Tensorflow2.0基础-笔记- 利用tf.data修改mnist多分类任务

import tensorflow as tf
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt
%matplotlib inline

(train_image,train_label),(test_image,test_label)=tf.keras.datasets.fashion_mnist.load_data()

train_image=train_image/255
test_image=test_image/255 #归一化

ds_train_image=tf.data.Dataset.from_tensor_slices(train_image) #将数据经过tf.data模块进行切片
ds_train_label=tf.data.Dataset.from_tensor_slices(train_label) 

ds_train=tf.data.Dataset.zip((ds_train_image,ds_train_label))#将图片和标签进行整合在一个元组中

ds_train=ds_train.shuffle(10000).repeat().batch(64) #训练数据打乱，无限重复，确定batch的大小

ds_test_image=tf.data.Dataset.from_tensor_slices(train_image)
ds_test_label=tf.data.Dataset.from_tensor_slices(train_label)
ds_test=tf.data.Dataset.zip((ds_test_image,ds_test_label))
ds_test=ds_test.batch(64)                           #测试数据不需要打乱，只需要batch

model=tf.keras.Sequential()
model.add(tf.keras.layers.Flatten(input_shape=(28,28))) 输入的是28*28的图片
model.add(tf.keras.layers.Dense(128,activation='relu'))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(128,activation='relu'))
model.add(tf.keras.layers.Dropout(0.5))
model.add(tf.keras.layers.Dense(10,activation='softmax'))

model.compile(optimizer='adam',
              loss='sparse_categorical_crossentropy',
              metrics=['acc'])

#一个batch有64个数据，一个epoch就就需要原始数据量/64个batch。
steps_per_epochs=train_image.shape[0]//64
history=model.fit(
          ds_train,
          epochs=5,
          steps_per_epoch=steps_per_epochs,
          validation_data=ds_test)