人工智能实战2019 第六次作业 续连元

项目	内容
这个作业属于哪个课程	https://edu.cnblogs.com/campus/buaa/BUAA-AI-2019
这个作业的要求在哪里	https://edu.cnblogs.com/campus/buaa/BUAA-AI-2019/homework/3091
我在这个课程的目标是	学习，了解并实践深度学习的实际工程应用
这个作业在哪个具体方面帮助我实现目标	感受参数对训练结果的影响，尝试调参
作业正文	如下

正文

一、提高模型准确率

发现适当增加神经网络节点数，同时增加epoch数，则可使准确率得到明显提升。而batch_size以及学习率的大小在一定范围内相对来说对结果影响较小。这里我选用9对神经节点数目（n1取[64,100,128]，n2取[16,32,50]），对每组给定的神经元数目，再以不同的学习率、epoch_size以及batch_size尝试。具体地，对于每组给定的神经元个数分别取LR=[0.07,0.11,0.15]，batch_size=[10,30,64]，m_epoch=[10,20,30]共27组参数。找到所有使准确率大于0.97的组合。

二、运行的结果

n1= 64 n2= 16
LR: 0.07 m_epoch: 10 batch_size: 10 rate: 0.9722
LR: 0.07 m_epoch: 10 batch_size: 30 rate: 0.9677(unqualified)
LR: 0.07 m_epoch: 10 batch_size: 64 rate: 0.9541(unqualified)
LR: 0.11 m_epoch: 10 batch_size: 10 rate: 0.9757
LR: 0.11 m_epoch: 10 batch_size: 30 rate: 0.9736
LR: 0.11 m_epoch: 10 batch_size: 64 rate: 0.9635(unqualified)
LR: 0.15 m_epoch: 10 batch_size: 10 rate: 0.9732
LR: 0.15 m_epoch: 10 batch_size: 30 rate: 0.974
LR: 0.15 m_epoch: 10 batch_size: 64 rate: 0.9665(unqualified)
LR: 0.07 m_epoch: 20 batch_size: 10 rate: 0.9767
LR: 0.07 m_epoch: 20 batch_size: 30 rate: 0.9766
LR: 0.07 m_epoch: 20 batch_size: 64 rate: 0.9693(unqualified)
LR: 0.11 m_epoch: 20 batch_size: 10 rate: 0.975
LR: 0.11 m_epoch: 20 batch_size: 30 rate: 0.9754
LR: 0.11 m_epoch: 20 batch_size: 64 rate: 0.9711
LR: 0.15 m_epoch: 20 batch_size: 10 rate: 0.9696(unqualified)
LR: 0.15 m_epoch: 20 batch_size: 30 rate: 0.9766
LR: 0.15 m_epoch: 20 batch_size: 64 rate: 0.9737
LR: 0.07 m_epoch: 30 batch_size: 10 rate: 0.977
LR: 0.07 m_epoch: 30 batch_size: 30 rate: 0.9749
LR: 0.07 m_epoch: 30 batch_size: 64 rate: 0.9706
LR: 0.11 m_epoch: 30 batch_size: 10 rate: 0.9756
LR: 0.11 m_epoch: 30 batch_size: 30 rate: 0.9766
LR: 0.11 m_epoch: 30 batch_size: 64 rate: 0.9733
LR: 0.15 m_epoch: 30 batch_size: 10 rate: 0.9745
LR: 0.15 m_epoch: 30 batch_size: 30 rate: 0.9767
LR: 0.15 m_epoch: 30 batch_size: 64 rate: 0.974
part finished
n1= 64 n2= 32
LR: 0.07 m_epoch: 10 batch_size: 10 rate: 0.9754
LR: 0.07 m_epoch: 10 batch_size: 30 rate: 0.9682(unqualified)
LR: 0.07 m_epoch: 10 batch_size: 64 rate: 0.9534(unqualified)
LR: 0.11 m_epoch: 10 batch_size: 10 rate: 0.974
LR: 0.11 m_epoch: 10 batch_size: 30 rate: 0.9713
LR: 0.11 m_epoch: 10 batch_size: 64 rate: 0.9595(unqualified)
LR: 0.15 m_epoch: 10 batch_size: 10 rate: 0.9746
LR: 0.15 m_epoch: 10 batch_size: 30 rate: 0.9748
LR: 0.15 m_epoch: 10 batch_size: 64 rate: 0.9658(unqualified)
LR: 0.07 m_epoch: 20 batch_size: 10 rate: 0.975
LR: 0.07 m_epoch: 20 batch_size: 30 rate: 0.9753
LR: 0.07 m_epoch: 20 batch_size: 64 rate: 0.9694(unqualified)
LR: 0.11 m_epoch: 20 batch_size: 10 rate: 0.9794
LR: 0.11 m_epoch: 20 batch_size: 30 rate: 0.9769
LR: 0.11 m_epoch: 20 batch_size: 64 rate: 0.972
LR: 0.15 m_epoch: 20 batch_size: 10 rate: 0.9766
LR: 0.15 m_epoch: 20 batch_size: 30 rate: 0.9759
LR: 0.15 m_epoch: 20 batch_size: 64 rate: 0.9726
LR: 0.07 m_epoch: 30 batch_size: 10 rate: 0.9769
LR: 0.07 m_epoch: 30 batch_size: 30 rate: 0.9757
LR: 0.07 m_epoch: 30 batch_size: 64 rate: 0.9736
LR: 0.11 m_epoch: 30 batch_size: 10 rate: 0.9782
LR: 0.11 m_epoch: 30 batch_size: 30 rate: 0.9762
LR: 0.11 m_epoch: 30 batch_size: 64 rate: 0.9749
LR: 0.15 m_epoch: 30 batch_size: 10 rate: 0.9785
LR: 0.15 m_epoch: 30 batch_size: 30 rate: 0.9778
LR: 0.15 m_epoch: 30 batch_size: 64 rate: 0.9768
part finished
n1= 64 n2= 50
LR: 0.07 m_epoch: 10 batch_size: 10 rate: 0.9755
LR: 0.07 m_epoch: 10 batch_size: 30 rate: 0.9672(unqualified)
LR: 0.07 m_epoch: 10 batch_size: 64 rate: 0.9563(unqualified)
LR: 0.11 m_epoch: 10 batch_size: 10 rate: 0.9786
LR: 0.11 m_epoch: 10 batch_size: 30 rate: 0.9743
LR: 0.11 m_epoch: 10 batch_size: 64 rate: 0.9648(unqualified)
LR: 0.15 m_epoch: 10 batch_size: 10 rate: 0.9766
LR: 0.15 m_epoch: 10 batch_size: 30 rate: 0.9761
LR: 0.15 m_epoch: 10 batch_size: 64 rate: 0.9688(unqualified)
LR: 0.07 m_epoch: 20 batch_size: 10 rate: 0.9759
LR: 0.07 m_epoch: 20 batch_size: 30 rate: 0.9784
LR: 0.07 m_epoch: 20 batch_size: 64 rate: 0.9685(unqualified)
LR: 0.11 m_epoch: 20 batch_size: 10 rate: 0.9781
LR: 0.11 m_epoch: 20 batch_size: 30 rate: 0.9776
LR: 0.11 m_epoch: 20 batch_size: 64 rate: 0.9719
LR: 0.15 m_epoch: 20 batch_size: 10 rate: 0.9802
LR: 0.15 m_epoch: 20 batch_size: 30 rate: 0.9765
LR: 0.15 m_epoch: 20 batch_size: 64 rate: 0.9758
LR: 0.07 m_epoch: 30 batch_size: 10 rate: 0.9788
LR: 0.07 m_epoch: 30 batch_size: 30 rate: 0.9762
LR: 0.07 m_epoch: 30 batch_size: 64 rate: 0.9735
LR: 0.11 m_epoch: 30 batch_size: 10 rate: 0.9787
LR: 0.11 m_epoch: 30 batch_size: 30 rate: 0.9766
LR: 0.11 m_epoch: 30 batch_size: 64 rate: 0.9754
LR: 0.15 m_epoch: 30 batch_size: 10 rate: 0.9797
LR: 0.15 m_epoch: 30 batch_size: 30 rate: 0.9788
LR: 0.15 m_epoch: 30 batch_size: 64 rate: 0.9779
part finished
n1= 100 n2= 16
LR: 0.07 m_epoch: 10 batch_size: 10 rate: 0.9722
LR: 0.07 m_epoch: 10 batch_size: 30 rate: 0.9667(unqualified)
LR: 0.07 m_epoch: 10 batch_size: 64 rate: 0.956(unqualified)
LR: 0.11 m_epoch: 10 batch_size: 10 rate: 0.9769
LR: 0.11 m_epoch: 10 batch_size: 30 rate: 0.9709
LR: 0.11 m_epoch: 10 batch_size: 64 rate: 0.964(unqualified)
LR: 0.15 m_epoch: 10 batch_size: 10 rate: 0.9742
LR: 0.15 m_epoch: 10 batch_size: 30 rate: 0.9751
LR: 0.15 m_epoch: 10 batch_size: 64 rate: 0.9678(unqualified)
LR: 0.07 m_epoch: 20 batch_size: 10 rate: 0.9782
LR: 0.07 m_epoch: 20 batch_size: 30 rate: 0.9723
LR: 0.07 m_epoch: 20 batch_size: 64 rate: 0.9682(unqualified)
LR: 0.11 m_epoch: 20 batch_size: 10 rate: 0.9781
LR: 0.11 m_epoch: 20 batch_size: 30 rate: 0.9771
LR: 0.11 m_epoch: 20 batch_size: 64 rate: 0.9698(unqualified)
LR: 0.15 m_epoch: 20 batch_size: 10 rate: 0.9797
LR: 0.15 m_epoch: 20 batch_size: 30 rate: 0.9771
LR: 0.15 m_epoch: 20 batch_size: 64 rate: 0.9741
LR: 0.07 m_epoch: 30 batch_size: 10 rate: 0.9802
LR: 0.07 m_epoch: 30 batch_size: 30 rate: 0.9751
LR: 0.07 m_epoch: 30 batch_size: 64 rate: 0.9704
LR: 0.11 m_epoch: 30 batch_size: 10 rate: 0.9805
LR: 0.11 m_epoch: 30 batch_size: 30 rate: 0.9767
LR: 0.11 m_epoch: 30 batch_size: 64 rate: 0.9749
LR: 0.15 m_epoch: 30 batch_size: 10 rate: 0.9803
LR: 0.15 m_epoch: 30 batch_size: 30 rate: 0.9777
LR: 0.15 m_epoch: 30 batch_size: 64 rate: 0.9754
part finished
n1= 100 n2= 32
LR: 0.07 m_epoch: 10 batch_size: 10 rate: 0.9774
LR: 0.07 m_epoch: 10 batch_size: 30 rate: 0.9671(unqualified)
LR: 0.07 m_epoch: 10 batch_size: 64 rate: 0.9548(unqualified)
LR: 0.11 m_epoch: 10 batch_size: 10 rate: 0.9788
LR: 0.11 m_epoch: 10 batch_size: 30 rate: 0.9732
LR: 0.11 m_epoch: 10 batch_size: 64 rate: 0.9673(unqualified)
LR: 0.15 m_epoch: 10 batch_size: 10 rate: 0.975
LR: 0.15 m_epoch: 10 batch_size: 30 rate: 0.9758
LR: 0.15 m_epoch: 10 batch_size: 64 rate: 0.9687(unqualified)
LR: 0.07 m_epoch: 20 batch_size: 10 rate: 0.9791
LR: 0.07 m_epoch: 20 batch_size: 30 rate: 0.9766
LR: 0.07 m_epoch: 20 batch_size: 64 rate: 0.969(unqualified)
LR: 0.11 m_epoch: 20 batch_size: 10 rate: 0.9806
LR: 0.11 m_epoch: 20 batch_size: 30 rate: 0.9768
LR: 0.11 m_epoch: 20 batch_size: 64 rate: 0.971
LR: 0.15 m_epoch: 20 batch_size: 10 rate: 0.9818
LR: 0.15 m_epoch: 20 batch_size: 30 rate: 0.9805
LR: 0.15 m_epoch: 20 batch_size: 64 rate: 0.9735
LR: 0.07 m_epoch: 30 batch_size: 10 rate: 0.9807
LR: 0.07 m_epoch: 30 batch_size: 30 rate: 0.9763
LR: 0.07 m_epoch: 30 batch_size: 64 rate: 0.9749
LR: 0.11 m_epoch: 30 batch_size: 10 rate: 0.9803
LR: 0.11 m_epoch: 30 batch_size: 30 rate: 0.9805
LR: 0.11 m_epoch: 30 batch_size: 64 rate: 0.9756
LR: 0.15 m_epoch: 30 batch_size: 10 rate: 0.9798
LR: 0.15 m_epoch: 30 batch_size: 30 rate: 0.9789
LR: 0.15 m_epoch: 30 batch_size: 64 rate: 0.9771
part finished
n1= 100 n2= 50
LR: 0.07 m_epoch: 10 batch_size: 10 rate: 0.9757
LR: 0.07 m_epoch: 10 batch_size: 30 rate: 0.968(unqualified)
LR: 0.07 m_epoch: 10 batch_size: 64 rate: 0.9543(unqualified)
LR: 0.11 m_epoch: 10 batch_size: 10 rate: 0.9761
LR: 0.11 m_epoch: 10 batch_size: 30 rate: 0.9742
LR: 0.11 m_epoch: 10 batch_size: 64 rate: 0.9648(unqualified)
LR: 0.15 m_epoch: 10 batch_size: 10 rate: 0.9762
LR: 0.15 m_epoch: 10 batch_size: 30 rate: 0.9756
LR: 0.15 m_epoch: 10 batch_size: 64 rate: 0.9695(unqualified)
LR: 0.07 m_epoch: 20 batch_size: 10 rate: 0.98
LR: 0.07 m_epoch: 20 batch_size: 30 rate: 0.9771
LR: 0.07 m_epoch: 20 batch_size: 64 rate: 0.9691(unqualified)
LR: 0.11 m_epoch: 20 batch_size: 10 rate: 0.9813
LR: 0.11 m_epoch: 20 batch_size: 30 rate: 0.979
LR: 0.11 m_epoch: 20 batch_size: 64 rate: 0.9739
LR: 0.15 m_epoch: 20 batch_size: 10 rate: 0.9796
LR: 0.15 m_epoch: 20 batch_size: 30 rate: 0.9786
LR: 0.15 m_epoch: 20 batch_size: 64 rate: 0.9769
LR: 0.07 m_epoch: 30 batch_size: 10 rate: 0.9797
LR: 0.07 m_epoch: 30 batch_size: 30 rate: 0.9798
LR: 0.07 m_epoch: 30 batch_size: 64 rate: 0.9744
LR: 0.11 m_epoch: 30 batch_size: 10 rate: 0.9805
LR: 0.11 m_epoch: 30 batch_size: 30 rate: 0.9801
LR: 0.11 m_epoch: 30 batch_size: 64 rate: 0.9786
LR: 0.15 m_epoch: 30 batch_size: 10 rate: 0.9811
LR: 0.15 m_epoch: 30 batch_size: 30 rate: 0.981
LR: 0.15 m_epoch: 30 batch_size: 64 rate: 0.9759
part finished
n1= 128 n2= 16
LR: 0.07 m_epoch: 10 batch_size: 10 rate: 0.9747
LR: 0.07 m_epoch: 10 batch_size: 30 rate: 0.9667(unqualified)
LR: 0.07 m_epoch: 10 batch_size: 64 rate: 0.952(unqualified)
LR: 0.11 m_epoch: 10 batch_size: 10 rate: 0.978
LR: 0.11 m_epoch: 10 batch_size: 30 rate: 0.973
LR: 0.11 m_epoch: 10 batch_size: 64 rate: 0.9608(unqualified)
LR: 0.15 m_epoch: 10 batch_size: 10 rate: 0.9743
LR: 0.15 m_epoch: 10 batch_size: 30 rate: 0.9738
LR: 0.15 m_epoch: 10 batch_size: 64 rate: 0.9674(unqualified)
LR: 0.07 m_epoch: 20 batch_size: 10 rate: 0.9806
LR: 0.07 m_epoch: 20 batch_size: 30 rate: 0.9735
LR: 0.07 m_epoch: 20 batch_size: 64 rate: 0.9662(unqualified)
LR: 0.11 m_epoch: 20 batch_size: 10 rate: 0.9799
LR: 0.11 m_epoch: 20 batch_size: 30 rate: 0.9785
LR: 0.11 m_epoch: 20 batch_size: 64 rate: 0.9727
LR: 0.15 m_epoch: 20 batch_size: 10 rate: 0.9797
LR: 0.15 m_epoch: 20 batch_size: 30 rate: 0.9777
LR: 0.15 m_epoch: 20 batch_size: 64 rate: 0.978
LR: 0.07 m_epoch: 30 batch_size: 10 rate: 0.9786
LR: 0.07 m_epoch: 30 batch_size: 30 rate: 0.9758
LR: 0.07 m_epoch: 30 batch_size: 64 rate: 0.9744
LR: 0.11 m_epoch: 30 batch_size: 10 rate: 0.9808
LR: 0.11 m_epoch: 30 batch_size: 30 rate: 0.9791
LR: 0.11 m_epoch: 30 batch_size: 64 rate: 0.9762
LR: 0.15 m_epoch: 30 batch_size: 10 rate: 0.9822
LR: 0.15 m_epoch: 30 batch_size: 30 rate: 0.9816
LR: 0.15 m_epoch: 30 batch_size: 64 rate: 0.9767
part finished
n1= 128 n2= 32
LR: 0.07 m_epoch: 10 batch_size: 10 rate: 0.9769
LR: 0.07 m_epoch: 10 batch_size: 30 rate: 0.9675(unqualified)
LR: 0.07 m_epoch: 10 batch_size: 64 rate: 0.9542(unqualified)
LR: 0.11 m_epoch: 10 batch_size: 10 rate: 0.978
LR: 0.11 m_epoch: 10 batch_size: 30 rate: 0.9715
LR: 0.11 m_epoch: 10 batch_size: 64 rate: 0.9582(unqualified)
LR: 0.15 m_epoch: 10 batch_size: 10 rate: 0.9778
LR: 0.15 m_epoch: 10 batch_size: 30 rate: 0.976
LR: 0.15 m_epoch: 10 batch_size: 64 rate: 0.9666(unqualified)
LR: 0.07 m_epoch: 20 batch_size: 10 rate: 0.9794
LR: 0.07 m_epoch: 20 batch_size: 30 rate: 0.9753
LR: 0.07 m_epoch: 20 batch_size: 64 rate: 0.9677(unqualified)
LR: 0.11 m_epoch: 20 batch_size: 10 rate: 0.9809
LR: 0.11 m_epoch: 20 batch_size: 30 rate: 0.976
LR: 0.11 m_epoch: 20 batch_size: 64 rate: 0.9705
LR: 0.15 m_epoch: 20 batch_size: 10 rate: 0.9811
LR: 0.15 m_epoch: 20 batch_size: 30 rate: 0.9784
LR: 0.15 m_epoch: 20 batch_size: 64 rate: 0.9755
LR: 0.07 m_epoch: 30 batch_size: 10 rate: 0.9792
LR: 0.07 m_epoch: 30 batch_size: 30 rate: 0.9784
LR: 0.07 m_epoch: 30 batch_size: 64 rate: 0.9718
LR: 0.11 m_epoch: 30 batch_size: 10 rate: 0.9806
LR: 0.11 m_epoch: 30 batch_size: 30 rate: 0.9804
LR: 0.11 m_epoch: 30 batch_size: 64 rate: 0.975
LR: 0.15 m_epoch: 30 batch_size: 10 rate: 0.9801
LR: 0.15 m_epoch: 30 batch_size: 30 rate: 0.9796
LR: 0.15 m_epoch: 30 batch_size: 64 rate: 0.9777
part finished
n1= 128 n2= 50
LR: 0.07 m_epoch: 10 batch_size: 10 rate: 0.9754
LR: 0.07 m_epoch: 10 batch_size: 30 rate: 0.9689(unqualified)
LR: 0.07 m_epoch: 10 batch_size: 64 rate: 0.9521(unqualified)
LR: 0.11 m_epoch: 10 batch_size: 10 rate: 0.979
LR: 0.11 m_epoch: 10 batch_size: 30 rate: 0.9728
LR: 0.11 m_epoch: 10 batch_size: 64 rate: 0.9609(unqualified)
LR: 0.15 m_epoch: 10 batch_size: 10 rate: 0.9789
LR: 0.15 m_epoch: 10 batch_size: 30 rate: 0.9757
LR: 0.15 m_epoch: 10 batch_size: 64 rate: 0.9681(unqualified)
LR: 0.07 m_epoch: 20 batch_size: 10 rate: 0.9788
LR: 0.07 m_epoch: 20 batch_size: 30 rate: 0.975
LR: 0.07 m_epoch: 20 batch_size: 64 rate: 0.9655(unqualified)
LR: 0.11 m_epoch: 20 batch_size: 10 rate: 0.9811
LR: 0.11 m_epoch: 20 batch_size: 30 rate: 0.9774
LR: 0.11 m_epoch: 20 batch_size: 64 rate: 0.9715
LR: 0.15 m_epoch: 20 batch_size: 10 rate: 0.9805
LR: 0.15 m_epoch: 20 batch_size: 30 rate: 0.9794
LR: 0.15 m_epoch: 20 batch_size: 64 rate: 0.9757
LR: 0.07 m_epoch: 30 batch_size: 10 rate: 0.9798
LR: 0.07 m_epoch: 30 batch_size: 30 rate: 0.9776
LR: 0.07 m_epoch: 30 batch_size: 64 rate: 0.9728
LR: 0.11 m_epoch: 30 batch_size: 10 rate: 0.9803
LR: 0.11 m_epoch: 30 batch_size: 30 rate: 0.9791
LR: 0.11 m_epoch: 30 batch_size: 64 rate: 0.9755
LR: 0.15 m_epoch: 30 batch_size: 10 rate: 0.9815
LR: 0.15 m_epoch: 30 batch_size: 30 rate: 0.9803
LR: 0.15 m_epoch: 30 batch_size: 64 rate: 0.9775
part finished

三、最终的准确度结果及loss下降曲线

可以看到，对于尝试的所有组合，n1=128，n2=16，LR=0.15，m_epoch:=30，batch_size=10时，获得到最大rate=0.9822，对应下降曲线如下：