tensorflow2.3实现mnist数据分类

tensorflow2.3实现mnist数据分类(CNN)

MNIST数据集是机器学习领域中非常经典的一个数据集，由60000个训练样本和10000个测试样本组成，每个样本都是一张28 * 28像素的灰度手写数字图片。

我们用CNN对其进行深度学习分类。
导入包

import tensorflow as tf
import numpy as np
import pathlib

查看tensorflow的版本

print('Tensorflow version: {}'.format(tf.__version__))

Tensorflow version: 2.3

读取数据

(train_images, train_labels),(test_images, test_labels) = tf.keras.datasets.mnist.load_data()

数据进行归一化

train_images = train_images/255
test_images = test_images/255

利用tf.data模块构建dataset

ds_train_img = tf.data.Dataset.from_tensor_slices(train_images)
ds_train_img

<TensorSliceDataset shapes: (28, 28), types: tf.float64>

label利用tf.data模块构建dataset

ds_train_lab = tf.data.Dataset.from_tensor_slices(train_labels)
ds_train_lab

<TensorSliceDataset shapes: (), types: tf.uint8>

把image和label 以元组的形式zip到一起（也可以不用）

ds_train = tf.data.Dataset.zip((ds_train_img, ds_train_lab))

<ZipDataset shapes: ((28, 28), ()), types: (tf.float64, tf.uint8)>

构建训练集

ds_train = ds_train.shuffle(10000).repeat().batch(64)
ds_train

<BatchDataset shapes: ((None, 28, 28), (None,)), types: (tf.float64, tf.uint8)>

构建测试集

ds_test = tf.data.Dataset.from_tensor_slices((test_images, test_labels))
ds_test

<TensorSliceDataset shapes: ((28, 28), ()), types: (tf.float64, tf.uint8)>

测试集也要以bacht的形式输入

ds_test = ds_test.batch(64)

建立模型

model = tf.keras.Sequential()
model.add(tf.keras.layers.Flatten(input_shape=(28, 28)))
model.add(tf.keras.layers.Dense(128, activation='relu'))
model.add(tf.keras.layers.Dense(10,activation='softmax'))

模型编译

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

设置批次

steps_per_epochs=train_images.shape[0]//64

模型训练

model.fit(ds_train, epochs=5, steps_per_epoch=steps_per_epochs, validation_data=ds_test, validation_steps=10000//64)

Train for 937 steps, validate for 156 steps
Epoch 1/5
937/937 [==============================] - 5s 5ms/step - loss: 0.3036 - acc: 0.9154 - val_loss: 0.1700 - val_acc: 0.9510
Epoch 2/5
937/937 [==============================] - 4s 4ms/step - loss: 0.1381 - acc: 0.9600 - val_loss: 0.1229 - val_acc: 0.9638
Epoch 3/5
937/937 [==============================] - 4s 4ms/step - loss: 0.0952 - acc: 0.9725 - val_loss: 0.1010 - val_acc: 0.9690
Epoch 4/5
937/937 [==============================] - 4s 4ms/step - loss: 0.0730 - acc: 0.9777 - val_loss: 0.0909 - val_acc: 0.9714
Epoch 5/5
937/937 [==============================] - 4s 4ms/step - loss: 0.0573 - acc: 0.9826 - val_loss: 0.0781 - val_acc: 0.9760
<tensorflow.python.keras.callbacks.History at 0x7f3d408c9898>

参考文献：
https://study.163.com/course/introduction/1004573006.htm