1030998230,rb方差:7.054357528687
rb均值: 38.9384193420,rb方差:11.156447410583
rb均值: -27.0931320190,rb方差:6.215202808380
时间: 37038.695312 ms
train Classification result: 81.75% ok (used 49984 images)
时间: 2713.370117 ms
Test Classification result: 10.02% ok (used 9984 images)
轮次：53
learn rate：0.0007
rb均值: 1.0481296778,rb方差:5.114522933960
rb均值: -57.5020790100,rb方差:23.686534881592
rb均值: -71.2897109985,rb方差:62.737396240234
rb均值: -63.3181304932,rb方差:34.272014617920
rb均值: 48.7159843445,rb方差:12.991103172302
rb均值: -61.8474121094,rb方差:28.552257537842
rb均值: -38.3693237305,rb方差:7.650144100189
rb均值: -72.4577026367,rb方差:44.308525085449
rb均值: -41.5541725159,rb方差:7.610636234283
rb均值: 42.6009292603,rb方差:12.976784706116
rb均值: -30.2559452057,rb方差:7.719710826874
时间: 33649.710938 ms
train Classification result: 83.11% ok (used 49984 images)
时间: 2519.780029 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：54
learn rate：0.0007
rb均值: 1.2852791548,rb方差:6.000449180603
rb均值: -61.5036277771,rb方差:28.876888275146
rb均值: -75.9817199707,rb方差:67.766654968262
rb均值: -68.5399627686,rb方差:37.054290771484
rb均值: 37.1110229492,rb方差:5.030401706696
rb均值: 52.5443000793,rb方差:18.293304443359
rb均值: -67.0653762817,rb方差:30.164152145386
rb均值: -41.4650382996,rb方差:7.637140750885
rb均值: -78.1370773315,rb方差:42.767654418945
rb均值: -45.4713287354,rb方差:7.790681362152
rb均值: 46.7377319336,rb方差:13.765178680420
rb均值: -33.2645111084,rb方差:7.769547462463
时间: 33417.398438 ms
train Classification result: 84.66% ok (used 49984 images)
时间: 2712.822021 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：55
learn rate：0.0007
rb均值: 1.5597444773,rb方差:7.704373836517
rb均值: -66.0041809082,rb方差:30.983263015747
rb均值: -81.7714309692,rb方差:67.440917968750
rb均值: -73.5870666504,rb方差:38.964405059814
rb均值: 39.8702812195,rb方差:7.001320838928
rb均值: 56.2885818481,rb方差:19.306165695190
rb均值: -71.9961013794,rb方差:30.001926422119
rb均值: -44.5986480713,rb方差:9.201763153076
rb均值: -83.6486892700,rb方差:47.567600250244
rb均值: -49.0813255310,rb方差:9.562763214111
rb均值: 50.5838165283,rb方差:14.543864250183
rb均值: -35.2722625732,rb方差:10.680083274841
时间: 34138.246094 ms
train Classification result: 85.85% ok (used 49984 images)
时间: 2490.733887 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：56
learn rate：0.0007
rb均值: 1.3579657078,rb方差:5.501024723053
rb均值: 1.8324171305,rb方差:9.621965408325
rb均值: 37.4142913818,rb方差:5.278701782227
rb均值: -70.7144546509,rb方差:33.184204101562
rb均值: -86.8049926758,rb方差:76.156394958496
rb均值: -78.4885711670,rb方差:44.176235198975
rb均值: 42.9078025818,rb方差:9.030879974365
rb均值: 60.3489761353,rb方差:20.731372833252
rb均值: -77.1577072144,rb方差:35.844020843506
rb均值: -47.1757011414,rb方差:10.652269363403
rb均值: -89.9143905640,rb方差:53.244415283203
rb均值: -52.3006401062,rb方差:11.439117431641
rb均值: 54.5343818665,rb方差:13.625034332275
rb均值: -37.9267425537,rb方差:11.054612159729
时间: 33231.558594 ms
train Classification result: 87.46% ok (used 49984 images)
时间: 2514.360107 ms
Test Classification result: 10.01% ok (used 9984 images)
轮次：57
learn rate：0.0007
rb均值: 1.4437457323,rb方差:5.933919429779
rb均值: 1.5748572350,rb方差:5.364465713501
rb均值: 1.8954451084,rb方差:11.391463279724
rb均值: 40.2667884827,rb方差:5.490139007568
rb均值: -74.5405044556,rb方差:33.324504852295
rb均值: -91.5866622925,rb方差:76.999694824219
rb均值: -83.0806121826,rb方差:41.552520751953
rb均值: 45.8116188049,rb方差:8.020635604858
rb均值: 64.5955200195,rb方差:21.434436798096
rb均值: -81.4394836426,rb方差:33.126953125000
rb均值: -50.5877723694,rb方差:11.741532325745
rb均值: -95.6307983398,rb方差:56.307155609131
rb均值: -55.1624908447,rb方差:11.936563491821
rb均值: 57.8375968933,rb方差:13.392998695374
rb均值: -40.7312088013,rb方差:12.993888854980
时间: 33263.707031 ms
train Classification result: 88.29% ok (used 49984 images)
时间: 2531.896973 ms
Test Classification result: 10.00% ok (used 9984 images)
轮次：58
learn rate：0.0001
rb均值: 1.6656756401,rb方差:6.204547882080
rb均值: 1.5678222179,rb方差:5.786753654480
rb均值: 1.9366011620,rb方差:11.419604301453
rb均值: 40.4383621216,rb方差:5.942081928253
rb均值: -74.9249114990,rb方差:35.135211944580
rb均值: -92.4235610962,rb方差:66.489768981934
rb均值: -83.6810150146,rb方差:44.661430358887
rb均值: 45.8999977112,rb方差:8.092182159424
rb均值: 65.0669860840,rb方差:21.740514755249
rb均值: -81.8265075684,rb方差:32.648799896240
rb均值: -50.8589591980,rb方差:12.589338302612
rb均值: -96.2981491089,rb方差:60.390499114990
rb均值: -55.2549247742,rb方差:11.055593490601
rb均值: 58.4152069092,rb方差:14.747470855713
rb均值: -40.9571685791,rb方差:13.050477981567
时间: 33273.699219 ms
train Classification result: 92.04% ok (used 49984 images)
时间: 2522.981934 ms
Test Classification result: 85.87% ok (used 9984 images)
轮次：59
learn rate：0.0001
rb均值: 1.7110574245,rb方差:6.174195766449
rb均值: 1.5948240757,rb方差:5.873876571655
rb均值: 1.9767427444,rb方差:11.837670326233
rb均值: 40.7505493164,rb方差:5.719985961914
rb均值: -75.6612625122,rb方差:35.183692932129
rb均值: -93.1206741333,rb方差:64.738731384277
rb均值: -84.3220443726,rb方差:44.769012451172
rb均值: 46.2816276550,rb方差:8.029926300049
rb均值: 65.5764160156,rb方差:21.731914520264
rb均值: -82.5255889893,rb方差:31.480693817139
rb均值: -51.3099060059,rb方差:12.194223403931
rb均值: -97.0512695312,rb方差:59.829113006592
rb均值: -55.7225303650,rb方差:11.251307487488
rb均值: 58.9783134460,rb方差:14.961582183838
rb均值: -41.3694076538,rb方差:12.621570587158
时间: 33189.187500 ms
train Classification result: 93.77% ok (used 49984 images)
时间: 2529.014893 ms
Test Classification result: 85.99% ok (used 9984 images)
轮次：60
learn rate：1e-05
rb均值: 1.7222387791,rb方差:6.168077468872
rb均值: 1.5988073349,rb方差:5.886610031128
rb均值: 1.9727517366,rb方差:11.915425300598
rb均值: 40.7322349548,rb方差:5.566786289215
rb均值: -75.7362442017,rb方差:34.561126708984
rb均值: -93.0568542480,rb方差:62.616016387939
rb均值: -84.2969360352,rb方差:44.364295959473
rb均值: 46.3012084961,rb方差:7.834212303162
rb均值: 65.5969238281,rb方差:21.467796325684
rb均值: -82.6397476196,rb方差:31.320507049561
rb均值: -51.2847976685,rb方差:12.114958763123
rb均值: -97.1213455200,rb方差:58.062335968018
rb均值: -55.7121238708,rb方差:10.812603950500
rb均值: 58.9825019836,rb方差:14.499515533447
rb均值: -41.3320350647,rb方差:12.579503059387
时间: 33213.593750 ms
train Classification result: 94.75% ok (used 49984 images)
时间: 2524.956055 ms
Test Classification result: 86.22% ok (used 9984 images)
轮次：61
learn rate：1e-05
rb均值: 1.7327111959,rb方差:6.177073478699
rb均值: 1.5964027643,rb方差:5.890989780426
rb均值: 1.9689640999,rb方差:11.907920837402
rb均值: 40.7582664490,rb方差:5.554490089417
rb均值: -75.8127441406,rb方差:34.489757537842
rb均值: -93.1496276855,rb方差:62.252201080322
rb均值: -84.3826065063,rb方差:43.966594696045
rb均值: 46.3124389648,rb方差:7.857982635498
rb均值: 65.6307754517,rb方差:21.488002777100
rb均值: -82.6988143921,rb方差:31.365556716919
rb均值: -51.2910614014,rb方差:12.145358085632
rb均值: -97.1550903320,rb方差:57.838768005371
rb均值: -55.7345886230,rb方差:10.711233139038
rb均值: 58.9696044922,rb方差:14.318328857422
rb均值: -41.3247871399,rb方差:12.666104316711
时间: 33237.097656 ms
train Classification result: 95.05% ok (used 49984 images)
时间: 2485.888916 ms
Test Classification result: 86.28% ok (used 9984 images)
轮次：62
learn rate：1e-06
rb均值: 1.7336536646,rb方差:6.163188457489
rb均值: 1.5956763029,rb方差:5.871879100800
rb均值: 1.9707425833,rb方差:11.927271842957
rb均值: 40.7711906433,rb方差:5.547277450562
rb均值: -75.8296127319,rb方差:34.149002075195
rb均值: -93.1822662354,rb方差:61.892433166504
rb均值: -84.3783950806,rb方差:43.845024108887
rb均值: 46.3135261536,rb方差:7.865357875824
rb均值: 65.6442565918,rb方差:21.289556503296
rb均值: -82.6995544434,rb方差:31.106214523315
rb均值: -51.2963676453,rb方差:12.368973731995
rb均值: -97.1537399292,rb方差:57.663291931152
rb均值: -55.7356834412,rb方差:10.697630882263
rb均值: 58.9790077209,rb方差:14.427842140198
rb均值: -41.3399734497,rb方差:12.468928337097
时间: 33185.464844 ms
train Classification result: 95.05% ok (used 49984 images)
时间: 2522.812988 ms
Test Classification result: 86.28% ok (used 9984 images)
轮次：63
learn rate：1e-06
rb均值: 1.7337003946,rb方差:6.167066097260
rb均值: 1.5961546898,rb方差:5.876651287079
rb均值: 1.9705663919,rb方差:11.919945716858
rb均值: 40.7803230286,rb方差:5.459111213684
rb均值: -75.8379440308,rb方差:34.152576446533
rb均值: -93.1796340942,rb方差:61.740005493164
rb均值: -84.3992309570,rb方差:43.551094055176
rb均值: 46.3213272095,rb方差:7.701320171356
rb均值: 65.6606063843,rb方差:21.228170394897
rb均值: -82.7222290039,rb方差:31.019563674927
rb均值: -51.3066215515,rb方差:12.229062080383
rb均值: -97.1729354858,rb方差:57.718257904053
rb均值: -55.7397232056,rb方差:10.597153663635
rb均值: 58.9836883545,rb方差:14.290382385254
rb均值: -41.3313484192,rb方差:12.549730300903
时间: 33247.292969 ms
train Classification result: 95.04% ok (used 49984 images)
时间: 2498.124023 ms
Test Classification result: 86.32% ok (used 9984 images)
轮次：64
learn rate：1e-07
rb均值: 1.7342418432,rb方差:6.168128013611
rb均值: 1.5961583853,rb方差:5.875751495361
rb均值: 1.9709783792,rb方差:11.920775413513
rb均值: 40.7715530396,rb方差:5.548216819763
rb均值: -75.8261642456,rb方差:34.403861999512
rb均值: -93.1891021729,rb方差:61.707130432129
rb均值: -84.3868026733,rb方差:43.701858520508
rb均值: 46.3219795227,rb方差:7.793116092682
rb均值: 65.6526947021,rb方差:21.264612197876
rb均值: -82.7100524902,rb方差:31.118370056152
rb均值: -51.3152809143,rb方差:12.308557510376
rb均值: -97.1570129395,rb方差:57.783813476562
rb均值: -55.7350006104,rb方差:10.717393875122
rb均值: 58.9870567322,rb方差:14.390939712524
rb均值: -41.3360748291,rb方差:12.638366699219
时间: 33248.515625 ms
train Classification result: 95.13% ok (used 49984 images)
时间: 2469.784912 ms
Test Classification result: 86.31% ok (used 9984 images)
轮次：65
learn rate：1e-07
rb均值: 1.7344284058,rb方差:6.166381359100
rb均值: 1.5962694883,rb方差:5.875902175903
rb均值: 1.9706745148,rb方差:11.920172691345
rb均值: 40.7784957886,rb方差:5.523163318634
rb均值: -75.8445434570,rb方差:34.281490325928
rb均值: -93.1662139893,rb方差:61.540534973145
rb均值: -84.3910369873,rb方差:43.554218292236
rb均值: 46.3189201355,rb方差:7.765586376190
rb均值: 65.6554794312,rb方差:21.281215667725
rb均值: -82.7340698242,rb方差:31.167970657349
rb均值: -51.2973976135,rb方差:12.305869102478
rb均值: -97.1972122192,rb方差:57.287841796875
rb均值: -55.7529067993,rb方差:10.702785491943
rb均值: 58.9922714233,rb方差:14.244552612305
rb均值: -41.3435363770,rb方差:12.525722503662
时间: 33205.679688 ms
train Classification result: 95.14% ok (used 49984 images)
时间: 2489.801025 ms
Test Classification result: 86.35% ok (used 9984 images)
请按任意键继续. . .