Created
May 22, 2017 04:38
-
-
Save anxingle/4a8afa65f272d23aeef5cf39591124ef to your computer and use it in GitHub Desktop.
实验报告
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
# fcn全连接网络+RNN | |
1. 切片为单列 第一层全连接中neuron为1024,batch_size:3000(太大)。 170epoch错误率 0.27/0.31 | |
![enter description here][1] | |
第一层全连接neuron为512,batch_size:2500. 170epoch 错误率0.154/0.186;230epoch错误率0.154/0.186;280epoch错误率 0.063/0.078 | |
![enter description here][2] | |
2. 切片为2列 第一层全连接中neuron为1024,batch_size:2500。280epoch错误率0.093/0.127. | |
![enter description here][3] | |
第一层全连接中neuron为512,batch_size:2500. 260epoch错误率0.089/0.12. | |
![enter description here][4] | |
3. 切片为3列 第一层全连接中neuron为1024,batch_size:2500. | |
![enter description here][5] | |
第一层全连接中neuron为512,batch_size:2500. | |
![enter description here][6] | |
4. 切片为3列,步长为2列(按列读取数据)。 第一层全连接中neuron为1024.batch_size:2500(较难训练,470epoch时终止) | |
![enter description here][7] | |
第一层全连接中neuron为512.batch_size:2500(280epoch准确率为0) | |
![enter description here][8] | |
5. 切片为3列,步长为2列(按行读取数据)。第一层全连接中neuron为1024.(230epoch是准确率为0) | |
![enter description here][9] | |
第一层全连接中neuron为512.batch_size:2800 (210epoch错误率0.085/0.165) | |
![enter description here][10] | |
# conv卷积+RNN | |
1. 切片为5列,步长为4列。 batch_size:2700 | |
![enter description here][11] | |
2. 切片为5列,步长为3列。 batch_size:2700 | |
![enter description here][12] | |
3. 4. no pooling and result is not very good! | |
[1]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/1/1.jpg | |
[2]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/1/2.jpg | |
[3]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/2/1.jpg | |
[4]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/2/2.jpg | |
[5]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/3/1.jpg | |
[6]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/3/2.jpg | |
[7]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/4/1.jpg | |
[8]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/4/2.jpg | |
[9]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/5/1.jpg | |
[10]:https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/fcn/5/2.jpg | |
[11]: https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/conv/1/1.jpg | |
[12]: https://raw.githubusercontent.com/anxingle/anxingle.github.io/master/public/img/experiment_report/conv/2/2.jpg |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment