import paddle
import numpy as np
import matplotlib.pyplot as plt

import paddle.vision.transforms as T

# 数据的加载和预处理
transform = T.Normalize(mean=[127.5], std=[127.5])  #里面数值是根据数据集进行设置的
#像素值分布0-255组成图片，差值比较大会影响loss，影响性能，归一化到【-1，1】【0，1】梯度下降
#图像归一化处理，支持两种方式： 1. 用统一的均值和标准差值对图像的每个通道进行归一化处理； 2. 对每个通道指定不同的均值和标准差值进行归一化处理。
#如有问题可以参考官网API文档例程
# 训练数据集
train_dataset = paddle.vision.datasets.MNIST(mode='train', transform=transform)

# 评估数据集
eval_dataset = paddle.vision.datasets.MNIST(mode='test', transform=transform)

print('训练集样本量: {}，验证集样本量: {}'.format(len(train_dataset), len(eval_dataset)))

print('图片：')
print(type(train_dataset[0][0]))
print(train_dataset[0][0])
print('标签：')
print(type(train_dataset[0][1]))
print(train_dataset[0][1])

# 可视化展示
plt.figure() 
plt.imshow(train_dataset[0][0].reshape([28,28]), cmap=plt.cm.binary)  #单通道图片
plt.show()

#numpy ndarray，归一化后的图像。

# 模型网络结构搭建，线性全连接，1个隐藏层
network = paddle.nn.Sequential(
    paddle.nn.Flatten(),           # 拉平，将 (28, 28) => (784) 1维数组
    paddle.nn.Linear(784, 512),    # 隐层：线性变换层
    paddle.nn.ReLU(),              # 激活函数--保持梯度
    paddle.nn.Linear(512, 10)      # 输出层
)

# 模型封装
model = paddle.Model(network)

# 模型可视化 检验创建模型对错
model.summary((1, 28, 28))

Layer (type)       Input Shape          Output Shape         Param #    
===========================================================================
   Flatten-1       [[1, 28, 28]]           [1, 784]              0       
   Linear-1          [[1, 784]]            [1, 512]           401,920    
    ReLU-1           [[1, 512]]            [1, 512]              0       
   Linear-2          [[1, 512]]            [1, 10]             5,130     
===========================================================================
Total params: 407,050
Trainable params: 407,050
Non-trainable params: 0
---------------------------------------------------------------------------
Input size (MB): 0.00
Forward/backward pass size (MB): 0.01
Params size (MB): 1.55
Estimated Total Size (MB): 1.57
---------------------------------------------------------------------------

{'total_params': 407050, 'trainable_params': 407050}

# 配置优化器、损失函数、评估指标
model.prepare(paddle.optimizer.Adam(learning_rate=0.001, parameters=network.parameters()),
              paddle.nn.CrossEntropyLoss(),
              paddle.metric.Accuracy())
#softmax已经直接在调用函数中写进去了             
# 启动模型全流程训练
model.fit(train_dataset,  # 训练数据集
          eval_dataset,   # 评估数据集   ，同时判断训练的好坏是否存在过拟合欠拟合
          epochs=5,       # 训练的总轮次，所有数据集训练次数
          batch_size=64,  # 训练使用的分批大小
          verbose=1)      # 日志展示形式  1：每条显示  0：不显示

The loss value printed in the log is the current step, and the metric is the average value of previous step.
Epoch 1/5
step  20/938 [..............................] - loss: 0.6511 - acc: 0.6156 - ETA: 14s - 15ms/ste
/opt/conda/envs/python35-paddle120-env/lib/python3.7/site-packages/paddle/fluid/layers/utils.py:77: DeprecationWarning: Using or importing the ABCs from 'collections' instead of from 'collections.abc' is deprecated, and in 3.8 it will stop working
  return (isinstance(seq, collections.Sequence) and

step  30/938 [..............................] - loss: 0.4588 - acc: 0.6885 - ETA: 12s - 13ms/step

step  40/938 [>.............................] - loss: 0.5677 - acc: 0.7301 - ETA: 11s - 13ms/step
step 938/938 [==============================] - loss: 0.2502 - acc: 0.9135 - 12ms/step         
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 157/157 [==============================] - loss: 0.0397 - acc: 0.9527 - 8ms/step         
Eval samples: 10000
Epoch 2/5
step 938/938 [==============================] - loss: 0.0424 - acc: 0.9600 - 19ms/step         
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 157/157 [==============================] - loss: 0.0038 - acc: 0.9628 - 8ms/step         
Eval samples: 10000
Epoch 3/5
step 938/938 [==============================] - loss: 0.0244 - acc: 0.9687 - 19ms/step        
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 157/157 [==============================] - loss: 0.0068 - acc: 0.9687 - 8ms/step        
Eval samples: 10000
Epoch 4/5
step 938/938 [==============================] - loss: 0.0076 - acc: 0.9740 - 20ms/step         
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 157/157 [==============================] - loss: 0.0055 - acc: 0.9614 - 8ms/step         
Eval samples: 10000
Epoch 5/5
step 938/938 [==============================] - loss: 0.0781 - acc: 0.9770 - 19ms/step         
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 157/157 [==============================] - loss: 0.0024 - acc: 0.9756 - 8ms/step   

# 模型评估，根据prepare接口配置的loss和metric进行返回
result = model.evaluate(eval_dataset, verbose=1)

print(result)
#得到最后的精度指标

# 进行预测操作
result = model.predict(eval_dataset)

# 定义画图方法
def show_img(img, predict):
    plt.figure()
    plt.title('predict: {}'.format(predict))
    plt.imshow(img.reshape([28, 28]), cmap=plt.cm.binary)
    plt.show()

# 抽样展示
indexs = [2, 15, 38, 211]

for idx in indexs:
    show_img(eval_dataset[idx][0], np.argmax(result[0][idx]))

# 读取单张图片
image = eval_dataset[501][0]

# 单张图片预测
result = model.predict_batch([image])  #numpy形式数据
print(result)
print(np.argmax(result))
#是个数组，在这个数组里找最大值，最大值对应的下标就是预测值0--9
# 可视化结果
show_img(image, np.argmax(result))
#np.argmax找到下标

[array([[ -6.519223  , -10.6420555 ,  -1.3091288 ,   0.15322888,
         -4.734633  ,  -4.65111   , -16.480547  ,  -0.9590389 ,
         -4.847525  ,  11.187076  ]], dtype=float32)]

model.save('mnist')

from paddle.static import InputSpec


# 模型封装，为了后面保存预测模型，这里传入了inputs参数
model_2 = paddle.Model(network, inputs=[InputSpec(shape=[-1, 28, 28], dtype='float32', name='image')])
#预测模型部署的时候需要知道输入模型的形状；-1表示batchsize大小，-1表示灵活的 后续可以再设值

# 加载之前保存的阶段训练模型
model_2.load('mnist')

# 模型配置
model_2.prepare(paddle.optimizer.Adam(learning_rate=0.001, parameters=network.parameters()),
                paddle.nn.CrossEntropyLoss(),
                paddle.metric.Accuracy())
#CrossEntropyLoss分类，
# 模型全流程训练
model_2.fit(train_dataset, 
            eval_dataset,
            epochs=2,
            batch_size=64,
            verbose=1)
#继续调优训练epochs=2可以少一点了

The loss value printed in the log is the current step, and the metric is the average value of previous step.
Epoch 1/2
step 938/938 [==============================] - loss: 0.1485 - acc: 0.9785 - 10ms/step        
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 157/157 [==============================] - loss: 9.0207e-04 - acc: 0.9709 - 9ms/step     
Eval samples: 10000
Epoch 2/2
step 938/938 [==============================] - loss: 0.0094 - acc: 0.9820 - 13ms/step         
Eval begin...
The loss value printed in the log is the current batch, and the metric is the average value of previous step.
step 157/157 [==============================] - loss: 1.1729e-04 - acc: 0.9752 - 8ms/step       
Eval samples: 10000

# 保存用于后续推理部署的模型
model_2.save('infer/mnist', training=False)
#training=False保存部署模型了，不在调优！