why two different mini-batch Accuracy in CNN
显示 更早的评论
I am trying train a CNN.GPU device is Nvidia 1050.
My code
train_data_total=img;
label_4=YTrain;
layers_first = [imageInputLayer([32 32 3],'Normalization','none');
convolution2dLayer(5,130);
reluLayer();
maxPooling2dLayer(2,'Stride',2);
convolution2dLayer(5,180);
reluLayer();
maxPooling2dLayer(2,'Stride',2);
fullyConnectedLayer(256);
reluLayer();
fullyConnectedLayer(2);
softmaxLayer();
classificationLayer()];
opts_first = trainingOptions('sgdm','MiniBatchSize',256,'MaxEpochs',70 ...
,'InitialLearnRate',0.01,'Momentum',0,'Shuffle','once');
train_data_total=imresize(train_data_total,[32 32]);
net_first = trainNetwork(train_data_total,label_4,layers_first,opts_first);
YTrain_output1=classify(net_first,train_data_total);
train_accuracy1 = sum(YTrain_output1 == label_4)/numel(label_4)
My question is why Mini-batch Accuracy is around 50%.
And another computer using the same code and same input has Mini-batch Accuracy is around 98%.
Anyone has an idea of this
4 个评论
Arthur Chien
2017-5-2
Joss Knight
2017-5-2
Are you using MATLAB R2016b?
Arthur Chien
2017-5-3
Joss Knight
2017-5-13
回答（1 个）
Joss Knight
2017-5-16
0 个投票
3 个评论
Arthur Chien
2017-5-17
Joss Knight
2017-5-17
Please accept the answer.
shefali saxena
2019-1-19
hello sir
i am using Matlab R2017b
I am facing the same problem when traing CNN for ECG signals
My Mini-batch Accuracy is around 50%. and Mini Batch loss is Fixed at 0.69xx.
how can i resolve this problem ???
Training on single CPU.
|=======================================================================|
| Epoch | Iteration | Time Elapsed | Mini-batch | Mini-batch | Base Learning|
| | | (seconds) | Loss | Accuracy | Rate |
|=======================================================================|
| 1 | 1 | 0.72 | 0.6930 | 70.00% | 0.0010 |
| 4 | 320 | 11.36 | 0.6931 | 50.00% | 0.0010 |
| 8 | 640 | 21.76 | 0.6929 | 70.00% | 0.0010 |
| 12 | 960 | 32.12 | 0.6937 | 40.00% | 0.0010 |
| 16 | 1280 | 42.50 | 0.6932 | 50.00% | 0.0010 |
| 20 | 1600 | 53.04 | 0.6932 | 50.00% | 0.0010 |
| 23 | 1920 | 63.60 | 0.6930 | 50.00% | 0.0010 |
| 27 | 2240 | 74.73 | 0.6929 | 70.00% | 0.0010 |
| 31 | 2560 | 85.56 | 0.6932 | 50.00% | 0.0010 |
| 35 | 2880 | 96.81 | 0.6929 | 80.00% | 0.0010 |
| 39 | 3200 | 107.51 | 0.6930 | 60.00% | 0.0010 |
| 42 | 3520 | 118.29 | 0.6938 | 40.00% | 0.0010 |
| 46 | 3840 | 129.85 | 0.6933 | 30.00% | 0.0010 |
| 50 | 4160 | 140.92 | 0.6946 | 30.00% | 0.0010 |
| 54 | 4480 | 151.81 | 0.6928 | 60.00% | 0.0010 |
| 58 | 4800 | 163.14 | 0.6936 | 30.00% | 0.0010 |
| 61 | 5120 | 174.09 | 0.6932 | 50.00% | 0.0010 |
| 65 | 5440 | 184.38 | 0.6933 | 40.00% | 0.0010 |
| 69 | 5760 | 194.80 | 0.6928 | 60.00% | 0.0010 |
| 73 | 6080 | 205.18 | 0.6938 | 40.00% | 0.0010 |
| 77 | 6400 | 215.87 | 0.6931 | 60.00% | 0.0010 |
| 80 | 6720 | 227.45 | 0.6934 | 30.00% | 0.0010 |
| 84 | 7040 | 239.33 | 0.6932 | 50.00% | 0.0010 |
| 88 | 7360 | 250.64 | 0.6930 | 70.00% | 0.0010 |
| 92 | 7680 | 261.30 | 0.6931 | 50.00% | 0.0010 |
| 96 | 8000 | 271.53 | 0.6931 | 60.00% | 0.0010 |
| 100 | 8320 | 282.09 | 0.6935 | 40.00% | 0.0010 |
| 100 | 8400 | 284.69 | 0.6937 | 30.00% | 0.0010 |
|=======================================================================|
accuracy = 0.5000
please help !!!!
类别
在 帮助中心 和 File Exchange 中查找有关 Deep Learning Toolbox 的更多信息
2017-5-2
2019-1-19
Community Treasure Hunt
Find the treasures in MATLAB Central and discover how the community can help you!
Start Hunting!
Translated by 
