CNN实战--mnist

CNN实战--mnist

dataprocessing

我一般把数据处理单独写一个函数

因为网上大多数都是直接在线下载做学习,导致与实际应用的情况不相符,所以我这是直接下载下来并读取,处理数据

这个数据类型文档说的很清楚

是图片二进制存储的(图片大小28*28),并且开头有一个magic num (需要跳过它)

不知道跳几位的可以多尝试一下不同的offset输出长度看是不是整除

具体数据处理可以看这个(虽然网上一搜就搜到了)

def read_data():

    with open('./t10k-labels.idx1-ubyte','rb') as f:
        y_test=np.frombuffer(f.read(),np.uint8,offset=8)
        y_test=tf.convert_to_tensor(y_test,tf.int32)
        # offset代表从第几个byte后面开始读取,0则是从头开始读 1byte=8bit
        # y_test=tf.one_hot(y_test,10)

    with open('./train-labels.idx1-ubyte','rb') as f:
        y_train=np.frombuffer(f.read(),np.uint8,offset=8)
        y_train=tf.convert_to_tensor(y_train,tf.int32)
        # 1*10000
        # y_train=tf.one_hot(y_train,10)

    with open('./t10k-images.idx3-ubyte', 'rb') as f:
        x_test = np.frombuffer(f.read(), np.uint8,offset=16).reshape(len(y_test), 28, 28,1)
        x_test=tf.convert_to_tensor(x_test,tf.float32)/255
    # #502098=28*28*60000

    with open('./train-images.idx3-ubyte', 'rb') as f:
        x_train = np.frombuffer(f.read(), np.uint8,offset=16).reshape(len(y_train),28,28,1)
        x_train=tf.convert_to_tensor(x_train,dtype=tf.float32)/255
    #78400=28*28*10000
    return x_train,y_train,x_test,y_test

train_model

#-*- coding:utf-8 -*-
# @Author : Dummerfu
# @Time : 2020/4/20 21:42
import tensorflow as tf
import data_processing
import numpy as np
import os

os.environ['TF_CPP_MIN_LOG_LEVEL'] = '2'

if __name__ == '__main__':

    # x: [60k, 28, 28,1], [10k, 28, 28,1]
    # y: [60k], [10k]
    x_train, y_train, x_test, y_test = data_processing.read_data()
    # print(y_test.shape,x_train.shape)
   

    model=tf.keras.models.Sequential([
        # 这里输入层还是要写单个输入的shape
        tf.keras.layers.Conv2D(input_shape=(28,28,1),filters=32,
                               kernel_size=(3,3),strides=(1,1),padding='SAME',activation='relu'),
        tf.keras.layers.MaxPool2D(pool_size=(2,2),strides=(2,2),padding='SAME'),
        
        tf.keras.layers.Conv2D(filters=64,kernel_size=(3,3),
                               strides=(1,1),padding='SAME',activation='relu'),
        tf.keras.layers.MaxPool2D(pool_size=(2,2),strides=(2,2),padding='SAME'),
        tf.keras.layers.Dropout(0.7),
        
        tf.keras.layers.Flatten(),
        
        # FC1
        tf.keras.layers.Dense(128,activation='relu'),
        tf.keras.layers.Dropout(0.5),
        # FC2|output
        tf.keras.layers.Dense(10,activation='softmax'),
    ])
    # 查看层的信息
    # print(model.summary())
	
    # 设置训练参数
    model.compile(optimizer='adam',loss='sparse_categorical_crossentropy',metrics=['accuracy'])
    
    # 训练(你甚至都不需要自己转onehot)
    # validation_split=x 将训练集*x变为测试集,进行预测
    # verbose=1 显示训练信息
    model.fit(x=x_train,y=y_train,batch_size=32,epochs=5,validation_split=0.3,verbose=1)
    train_loss,train_accu=model.evaluate(x=x_test,y=y_test)
    print(train_loss)
    print(train_accu)

这个才训练到 98.5%好垃圾

model save|restore

参考这个

有两种方式save

只保存weight和bias,不保存网络结构

这个知道就好了 其实是我懒得写,可以看那个链接里面写的

保存网络结构

import tensorflow as tf

# 这个model是前面的那个model类
model.save("path")
# model del
	# 这里的测试可以自己输入
    x_train,y_train,x_test,y_test=data_processing.read_data()
    restore_model= tf.keras.models.load_model('./my_model.ckpt')
    loss,acc=restore_model.evaluate(x_test,y_test)
    print(loss)
    print(acc)

predict

	# draw 当然自己随便写,预测数据还是得本地导入
    draw(x_test.numpy()[rad].reshape(28,28),y_test.numpy()[rad])
    restore_model= tf.keras.models.load_model('./my_model.ckpt')
    
    pro=np.argmax(restore_model.predict(x_test.numpy()[rad].reshape(1,28,28,1)))
    print('???',pro)
posted @ 2020-07-12 19:35  Sakura_Momoko  阅读(281)  评论(0编辑  收藏  举报