# MNIST Test

**Repository Path**: edwinjiang703/MNIST-Test

## Basic Information

- **Project Name**: MNIST Test
- **Description**: No description available
- **Primary Language**: Unknown
- **License**: Not specified
- **Default Branch**: master
- **Homepage**: None
- **GVP Project**: No

## Statistics

- **Stars**: 0
- **Forks**: 0
- **Created**: 2018-06-26
- **Last Updated**: 2020-12-19

## Categories & Tags

**Categories**: Uncategorized

**Tags**: None

## README

# MNIST Test

#### Description
1 添加两个隐含层。一层有500个神经元，一层有300个神经元
2 使用学习率动态减少的测率来训练模型
3 使用双曲正切函数为激活函数
4 使用AdamOptimizer算法
在迭代到第三次时，ACC就超过了0.98

Iter 0, Testing Accuracy= 0.9795, Learning Rate= 0.001<br/>
Iter 1, Testing Accuracy= 0.9795, Learning Rate= 0.0009<br/>
Iter 2, Testing Accuracy= 0.9815, Learning Rate= 0.00081<br/>
Iter 3, Testing Accuracy= 0.982, Learning Rate= 0.000729<br/>
Iter 4, Testing Accuracy= 0.9825, Learning Rate= 0.0006561<br/>
Iter 5, Testing Accuracy= 0.9824, Learning Rate= 0.00059049<br/>
Iter 6, Testing Accuracy= 0.9828, Learning Rate= 0.000531441<br/>

改成Relu激活函数，测试结果如下：
在迭代到第9次时，ACC达到0.9854，而且收敛速度也很快

Iter 0, Testing Accuracy= 0.9768, Learning Rate= 0.001<br/>
Iter 1, Testing Accuracy= 0.9787, Learning Rate= 0.0009<br/>
Iter 2, Testing Accuracy= 0.9825, Learning Rate= 0.00081<br/>
Iter 3, Testing Accuracy= 0.9806, Learning Rate= 0.000729<br/>
Iter 4, Testing Accuracy= 0.9841, Learning Rate= 0.0006561<br/>
Iter 5, Testing Accuracy= 0.9842, Learning Rate= 0.00059049<br/>
Iter 6, Testing Accuracy= 0.9845, Learning Rate= 0.000531441<br/>
Iter 7, Testing Accuracy= 0.9849, Learning Rate= 0.000478297<br/>
Iter 8, Testing Accuracy= 0.9854, Learning Rate= 0.000430467<br/>
Iter 9, Testing Accuracy= 0.9853, Learning Rate= 0.00038742<br/>
Iter 10, Testing Accuracy= 0.9849, Learning Rate= 0.000348678<br/>
Iter 11, Testing Accuracy= 0.9854, Learning Rate= 0.000313811<br/>
Iter 12, Testing Accuracy= 0.9854, Learning Rate= 0.00028243<br/>


使用GradientDescentOptimizer算法，计算很慢，收敛速度比AdamOptimizer差很多，性能不理想。测试结果如下：
Iter 0, Testing Accuracy= 0.8802, Learning Rate= 0.001<br/>
Iter 1, Testing Accuracy= 0.9012, Learning Rate= 0.0009<br/>
Iter 2, Testing Accuracy= 0.9113, Learning Rate= 0.00081<br/>
Iter 3, Testing Accuracy= 0.9156, Learning Rate= 0.000729<br/>
Iter 4, Testing Accuracy= 0.9207, Learning Rate= 0.0006561<br/>
Iter 5, Testing Accuracy= 0.9232, Learning Rate= 0.00059049<br/>
Iter 6, Testing Accuracy= 0.9253, Learning Rate= 0.000531441<br/>