tensorflow中过拟合以及网络超参数的选择实例笔记

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()
# 加载keras里的fashion_mnist数据

train_image = train_image/255  # 对训练集进行归一化
test_image = test_image/255

model = tf.keras.Sequential()
model.add(tf.keras.layers.Flatten(input_shape=(28,28)))  # 参考https://blog.csdn.net/qq_46244851/article/details/109584831
model.add(tf.keras.layers.Dense(128,activation='relu'))
model.add(tf.keras.layers.Dense(128,activation='relu'))
model.add(tf.keras.layers.Dense(128,activation='relu'))  # 层数多增加了2层
model.add(tf.keras.layers.Dense(10,activation='softmax'))  # 输出十个类，用softmax转化成概率值激活

model.compile(
    optimizer='adam',
    loss='sparse_categorical_crossentropy',  # 当label使用数字编码时就使用这个损失函数
    metrics=['acc']
)

history=model.fit(train_image,train_label,
          validation_data=(test_image,test_label),  # 提示训练集上的loss以及acc
          epochs=10)

Epoch 1/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.2287 - acc: 0.9129 - val_loss: 0.3322 - val_acc: 0.8850
Epoch 2/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.2208 - acc: 0.9160 - val_loss: 0.3488 - val_acc: 0.8822
Epoch 3/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.2139 - acc: 0.9197 - val_loss: 0.3382 - val_acc: 0.8868
Epoch 4/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.2092 - acc: 0.9203 - val_loss: 0.3514 - val_acc: 0.8868
Epoch 5/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.2046 - acc: 0.9225 - val_loss: 0.3459 - val_acc: 0.8883
Epoch 6/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.1958 - acc: 0.9251 - val_loss: 0.3621 - val_acc: 0.8821
Epoch 7/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.1911 - acc: 0.9268 - val_loss: 0.3806 - val_acc: 0.8863
Epoch 8/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.1843 - acc: 0.9296 - val_loss: 0.3520 - val_acc: 0.8926
Epoch 9/10
1875/1875 [==============================] - 3s 1ms/step - loss: 0.1807 - acc: 0.9304 - val_loss: 0.3510 - val_acc: 0.8886
Epoch 10/10
1875/1875 [==============================] - 3s 2ms/step - loss: 0.1767 - acc: 0.9316 - val_loss: 0.3967 - val_acc: 0.8846

history.history.keys()

dict_keys(['loss', 'acc', 'val_loss', 'val_acc'])

plt.plot(history.epoch,history.history.get('loss'),label='loss')plt.plot(history.epoch,history.history.get('val_loss'),label='val_loss')plt.legend()  # plt.legend（）函数主要的作用就是给图加上图例，就是蓝色表示loss，黄色是val_loss# plt.legend([x,y,z])里面的参数使用的是list的的形式将图表的的名称喂给这和函数。

<matplotlib.legend.Legend at 0x191cfb87d00>

plt.plot(history.epoch,history.history.get('acc'),label='acc')plt.plot(history.epoch,history.history.get('val_acc'),label='val_acc')plt.legend()

<matplotlib.legend.Legend at 0x191d04fae50>

model = tf.keras.Sequential()model.add(tf.keras.layers.Flatten(input_shape=(28,28)))  # 参考https://blog.csdn.net/qq_46244851/article/details/109584831model.add(tf.keras.layers.Dense(128,activation='relu'))model.add(tf.keras.layers.Dropout(0.5))  # 随机地丢去50%的隐藏层结点,抑制过拟合model.add(tf.keras.layers.Dense(128,activation='relu'))model.add(tf.keras.layers.Dropout(0.5))  # 随机地丢去50%的隐藏层结点model.add(tf.keras.layers.Dense(128,activation='relu'))  # 层数多增加了2层model.add(tf.keras.layers.Dropout(0.5))  # 随机地丢去50%的隐藏层结点model.add(tf.keras.layers.Dense(10,activation='softmax'))  # 输出十个类，用softmax转化成概率值激活

model.compile(    optimizer='adam',    loss='sparse_categorical_crossentropy',  # 当label使用数字编码时就使用这个损失函数    metrics=['acc'])

history=model.fit(train_image,train_label,          validation_data=(test_image,test_label),  # 提示训练集上的loss以及acc          epochs=10)

Epoch 1/101875/1875 [==============================] - 4s 2ms/step - loss: 0.8600 - acc: 0.6801 - val_loss: 0.5292 - val_acc: 0.8009Epoch 2/101875/1875 [==============================] - 4s 2ms/step - loss: 0.6184 - acc: 0.7778 - val_loss: 0.4645 - val_acc: 0.8250Epoch 3/101875/1875 [==============================] - 4s 2ms/step - loss: 0.5680 - acc: 0.7978 - val_loss: 0.4471 - val_acc: 0.8355Epoch 4/101875/1875 [==============================] - 4s 2ms/step - loss: 0.5364 - acc: 0.8112 - val_loss: 0.4488 - val_acc: 0.8409Epoch 5/101875/1875 [==============================] - 4s 2ms/step - loss: 0.5198 - acc: 0.8170 - val_loss: 0.4352 - val_acc: 0.8424Epoch 6/101875/1875 [==============================] - 4s 2ms/step - loss: 0.5074 - acc: 0.8200 - val_loss: 0.4235 - val_acc: 0.8459Epoch 7/101875/1875 [==============================] - 3s 2ms/step - loss: 0.4955 - acc: 0.8244 - val_loss: 0.4091 - val_acc: 0.8557Epoch 8/101875/1875 [==============================] - 3s 2ms/step - loss: 0.4900 - acc: 0.8287 - val_loss: 0.4125 - val_acc: 0.8562Epoch 9/101875/1875 [==============================] - 3s 2ms/step - loss: 0.4872 - acc: 0.8299 - val_loss: 0.4008 - val_acc: 0.8480Epoch 10/101875/1875 [==============================] - 3s 2ms/step - loss: 0.4778 - acc: 0.8340 - val_loss: 0.3936 - val_acc: 0.8535

model.summary()

Model: "sequential_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
flatten_1 (Flatten)          (None, 784)               0         
_________________________________________________________________
dense_4 (Dense)              (None, 128)               100480    
_________________________________________________________________
dropout (Dropout)            (None, 128)               0         
_________________________________________________________________
dense_5 (Dense)              (None, 128)               16512     
_________________________________________________________________
dropout_1 (Dropout)          (None, 128)               0         
_________________________________________________________________
dense_6 (Dense)              (None, 128)               16512     
_________________________________________________________________
dropout_2 (Dropout)          (None, 128)               0         
_________________________________________________________________
dense_7 (Dense)              (None, 10)                1290      
=================================================================
Total params: 134,794
Trainable params: 134,794
Non-trainable params: 0
_________________________________________________________________

plt.plot(history.epoch,history.history.get('loss'),label='loss')
plt.plot(history.epoch,history.history.get('val_loss'),label='val_loss')
plt.legend()  # plt.legend（）函数主要的作用就是给图加上图例，就是蓝色表示loss，黄色是val_loss
# plt.legend([x,y,z])里面的参数使用的是list的的形式将图表的的名称喂给这和函数。

<matplotlib.legend.Legend at 0x191cfc76df0>

plt.plot(history.epoch,history.history.get('acc'),label='acc')
plt.plot(history.epoch,history.history.get('val_acc'),label='val_acc')
plt.legend()

<matplotlib.legend.Legend at 0x191cfcbf6a0>