[Tensorflow2.X][原创]入门基础之fashion_mnist分类

[Tensorflow2.X][原创]入门基础之fashion_mnist分类

import tensorflow as tf

import matplotlib.pyplot as plt

import numpy as np

import pandas as pd

import sklearn

import os

import sys

 

 

def plot_learning_curves(history):

    pd.DataFrame(history.history).plot(figsize=(8, 5))

    plt.grid(True)

    plt.gca().set_ylim(0, 1)

    plt.show()

 

 

fashion_mnist = tf.keras.datasets.fashion_mnist # fashion_mnist数据集

(x_train_all, y_train_all), (x_test, y_test) = fashion_mnist.load_data() # 加载数据集

x_valid, x_train = x_train_all[:5000], x_train_all[5000:] # 从训练集中前5000个做为验证集,后面的作为训练集

y_valid, y_train = y_train_all[:5000], y_train_all[5000:]

 

model = tf.keras.models.Sequential() # 构建模型

model.add(tf.keras.layers.Flatten(input_shape=[28, 28])) #将输入扁平化，即转为一维向量[28x28,1]

model.add(tf.keras.layers.Dense(300, activation='sigmoid')) # 全连接层,300为卷积核大小WX=b

model.add(tf.keras.layers.Dense(100, activation='sigmoid')) # 全连接层,100为卷积核大小

model.add(tf.keras.layers.Dense(10, activation='softmax')) # 全连接层，输出10个预测值

 

# 另外一种网络模型搭建的写法

# model = tf.keras.models.Sequential([

# #     tf.keras.layers.Dense(300,activation='relu'),  # relu ==> max(0,x)

# #     tf.keras.layers.Dense(100,activation='relu'),  # softmax将向量转为概率分布

# #     tf.keras.layers.Dense(10,activation='softmax')

# # ])

 

print(model.summary()) #打印层结构

model.compile(loss='sparse_categorical_crossentropy', optimizer='sgd', metrics=['accuracy'])

history = model.fit(x_train, y_train, epochs=10, validation_data=(x_valid, y_valid))

 

plot_learning_curves(history)

 

model.summary()结构：

Model: "sequential"

_________________________________________________________________

Layer (type)                 Output Shape              Param #   

=================================================================

flatten (Flatten)            (None, 784)               0         

_________________________________________________________________

dense (Dense)                (None, 300)               235500    

_________________________________________________________________

dense_1 (Dense)              (None, 100)               30100     

_________________________________________________________________

dense_2 (Dense)              (None, 10)                1010      

=================================================================

Total params: 266,610

Trainable params: 266,610

Non-trainable params: 0

参数第2行计算出来是根据WX+b算出来的即

[None,784][784,300]+300=784x300+300=235500

参数第3行计算出来是根据WX+b算出来的即

[None,300][300,100]+100=300x100+100=30100其他依次类推

利用history打印训练变化参数的曲线：

 

最终结果：

loss: 0.5496 - accuracy: 0.8094 - val_loss: 0.5803 - val_accuracy: 0.7990