PaddlePaddle实现手写体数字识别模型

一 导入库及数据集：

 二 模型设计：

使用单层且没有非线性变换的模型，预测输入的图形数字值。其中，模型的输入为784维（28×28）数据，输出为1维数据。

 三 训练配置：

训练配置需要先生成模型实例（设为“训练”状态），再设置优化算法和学习率（使用随机梯度下降SGD，学习率设置为0.001）。

四 训练过程：

训练过程采用二层循环嵌套方式：

• 内层循环：负责整个数据集的一次遍历，遍历数据集采用分批次（batch）方式。
• 外层循环：定义遍历数据集的次数，本次训练中外层循环10次，通过参数EPOCH_NUM设置。

epoch_id: 0, batch_id: 0, loss is: [37.40072]
epoch_id: 0, batch_id: 1000, loss is: [6.511747]
epoch_id: 0, batch_id: 2000, loss is: [2.732864]
epoch_id: 0, batch_id: 3000, loss is: [3.4710536]
epoch_id: 1, batch_id: 0, loss is: [3.096552]
epoch_id: 1, batch_id: 1000, loss is: [2.7121267]
epoch_id: 1, batch_id: 2000, loss is: [3.9004908]
epoch_id: 1, batch_id: 3000, loss is: [5.8262606]
epoch_id: 2, batch_id: 0, loss is: [4.3189907]
epoch_id: 2, batch_id: 1000, loss is: [1.8050106]
epoch_id: 2, batch_id: 2000, loss is: [5.5903983]
epoch_id: 2, batch_id: 3000, loss is: [5.1830134]
epoch_id: 3, batch_id: 0, loss is: [2.2760863]
epoch_id: 3, batch_id: 1000, loss is: [5.3123417]
epoch_id: 3, batch_id: 2000, loss is: [3.207981]
epoch_id: 3, batch_id: 3000, loss is: [3.355655]
epoch_id: 4, batch_id: 0, loss is: [2.409387]
epoch_id: 4, batch_id: 1000, loss is: [4.8867397]
epoch_id: 4, batch_id: 2000, loss is: [1.474983]
epoch_id: 4, batch_id: 3000, loss is: [4.4998617]
epoch_id: 5, batch_id: 0, loss is: [2.2124193]
epoch_id: 5, batch_id: 1000, loss is: [5.57606]
epoch_id: 5, batch_id: 2000, loss is: [3.913291]
epoch_id: 5, batch_id: 3000, loss is: [2.5047076]
epoch_id: 6, batch_id: 0, loss is: [3.2275934]
epoch_id: 6, batch_id: 1000, loss is: [6.0397263]
epoch_id: 6, batch_id: 2000, loss is: [4.522304]
epoch_id: 6, batch_id: 3000, loss is: [5.8284745]
epoch_id: 7, batch_id: 0, loss is: [2.982819]
epoch_id: 7, batch_id: 1000, loss is: [2.6216135]
epoch_id: 7, batch_id: 2000, loss is: [6.423638]
epoch_id: 7, batch_id: 3000, loss is: [1.9560221]
epoch_id: 8, batch_id: 0, loss is: [2.6066465]
epoch_id: 8, batch_id: 1000, loss is: [4.659786]
epoch_id: 8, batch_id: 2000, loss is: [2.6956224]
epoch_id: 8, batch_id: 3000, loss is: [1.8081566]
epoch_id: 9, batch_id: 0, loss is: [1.3943908]
epoch_id: 9, batch_id: 1000, loss is: [6.2347713]
epoch_id: 9, batch_id: 2000, loss is: [3.3525836]
epoch_id: 9, batch_id: 3000, loss is: [2.8217497]

  完成代码后，自己写了一个28*28的手写数字进行了识别：