深度学习 Day 20——优化器对比实验

文章目录

深度学习 Day 20——优化器对比实验
- 一、前言
- 二、我的环境
- 三、前期工作
- - 1、设置GPU
  - 2、导入数据
  - 3、配置数据集
  - 4、数据可视化
- 三、构建模型
- 四、训练模型
- 五、模型评估
- - 1、Accuracy与Loss图
  - 2、评估模型
- 六、最后我想说

一、前言

🍨 本文为🔗365天深度学习训练营中的学习记录博客
🍦 参考文章：365天深度学习训练营-第11周：优化器对比实验（训练营内部成员可读）
🍖 原作者：K同学啊|接辅导、项目定制

在上一期数据增强实验中，我们将TensorFlow版本升级到了2.4.0，可能有些库会出现不兼容异常，大家需要版本对应一下。

本期博客，我们将着眼于深度学习中的各种优化器对比进行学习。

二、我的环境

电脑系统：Windows 11
语言环境：Python 3.8.5
编译器：DataSpell 2022.2
深度学习环境：TensorFlow 2.4.0
显卡及显存：RTX 3070 8G

三、前期工作

1、设置GPU

import tensorflow as tf
gpus = tf.config.list_physical_devices("GPU")

if gpus:
    gpu0 = gpus[0] #如果有多个GPU，仅使用第0个GPU
    tf.config.experimental.set_memory_growth(gpu0, True) #设置GPU显存用量按需使用
    tf.config.set_visible_devices([gpu0],"GPU")

from tensorflow          import keras
import matplotlib.pyplot as plt
import pandas            as pd
import numpy             as np
import warnings,os,PIL,pathlib

warnings.filterwarnings("ignore")             #忽略警告信息
plt.rcParams['font.sans-serif']    = ['SimHei']  # 用来正常显示中文标签
plt.rcParams['axes.unicode_minus'] = False    # 用来正常显示负号

2、导入数据

本期使用的数据集跟之前的好莱坞明星识别使用的数据集一样。

data_dir    = "/content/gdrive/MyDrive/data"
data_dir    = pathlib.Path(data_dir)
image_count = len(list(data_dir.glob('*/*')))
print("图片总数为：",image_count)

图片总数为： 1800

batch_size = 16
img_height = 336
img_width  = 336

train_ds = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir,
    validation_split=0.2,
    subset="training",
    seed=12,
    image_size=(img_height, img_width),
    batch_size=batch_size)

val_ds = tf.keras.preprocessing.image_dataset_from_directory(
    data_dir,
    validation_split=0.2,
    subset="validation",
    seed=12,
    image_size=(img_height, img_width),
    batch_size=batch_size)

Found 1800 files belonging to 17 classes.
Using 1440 files for training.
Found 1800 files belonging to 17 classes.
Using 360 files for validation.

查看一下数据文件标签：

class_names = train_ds.class_names
print(class_names)

['Angelina Jolie', 'Brad Pitt', 'Denzel Washington', 'Hugh Jackman', 'Jennifer Lawrence', 'Johnny Depp', 'Kate Winslet', 'Leonardo DiCaprio', 'Megan Fox', 'Natalie Portman', 'Nicole Kidman', 'Robert Downey Jr', 'Sandra Bullock', 'Scarlett Johansson', 'Tom Cruise', 'Tom Hanks', 'Will Smith']

3、配置数据集

AUTOTUNE = tf.data.AUTOTUNE

def train_preprocessing(image,label):
    return (image/255.0,label)

train_ds = (
    train_ds.cache()
    .shuffle(1000)
    .map(train_preprocessing)    # 这里可以设置预处理函数
#     .batch(batch_size)           # 在image_dataset_from_directory处已经设置了batch_size
    .prefetch(buffer_size=AUTOTUNE)
)

val_ds = (
    val_ds.cache()
    .shuffle(1000)
    .map(train_preprocessing)    # 这里可以设置预处理函数
#     .batch(batch_size)         # 在image_dataset_from_directory处已经设置了batch_size
    .prefetch(buffer_size=AUTOTUNE)
)

4、数据可视化

plt.figure(figsize=(10, 8))  # 图形的宽为10高为5
plt.suptitle("数据展示")

for images, labels in train_ds.take(1):
    for i in range(15):
        plt.subplot(4, 5, i + 1)
        plt.xticks([])
        plt.yticks([])
        plt.grid(False)

        # 显示图片
        plt.imshow(images[i])
        # 显示标签
        plt.xlabel(class_names[labels[i]-1])

plt.show()

在这里插入图片描述

三、构建模型

from tensorflow.keras.layers import Dropout,Dense,BatchNormalization
from tensorflow.keras.models import Model

def create_model(optimizer='adam'):
    # 加载预训练模型
    vgg16_base_model = tf.keras.applications.vgg16.VGG16(weights='imagenet',
                                                                include_top=False,
                                                                input_shape=(img_width, img_height, 3),
                                                                pooling='avg')
    for layer in vgg16_base_model.layers:
        layer.trainable = False

    X = vgg16_base_model.output
    
    X = Dense(170, activation='relu')(X)
    X = BatchNormalization()(X)
    X = Dropout(0.5)(X)

    output = Dense(len(class_names), activation='softmax')(X)
    vgg16_model = Model(inputs=vgg16_base_model.input, outputs=output)

    vgg16_model.compile(optimizer=optimizer,
                        loss='sparse_categorical_crossentropy',
                        metrics=['accuracy'])
    return vgg16_model

model1 = create_model(optimizer=tf.keras.optimizers.Adam())
model2 = create_model(optimizer=tf.keras.optimizers.SGD())
model2.summary()

打印的网络结构：

Downloading data from https://storage.googleapis.com/tensorflow/keras-applications/vgg16/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5
58892288/58889256 [==============================] - 8s 0us/step
Model: "model_1"
_________________________________________________________________
Layer (type)                 Output Shape              Param #   
=================================================================
input_2 (InputLayer)         [(None, 336, 336, 3)]     0         
_________________________________________________________________
block1_conv1 (Conv2D)        (None, 336, 336, 64)      1792      
_________________________________________________________________
block1_conv2 (Conv2D)        (None, 336, 336, 64)      36928     
_________________________________________________________________
block1_pool (MaxPooling2D)   (None, 168, 168, 64)      0         
_________________________________________________________________
block2_conv1 (Conv2D)        (None, 168, 168, 128)     73856     
_________________________________________________________________
block2_conv2 (Conv2D)        (None, 168, 168, 128)     147584    
_________________________________________________________________
block2_pool (MaxPooling2D)   (None, 84, 84, 128)       0         
_________________________________________________________________
block3_conv1 (Conv2D)        (None, 84, 84, 256)       295168    
_________________________________________________________________
block3_conv2 (Conv2D)        (None, 84, 84, 256)       590080    
_________________________________________________________________
block3_conv3 (Conv2D)        (None, 84, 84, 256)       590080    
_________________________________________________________________
block3_pool (MaxPooling2D)   (None, 42, 42, 256)       0         
_________________________________________________________________
block4_conv1 (Conv2D)        (None, 42, 42, 512)       1180160   
_________________________________________________________________
block4_conv2 (Conv2D)        (None, 42, 42, 512)       2359808   
_________________________________________________________________
block4_conv3 (Conv2D)        (None, 42, 42, 512)       2359808   
_________________________________________________________________
block4_pool (MaxPooling2D)   (None, 21, 21, 512)       0         
_________________________________________________________________
block5_conv1 (Conv2D)        (None, 21, 21, 512)       2359808   
_________________________________________________________________
block5_conv2 (Conv2D)        (None, 21, 21, 512)       2359808   
_________________________________________________________________
block5_conv3 (Conv2D)        (None, 21, 21, 512)       2359808   
_________________________________________________________________
block5_pool (MaxPooling2D)   (None, 10, 10, 512)       0         
_________________________________________________________________
global_average_pooling2d_1 ( (None, 512)               0         
_________________________________________________________________
dense_2 (Dense)              (None, 170)               87210     
_________________________________________________________________
batch_normalization_1 (Batch (None, 170)               680       
_________________________________________________________________
dropout_1 (Dropout)          (None, 170)               0         
_________________________________________________________________
dense_3 (Dense)              (None, 17)                2907      
=================================================================
Total params: 14,805,485
Trainable params: 90,457
Non-trainable params: 14,715,028
_________________________________________________________________

在这里我们是直接从网上下载VGG16模型并使用，可能会出现下载失败的情况，例如：

Exception: URL fetch failure on https://storage.googleapis.com/tensorflow/keras-applications/vgg16/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5: None -- [WinError 10054] 远程主机强迫关闭了一个现有的连接。

这种情况就是网络问题，导致无法下载，可以多试几次看看，如果一直都无法下载的话，可以直接进入上面错误中的网址：https://storage.googleapis.com/tensorflow/keras-applications/vgg16/vgg16_weights_tf_dim_ordering_tf_kernels_notop.h5它就会自动下载该模型，我们将其保存在项目文件夹中，然后我们在上面代码调用VGG16模型的时候里面的weights参数的值改成下载的VGG模型对应的地址即可。

在这里我们使用了两种优化器进行对比，Adam和SGD并对其两者进行简单的介绍：

Adam
```
keras.optimizers.Adam(lr=0.001, beta_1=0.9, beta_2=0.999, epsilon=None, decay=0.0, amsgrad=False)
```
它利用梯度的一阶矩估计和二阶矩估计动态调整每个参数的学习率。Adam的优点主要在于经过偏置校正后，每一次迭代学习率都有个确定范围，使得参数比较平稳。堆内存的需求比较小，也适用于大数据集和更高维空间的模型。
SGD
```
keras.optimizers.SGD(lr=0.01, momentum=0.0, decay=0.0, nesterov=False)
```
它是一种随机梯度下降优化器，SGD就是每一次迭代计算mini-batch的梯度，然后对参数进行更新，是最常见的优化方法了。

四、训练模型

NO_EPOCHS = 50

history_model1  = model1.fit(train_ds, epochs=NO_EPOCHS, verbose=1, validation_data=val_ds)
history_model2  = model2.fit(train_ds, epochs=NO_EPOCHS, verbose=1, validation_data=val_ds)

Epoch 1/50
90/90 [==============================] - 113s 1s/step - loss: 2.8072 - accuracy: 0.1535 - val_loss: 2.7235 - val_accuracy: 0.0556
Epoch 2/50
90/90 [==============================] - 20s 221ms/step - loss: 2.0860 - accuracy: 0.3243 - val_loss: 2.4607 - val_accuracy: 0.2833
Epoch 3/50
90/90 [==============================] - 21s 238ms/step - loss: 1.8125 - accuracy: 0.4132 - val_loss: 2.2316 - val_accuracy: 0.2972
Epoch 4/50
90/90 [==============================] - 20s 224ms/step - loss: 1.5680 - accuracy: 0.5146 - val_loss: 1.9419 - val_accuracy: 0.4361
Epoch 5/50
90/90 [==============================] - 20s 225ms/step - loss: 1.4038 - accuracy: 0.5681 - val_loss: 1.6831 - val_accuracy: 0.4833
Epoch 6/50
90/90 [==============================] - 20s 224ms/step - loss: 1.2327 - accuracy: 0.6153 - val_loss: 1.6376 - val_accuracy: 0.4944
Epoch 7/50
90/90 [==============================] - 20s 223ms/step - loss: 1.1563 - accuracy: 0.6486 - val_loss: 1.6727 - val_accuracy: 0.4417
Epoch 8/50
90/90 [==============================] - 20s 224ms/step - loss: 1.0707 - accuracy: 0.6694 - val_loss: 1.4806 - val_accuracy: 0.5250
Epoch 9/50
90/90 [==============================] - 20s 224ms/step - loss: 0.9549 - accuracy: 0.7125 - val_loss: 1.6010 - val_accuracy: 0.4889
Epoch 10/50
90/90 [==============================] - 20s 224ms/step - loss: 0.8829 - accuracy: 0.7347 - val_loss: 1.7179 - val_accuracy: 0.4611
Epoch 11/50
90/90 [==============================] - 20s 223ms/step - loss: 0.8417 - accuracy: 0.7389 - val_loss: 1.7174 - val_accuracy: 0.4833
Epoch 12/50
90/90 [==============================] - 20s 225ms/step - loss: 0.7601 - accuracy: 0.7708 - val_loss: 1.5996 - val_accuracy: 0.4833
Epoch 13/50
90/90 [==============================] - 20s 224ms/step - loss: 0.7254 - accuracy: 0.7757 - val_loss: 1.6183 - val_accuracy: 0.5278
Epoch 14/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6863 - accuracy: 0.8014 - val_loss: 1.7551 - val_accuracy: 0.4722
Epoch 15/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6336 - accuracy: 0.8069 - val_loss: 1.8830 - val_accuracy: 0.4639
Epoch 16/50
90/90 [==============================] - 20s 224ms/step - loss: 0.5819 - accuracy: 0.8319 - val_loss: 1.4917 - val_accuracy: 0.5389
Epoch 17/50
90/90 [==============================] - 20s 224ms/step - loss: 0.5748 - accuracy: 0.8340 - val_loss: 1.8751 - val_accuracy: 0.4694
Epoch 18/50
90/90 [==============================] - 20s 223ms/step - loss: 0.5219 - accuracy: 0.8396 - val_loss: 2.0875 - val_accuracy: 0.4861
Epoch 19/50
90/90 [==============================] - 20s 224ms/step - loss: 0.4934 - accuracy: 0.8556 - val_loss: 1.9038 - val_accuracy: 0.5028
Epoch 20/50
90/90 [==============================] - 20s 224ms/step - loss: 0.4942 - accuracy: 0.8514 - val_loss: 1.6452 - val_accuracy: 0.5444
Epoch 21/50
90/90 [==============================] - 20s 224ms/step - loss: 0.4933 - accuracy: 0.8431 - val_loss: 2.1585 - val_accuracy: 0.4472
Epoch 22/50
90/90 [==============================] - 20s 225ms/step - loss: 0.4514 - accuracy: 0.8701 - val_loss: 2.0218 - val_accuracy: 0.4972
Epoch 23/50
90/90 [==============================] - 20s 223ms/step - loss: 0.4458 - accuracy: 0.8694 - val_loss: 1.6499 - val_accuracy: 0.5417
Epoch 24/50
90/90 [==============================] - 20s 224ms/step - loss: 0.3927 - accuracy: 0.8917 - val_loss: 2.3310 - val_accuracy: 0.4222
Epoch 25/50
90/90 [==============================] - 20s 224ms/step - loss: 0.3870 - accuracy: 0.8854 - val_loss: 1.6200 - val_accuracy: 0.5583
Epoch 26/50
90/90 [==============================] - 20s 224ms/step - loss: 0.3800 - accuracy: 0.8861 - val_loss: 1.9285 - val_accuracy: 0.5361
Epoch 27/50
90/90 [==============================] - 20s 224ms/step - loss: 0.3792 - accuracy: 0.8771 - val_loss: 2.3675 - val_accuracy: 0.4806
Epoch 28/50
90/90 [==============================] - 20s 224ms/step - loss: 0.3321 - accuracy: 0.8986 - val_loss: 1.7445 - val_accuracy: 0.5500
Epoch 29/50
90/90 [==============================] - 20s 224ms/step - loss: 0.3185 - accuracy: 0.9076 - val_loss: 1.7202 - val_accuracy: 0.5639
Epoch 30/50
90/90 [==============================] - 20s 224ms/step - loss: 0.3436 - accuracy: 0.8958 - val_loss: 1.6614 - val_accuracy: 0.5667
Epoch 31/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2917 - accuracy: 0.9118 - val_loss: 2.0079 - val_accuracy: 0.5500
Epoch 32/50
90/90 [==============================] - 20s 224ms/step - loss: 0.3325 - accuracy: 0.8868 - val_loss: 2.0677 - val_accuracy: 0.5028
Epoch 33/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2879 - accuracy: 0.9146 - val_loss: 1.6412 - val_accuracy: 0.6028
Epoch 34/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2856 - accuracy: 0.9111 - val_loss: 2.1213 - val_accuracy: 0.5222
Epoch 35/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2645 - accuracy: 0.9153 - val_loss: 2.0940 - val_accuracy: 0.5222
Epoch 36/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2528 - accuracy: 0.9160 - val_loss: 1.8489 - val_accuracy: 0.5389
Epoch 37/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2553 - accuracy: 0.9208 - val_loss: 1.8388 - val_accuracy: 0.5583
Epoch 38/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2362 - accuracy: 0.9285 - val_loss: 1.8624 - val_accuracy: 0.5667
Epoch 39/50
90/90 [==============================] - 20s 223ms/step - loss: 0.2245 - accuracy: 0.9229 - val_loss: 1.9156 - val_accuracy: 0.5639
Epoch 40/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2198 - accuracy: 0.9333 - val_loss: 2.2192 - val_accuracy: 0.5556
Epoch 41/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2144 - accuracy: 0.9278 - val_loss: 1.8951 - val_accuracy: 0.5833
Epoch 42/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2074 - accuracy: 0.9389 - val_loss: 2.0159 - val_accuracy: 0.5500
Epoch 43/50
90/90 [==============================] - 20s 225ms/step - loss: 0.2166 - accuracy: 0.9257 - val_loss: 2.2641 - val_accuracy: 0.5111
Epoch 44/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2312 - accuracy: 0.9264 - val_loss: 2.0438 - val_accuracy: 0.5750
Epoch 45/50
90/90 [==============================] - 20s 223ms/step - loss: 0.2248 - accuracy: 0.9257 - val_loss: 2.2686 - val_accuracy: 0.5472
Epoch 46/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2102 - accuracy: 0.9375 - val_loss: 2.2441 - val_accuracy: 0.5583
Epoch 47/50
90/90 [==============================] - 20s 224ms/step - loss: 0.2120 - accuracy: 0.9340 - val_loss: 2.3860 - val_accuracy: 0.5361
Epoch 48/50
90/90 [==============================] - 20s 224ms/step - loss: 0.1959 - accuracy: 0.9354 - val_loss: 2.4052 - val_accuracy: 0.5167
Epoch 49/50
90/90 [==============================] - 20s 224ms/step - loss: 0.1699 - accuracy: 0.9521 - val_loss: 2.5167 - val_accuracy: 0.5250
Epoch 50/50
90/90 [==============================] - 20s 224ms/step - loss: 0.1645 - accuracy: 0.9528 - val_loss: 2.1405 - val_accuracy: 0.5722
Epoch 1/50
90/90 [==============================] - 21s 226ms/step - loss: 3.0785 - accuracy: 0.0986 - val_loss: 2.7949 - val_accuracy: 0.1000
Epoch 2/50
90/90 [==============================] - 20s 223ms/step - loss: 2.5472 - accuracy: 0.1924 - val_loss: 2.6379 - val_accuracy: 0.1583
Epoch 3/50
90/90 [==============================] - 20s 225ms/step - loss: 2.2651 - accuracy: 0.2694 - val_loss: 2.4596 - val_accuracy: 0.2528
Epoch 4/50
90/90 [==============================] - 20s 224ms/step - loss: 2.0612 - accuracy: 0.3389 - val_loss: 2.2347 - val_accuracy: 0.3389
Epoch 5/50
90/90 [==============================] - 20s 224ms/step - loss: 1.9508 - accuracy: 0.3653 - val_loss: 2.0695 - val_accuracy: 0.3972
Epoch 6/50
90/90 [==============================] - 20s 224ms/step - loss: 1.8406 - accuracy: 0.4021 - val_loss: 1.9282 - val_accuracy: 0.3917
Epoch 7/50
90/90 [==============================] - 20s 224ms/step - loss: 1.7565 - accuracy: 0.4451 - val_loss: 1.8469 - val_accuracy: 0.4111
Epoch 8/50
90/90 [==============================] - 20s 223ms/step - loss: 1.6587 - accuracy: 0.4667 - val_loss: 1.7935 - val_accuracy: 0.4306
Epoch 9/50
90/90 [==============================] - 20s 224ms/step - loss: 1.5934 - accuracy: 0.4889 - val_loss: 1.6561 - val_accuracy: 0.4528
Epoch 10/50
90/90 [==============================] - 20s 223ms/step - loss: 1.5516 - accuracy: 0.4854 - val_loss: 1.7235 - val_accuracy: 0.3944
Epoch 11/50
90/90 [==============================] - 20s 224ms/step - loss: 1.4753 - accuracy: 0.5403 - val_loss: 1.6903 - val_accuracy: 0.4333
Epoch 12/50
90/90 [==============================] - 20s 223ms/step - loss: 1.4309 - accuracy: 0.5389 - val_loss: 1.6633 - val_accuracy: 0.4556
Epoch 13/50
90/90 [==============================] - 20s 225ms/step - loss: 1.4168 - accuracy: 0.5437 - val_loss: 1.6759 - val_accuracy: 0.4667
Epoch 14/50
90/90 [==============================] - 20s 223ms/step - loss: 1.3726 - accuracy: 0.5701 - val_loss: 1.7004 - val_accuracy: 0.4667
Epoch 15/50
90/90 [==============================] - 20s 224ms/step - loss: 1.2890 - accuracy: 0.5924 - val_loss: 1.6371 - val_accuracy: 0.4639
Epoch 16/50
90/90 [==============================] - 20s 223ms/step - loss: 1.2669 - accuracy: 0.6139 - val_loss: 1.5207 - val_accuracy: 0.4806
Epoch 17/50
90/90 [==============================] - 20s 223ms/step - loss: 1.2238 - accuracy: 0.6097 - val_loss: 1.5294 - val_accuracy: 0.4972
Epoch 18/50
90/90 [==============================] - 20s 224ms/step - loss: 1.1582 - accuracy: 0.6375 - val_loss: 1.4838 - val_accuracy: 0.5111
Epoch 19/50
90/90 [==============================] - 20s 223ms/step - loss: 1.1518 - accuracy: 0.6271 - val_loss: 1.5244 - val_accuracy: 0.5111
Epoch 20/50
90/90 [==============================] - 20s 224ms/step - loss: 1.1324 - accuracy: 0.6438 - val_loss: 1.5217 - val_accuracy: 0.4917
Epoch 21/50
90/90 [==============================] - 21s 237ms/step - loss: 1.0931 - accuracy: 0.6590 - val_loss: 1.4744 - val_accuracy: 0.5056
Epoch 22/50
90/90 [==============================] - 20s 224ms/step - loss: 1.0524 - accuracy: 0.6667 - val_loss: 1.4386 - val_accuracy: 0.5167
Epoch 23/50
90/90 [==============================] - 20s 224ms/step - loss: 1.0196 - accuracy: 0.6729 - val_loss: 1.4282 - val_accuracy: 0.5278
Epoch 24/50
90/90 [==============================] - 20s 224ms/step - loss: 1.0143 - accuracy: 0.6924 - val_loss: 1.5158 - val_accuracy: 0.5361
Epoch 25/50
90/90 [==============================] - 20s 224ms/step - loss: 0.9708 - accuracy: 0.6875 - val_loss: 1.5623 - val_accuracy: 0.4806
Epoch 26/50
90/90 [==============================] - 20s 223ms/step - loss: 0.9651 - accuracy: 0.6875 - val_loss: 1.3693 - val_accuracy: 0.5611
Epoch 27/50
90/90 [==============================] - 20s 223ms/step - loss: 0.9384 - accuracy: 0.7076 - val_loss: 1.4377 - val_accuracy: 0.5556
Epoch 28/50
90/90 [==============================] - 20s 224ms/step - loss: 0.8951 - accuracy: 0.7285 - val_loss: 1.4171 - val_accuracy: 0.5222
Epoch 29/50
90/90 [==============================] - 20s 224ms/step - loss: 0.8706 - accuracy: 0.7340 - val_loss: 1.6458 - val_accuracy: 0.5167
Epoch 30/50
90/90 [==============================] - 20s 224ms/step - loss: 0.8520 - accuracy: 0.7375 - val_loss: 1.4419 - val_accuracy: 0.5139
Epoch 31/50
90/90 [==============================] - 20s 223ms/step - loss: 0.8547 - accuracy: 0.7188 - val_loss: 1.2940 - val_accuracy: 0.5889
Epoch 32/50
90/90 [==============================] - 20s 223ms/step - loss: 0.8222 - accuracy: 0.7424 - val_loss: 1.4509 - val_accuracy: 0.5528
Epoch 33/50
90/90 [==============================] - 20s 223ms/step - loss: 0.8406 - accuracy: 0.7299 - val_loss: 1.4598 - val_accuracy: 0.5306
Epoch 34/50
90/90 [==============================] - 20s 225ms/step - loss: 0.7983 - accuracy: 0.7528 - val_loss: 1.5114 - val_accuracy: 0.5472
Epoch 35/50
90/90 [==============================] - 20s 224ms/step - loss: 0.7992 - accuracy: 0.7403 - val_loss: 1.4475 - val_accuracy: 0.5750
Epoch 36/50
90/90 [==============================] - 20s 224ms/step - loss: 0.7557 - accuracy: 0.7569 - val_loss: 1.5024 - val_accuracy: 0.5389
Epoch 37/50
90/90 [==============================] - 20s 224ms/step - loss: 0.7298 - accuracy: 0.7681 - val_loss: 1.4272 - val_accuracy: 0.5389
Epoch 38/50
90/90 [==============================] - 20s 223ms/step - loss: 0.7378 - accuracy: 0.7632 - val_loss: 1.3973 - val_accuracy: 0.5778
Epoch 39/50
90/90 [==============================] - 20s 224ms/step - loss: 0.7025 - accuracy: 0.7875 - val_loss: 1.3738 - val_accuracy: 0.5500
Epoch 40/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6812 - accuracy: 0.7958 - val_loss: 1.5651 - val_accuracy: 0.5361
Epoch 41/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6646 - accuracy: 0.7854 - val_loss: 1.4765 - val_accuracy: 0.5667
Epoch 42/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6477 - accuracy: 0.8021 - val_loss: 1.5985 - val_accuracy: 0.5361
Epoch 43/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6508 - accuracy: 0.8042 - val_loss: 1.3467 - val_accuracy: 0.5667
Epoch 44/50
90/90 [==============================] - 20s 225ms/step - loss: 0.6539 - accuracy: 0.7889 - val_loss: 1.3919 - val_accuracy: 0.5778
Epoch 45/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6402 - accuracy: 0.8104 - val_loss: 1.3426 - val_accuracy: 0.5917
Epoch 46/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6178 - accuracy: 0.8076 - val_loss: 1.4094 - val_accuracy: 0.5833
Epoch 47/50
90/90 [==============================] - 20s 223ms/step - loss: 0.6083 - accuracy: 0.8000 - val_loss: 1.3747 - val_accuracy: 0.5750
Epoch 48/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6079 - accuracy: 0.8028 - val_loss: 1.5148 - val_accuracy: 0.5583
Epoch 49/50
90/90 [==============================] - 20s 224ms/step - loss: 0.6115 - accuracy: 0.8000 - val_loss: 1.9661 - val_accuracy: 0.4556
Epoch 50/50
90/90 [==============================] - 20s 224ms/step - loss: 0.5785 - accuracy: 0.8146 - val_loss: 1.4971 - val_accuracy: 0.5500

五、模型评估

1、Accuracy与Loss图

from matplotlib.ticker import MultipleLocator
plt.rcParams['savefig.dpi'] = 300 #图片像素
plt.rcParams['figure.dpi']  = 300 #分辨率

acc1     = history_model1.history['accuracy']
acc2     = history_model2.history['accuracy']
val_acc1 = history_model1.history['val_accuracy']
val_acc2 = history_model2.history['val_accuracy']

loss1     = history_model1.history['loss']
loss2     = history_model2.history['loss']
val_loss1 = history_model1.history['val_loss']
val_loss2 = history_model2.history['val_loss']

epochs_range = range(len(acc1))

plt.figure(figsize=(16, 4))
plt.subplot(1, 2, 1)

plt.plot(epochs_range, acc1, label='Training Accuracy-Adam')
plt.plot(epochs_range, acc2, label='Training Accuracy-SGD')
plt.plot(epochs_range, val_acc1, label='Validation Accuracy-Adam')
plt.plot(epochs_range, val_acc2, label='Validation Accuracy-SGD')
plt.legend(loc='lower right')
plt.title('Training and Validation Accuracy')
# 设置刻度间隔，x轴每1一个刻度
ax = plt.gca()
ax.xaxis.set_major_locator(MultipleLocator(1))

plt.subplot(1, 2, 2)
plt.plot(epochs_range, loss1, label='Training Loss-Adam')
plt.plot(epochs_range, loss2, label='Training Loss-SGD')
plt.plot(epochs_range, val_loss1, label='Validation Loss-Adam')
plt.plot(epochs_range, val_loss2, label='Validation Loss-SGD')
plt.legend(loc='upper right')
plt.title('Training and Validation Loss')
   
# 设置刻度间隔，x轴每1一个刻度
ax = plt.gca()
ax.xaxis.set_major_locator(MultipleLocator(1))

plt.show()

在这里插入图片描述

2、评估模型

def test_accuracy_report(model):
    score = model.evaluate(val_ds, verbose=0)
    print('Loss function: %s, accuracy:' % score[0], score[1])
    
test_accuracy_report(model2)

Loss function: 1.49705171585083, accuracy: 0.550000011920929

六、最后我想说

本期的博客到这里就结束了，最近我的电脑出现了一些问题导致无法对复杂的模型进行训练，最近考虑准备重装一下系统并清理一下电脑了，本次实验我是在Google Colaboratory上run的，这个平台目前我使用下来感觉不还错，提供免费的算力对我来说够了，大家如果自己的电脑配置不够的话也可以去试试。