Digit_Recognizer 流程： 1. 数据预处理 1.1 载入数据 1.2 check 数据 1.3 标准化 1.4 Reshape 1.5 Label encoding 1.6 Split training and valdiation set(划分训练集和测试集) 2. 搭建模型 2.1 Set the CNN model my CNN architechture is: [[Conv2D->relu]2 -> MaxPool2D -> Dropout]2 -> Flatten -> Dense -> Dropout -> Out 2.2 Set the optimizer and annealer(设置优化器等，使用 LR 衰减学习率) 3. Data augmentation 数据扩增 （使用数据扩增方法避免过拟合，常见的数据扩增方法有：灰度,水平翻转、垂直翻转,随机裁剪,反转,旋转,等等） 4. Evaluate the model 4.1 Plt training and validation curves 4.2 Confusion matrix (Confusion matrix can be very helpfull to see your model drawbacks) 4.3 可视化错误分类 5. Prediction and submition 5.1 Prediction and submition results