深度学习实验--初步探索数据增强、优化器对模型的影响

🍨 本文为🔗365天深度学习训练营中的学习记录博客
🍖 原作者：K同学啊

前言

这次主要是探究，优化器、数据增强对模型训练的影响；
基础篇还剩下几个, 后面的难度会逐步提升;
越学，越觉得这个东西很玄学，没有过硬实力真的很难把控；
欢迎收藏 + 关注, 本人会持续更新.

文章目录

1、实验
- 优化器对比实验
- 优化器结合数据增强
- - 分别进行与不进行数据增强，跑20轮
  - 分别进行与不进行数据增强，跑50轮
2、代码实现
- 1、数据处理
- - 1、导入库
  - 2、数据导入与划分数据集
  - 3、展示一部分数据
  - 4、数据归一化和内存加速
  - 5、数据增强
  - 6、数据整合
- 2、模型构建和实验
- 3、模型训练与实验验证
- 4、结果检验

1、实验

💁‍♂ 提示：本次实验并不一定具有广泛意义，具体也是需要更具不同场景进行不同分析，这里仅仅作为简单的对比，初步了解不同优化器的区别。

优化器对比实验

目的：对比Adam和SGD优化器
条件：用优化版VGG16(详情请看代码实现)对人脸进行识别分类，训练50轮，效果如下：

在这里插入图片描述

⛹️‍♀️ 分析

Adam 每一次训练的时候是结合动量和自适应学习方法，可以自动调整学习率；
SGD是每一次训练一批数据的时候，是用一个一个样本进行训练的，比较简单；
效果分析：Adam比较快提高准确率，在训练初期的时候效果好，但是在后期的时候SGD综合效果更好一点点。

优化器结合数据增强

分别进行与不进行数据增强，跑20轮

效果如图：

不进行

在这里插入图片描述

进行

在这里插入图片描述

初步分析：

从准确率和损失率来看，发现数据增强后效果较好，但是都存在过拟合的情况，下面加大轮次进行训练。

分别进行与不进行数据增强，跑50轮

效果如图：

不进行

在这里插入图片描述

进行

在这里插入图片描述

初步分析：

从准确率和损失率来看，发现在加大训练轮次后，进行数据增强效果更好，因为解决了过拟合的情况，不进行数据增强依然存在过拟合情况；
效果：从模型效果来看，想要继续优化，最好的办法就是换更好的模型(本文用的是VGG16)，后面可以用Resnet这些模型进行再一次优化。

总结：

数据、优化器均对效果有不同影响，我感觉这也是深度学习很难得地方，很难把控，解决方法，我感觉只有多实践，多看论文，多复现论文积累经验。

2、代码实现

这里完整代码是50轮数据增强代码，不数据增强，只需要不允许

1、数据处理

1、导入库

import tensorflow as tf
import numpy as np 
from tensorflow.keras import models, datasets, layersgpu = tf.config.list_physical_devices("GPU")print(gpu)

[PhysicalDevice(name='/physical_device:GPU:0', device_type='GPU')]

2、数据导入与划分数据集

这里查看图像数据分类信息

import os, pathlibdata_dir = './data/'
data_dir = pathlib.Path(data_dir)personClassNames = os.listdir(data_dir)personClassNames

['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']

划分: 训练集：验证集 = 8 ： 2

batch_size = 16
image_width, image_height = 336, 336train_ds = tf.keras.preprocessing.image_dataset_from_directory('./data/',subset='training',validation_split=0.2,batch_size=batch_size,image_size=(image_width, image_height),seed=42,shuffle=True
)val_ds = tf.keras.preprocessing.image_dataset_from_directory('./data/',subset='validation',validation_split=0.2,batch_size=batch_size,image_size=(image_width, image_height),seed=42,shuffle=True
)

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

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

查看数据划分后的格式

one_batch_data = next(iter(train_ds))one_batch_data_images, one_batcg_data_labels = one_batch_dataprint("images [N, W, H, C]: ",one_batch_data_images.shape)
print("labels: ", one_batcg_data_labels)

images [N, W, H, C]:  (16, 336, 336, 3)
labels:  tf.Tensor([13 11 15  4  5  3 13  5 15  7  3  5  9 16 13  1], shape=(16,), dtype=int32)

3、展示一部分数据

import matplotlib.pyplot as plt plt.figure(figsize=(20, 10))for i in range(10):plt.subplot(5, 10, i + 1)plt.imshow(one_batch_data_images[i].numpy().astype('uint8'))plt.title(personClassNames[one_batcg_data_labels[i]])plt.axis('off')plt.show()

在这里插入图片描述

4、数据归一化和内存加速

AUTOTUNE = tf.data.AUTOTUNEdef train_preprocessing(image,label):return (image/255.0,label)train_ds = (train_ds.cache().shuffle(1000).map(train_preprocessing)    .prefetch(buffer_size=AUTOTUNE)
)val_ds = (val_ds.cache().shuffle(1000).map(train_preprocessing)    .prefetch(buffer_size=AUTOTUNE)
)

5、数据增强

这里对图像进行旋转

data_augmental = tf.keras.Sequential([tf.keras.layers.experimental.preprocessing.RandomFlip("horizontal_and_vertical"),tf.keras.layers.experimental.preprocessing.RandomRotation(0.2)
])

随机取一张图片看效果

# 取归一化后数据temp_data = next(iter(train_ds))temp_data_images, _ = temp_datatest_image = tf.expand_dims(temp_data_images[i], 0)# 随机旋转9张
plt.figure(figsize=(8, 8))
for i in range(9):data_strength = data_augmental(test_image)plt.subplot(3, 3, i + 1)plt.imshow(data_strength[0])plt.axis("off")

在这里插入图片描述

6、数据整合

对训练集进行数据增强

AUTOTUNE = tf.data.AUTOTUNEdef prepare(ds):ds = ds.map(lambda x, y : (data_augmental(x), y), num_parallel_calls=AUTOTUNE)return ds train_ds = prepare(train_ds)

2、模型构建和实验

这里采用VGG16模型，优化器默认使用“adam”，本次实验主要对比“adam”和“sgd", 池化层：采用平均池化
修改：全连接层大幅度降低计算量

from tensorflow.keras.applications import VGG16
from tensorflow.keras.layers import Dense, Flatten, Dropout, BatchNormalization
from tensorflow.keras.models import Modeldef create_model(optimizer='adam'):# VGG16模型vgg16 = VGG16(include_top=False, weights='imagenet', input_shape=(image_width, image_height, 3), pooling='avg')# 冻结权重for layer in vgg16.layers:layer.trainable = False# 获取卷积输出x = vgg16.output# 展开x = Flatten()(x)x = Dense(170, activation='relu')(x)  # 修改全连接层x = BatchNormalization()(x)x = Dropout(0.5)(x)# 分类输出output = Dense(len(personClassNames), activation='softmax')(x)# 模型整合vgg16_model = Model(inputs=vgg16.input, outputs=output)# 超参数设计vgg16_model.compile(optimizer=optimizer,loss='sparse_categorical_crossentropy',metrics=['accuracy'])return vgg16_model

实验：分别检验SGD和Aadm优化器的效果

model1 = create_model(optimizer=tf.keras.optimizers.Adam())
model2 = create_model(optimizer=tf.keras.optimizers.SGD())# 输出一个结构展示
model1.summary()

Model: "model"
_________________________________________________________________Layer (type)                Output Shape              Param #   
=================================================================input_1 (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 (G  (None, 512)              0         lobalAveragePooling2D)                                          flatten (Flatten)           (None, 512)               0         dense (Dense)               (None, 170)               87210     batch_normalization (BatchN  (None, 170)              680       ormalization)                                                   dropout (Dropout)           (None, 170)               0         dense_1 (Dense)             (None, 17)                2907      =================================================================
Total params: 14,805,485
Trainable params: 90,457
Non-trainable params: 14,715,028
_________________________________________________________________

3、模型训练与实验验证

epochs = 50history_model1 = model1.fit(train_ds,validation_data=val_ds,verbose=1,epochs=epochs
)history_model2 = model2.fit(train_ds,validation_data=val_ds,verbose=1,epochs=epochs
)

Epoch 1/50

2024-11-29 20:04:09.371628: I tensorflow/stream_executor/cuda/cuda_dnn.cc:384] Loaded cuDNN version 8101

 5/90 [>.............................] - ETA: 3s - loss: 3.6247 - accuracy: 0.0500    2024-11-29 20:04:11.269447: I tensorflow/stream_executor/cuda/cuda_blas.cc:1786] TensorFloat-32 will be used for the matrix multiplication. This will only be logged once.

90/90 [==============================] - 11s 65ms/step - loss: 2.9974 - accuracy: 0.1243 - val_loss: 2.7417 - val_accuracy: 0.0889
Epoch 2/50
90/90 [==============================] - 7s 65ms/step - loss: 2.4747 - accuracy: 0.2313 - val_loss: 2.5382 - val_accuracy: 0.1611
Epoch 3/50
90/90 [==============================] - 8s 68ms/step - loss: 2.1940 - accuracy: 0.2771 - val_loss: 2.3752 - val_accuracy: 0.2222
Epoch 4/50
90/90 [==============================] - 9s 62ms/step - loss: 2.0498 - accuracy: 0.3292 - val_loss: 2.1286 - val_accuracy: 0.3194
Epoch 5/50
90/90 [==============================] - 7s 61ms/step - loss: 1.9032 - accuracy: 0.3688 - val_loss: 2.0959 - val_accuracy: 0.2500
Epoch 6/50
90/90 [==============================] - 7s 64ms/step - loss: 1.8406 - accuracy: 0.3958 - val_loss: 1.8965 - val_accuracy: 0.3361
Epoch 7/50
90/90 [==============================] - 7s 60ms/step - loss: 1.7796 - accuracy: 0.4160 - val_loss: 1.8721 - val_accuracy: 0.3444
Epoch 8/50
90/90 [==============================] - 7s 67ms/step - loss: 1.6878 - accuracy: 0.4417 - val_loss: 1.8793 - val_accuracy: 0.3694
Epoch 9/50
90/90 [==============================] - 7s 67ms/step - loss: 1.5962 - accuracy: 0.4806 - val_loss: 2.1096 - val_accuracy: 0.3778
Epoch 10/50
90/90 [==============================] - 8s 65ms/step - loss: 1.5721 - accuracy: 0.4736 - val_loss: 1.8630 - val_accuracy: 0.3639
Epoch 11/50
90/90 [==============================] - 7s 67ms/step - loss: 1.5559 - accuracy: 0.4771 - val_loss: 1.8641 - val_accuracy: 0.3667
Epoch 12/50
90/90 [==============================] - 8s 59ms/step - loss: 1.5054 - accuracy: 0.4993 - val_loss: 1.6805 - val_accuracy: 0.4222
Epoch 13/50
90/90 [==============================] - 7s 62ms/step - loss: 1.4850 - accuracy: 0.5153 - val_loss: 1.7722 - val_accuracy: 0.4139
Epoch 14/50
90/90 [==============================] - 7s 64ms/step - loss: 1.4857 - accuracy: 0.5014 - val_loss: 1.9245 - val_accuracy: 0.4000
Epoch 15/50
90/90 [==============================] - 8s 63ms/step - loss: 1.4274 - accuracy: 0.5299 - val_loss: 1.6763 - val_accuracy: 0.4306
Epoch 16/50
90/90 [==============================] - 7s 58ms/step - loss: 1.4398 - accuracy: 0.5347 - val_loss: 1.8342 - val_accuracy: 0.4139
Epoch 17/50
90/90 [==============================] - 7s 68ms/step - loss: 1.3925 - accuracy: 0.5514 - val_loss: 1.9606 - val_accuracy: 0.3722
Epoch 18/50
90/90 [==============================] - 7s 64ms/step - loss: 1.3788 - accuracy: 0.5576 - val_loss: 1.9656 - val_accuracy: 0.4056
Epoch 19/50
90/90 [==============================] - 7s 66ms/step - loss: 1.3547 - accuracy: 0.5507 - val_loss: 1.7001 - val_accuracy: 0.4472
Epoch 20/50
90/90 [==============================] - 7s 63ms/step - loss: 1.3617 - accuracy: 0.5299 - val_loss: 1.7058 - val_accuracy: 0.4139
Epoch 21/50
90/90 [==============================] - 8s 61ms/step - loss: 1.3571 - accuracy: 0.5500 - val_loss: 1.7327 - val_accuracy: 0.4111
Epoch 22/50
90/90 [==============================] - 7s 64ms/step - loss: 1.3489 - accuracy: 0.5562 - val_loss: 1.8420 - val_accuracy: 0.4139
Epoch 23/50
90/90 [==============================] - 7s 71ms/step - loss: 1.3550 - accuracy: 0.5444 - val_loss: 1.5965 - val_accuracy: 0.4944
Epoch 24/50
90/90 [==============================] - 7s 63ms/step - loss: 1.3284 - accuracy: 0.5597 - val_loss: 1.7336 - val_accuracy: 0.3944
Epoch 25/50
90/90 [==============================] - 7s 66ms/step - loss: 1.3051 - accuracy: 0.5604 - val_loss: 2.1655 - val_accuracy: 0.3333
Epoch 26/50
90/90 [==============================] - 7s 66ms/step - loss: 1.2735 - accuracy: 0.5813 - val_loss: 1.8967 - val_accuracy: 0.3972
Epoch 27/50
90/90 [==============================] - 8s 58ms/step - loss: 1.2642 - accuracy: 0.5813 - val_loss: 1.8246 - val_accuracy: 0.4556
Epoch 28/50
90/90 [==============================] - 7s 63ms/step - loss: 1.2473 - accuracy: 0.5778 - val_loss: 1.8763 - val_accuracy: 0.4167
Epoch 29/50
90/90 [==============================] - 7s 61ms/step - loss: 1.2544 - accuracy: 0.5868 - val_loss: 1.8783 - val_accuracy: 0.4278
Epoch 30/50
90/90 [==============================] - 7s 62ms/step - loss: 1.2677 - accuracy: 0.5958 - val_loss: 1.9555 - val_accuracy: 0.3667
Epoch 31/50
90/90 [==============================] - 7s 63ms/step - loss: 1.2191 - accuracy: 0.5847 - val_loss: 2.1074 - val_accuracy: 0.3333
Epoch 32/50
90/90 [==============================] - 7s 64ms/step - loss: 1.1922 - accuracy: 0.5903 - val_loss: 1.8136 - val_accuracy: 0.4111
Epoch 33/50
90/90 [==============================] - 7s 71ms/step - loss: 1.2122 - accuracy: 0.5826 - val_loss: 1.8247 - val_accuracy: 0.4417
Epoch 34/50
90/90 [==============================] - 7s 65ms/step - loss: 1.1953 - accuracy: 0.5944 - val_loss: 1.7219 - val_accuracy: 0.4861
Epoch 35/50
90/90 [==============================] - 7s 59ms/step - loss: 1.2409 - accuracy: 0.5750 - val_loss: 2.0469 - val_accuracy: 0.4028
Epoch 36/50
90/90 [==============================] - 7s 64ms/step - loss: 1.1802 - accuracy: 0.5972 - val_loss: 1.7781 - val_accuracy: 0.4000
Epoch 37/50
90/90 [==============================] - 7s 65ms/step - loss: 1.2651 - accuracy: 0.5715 - val_loss: 1.7563 - val_accuracy: 0.4278
Epoch 38/50
90/90 [==============================] - 7s 62ms/step - loss: 1.1964 - accuracy: 0.5806 - val_loss: 1.7468 - val_accuracy: 0.4778
Epoch 39/50
90/90 [==============================] - 7s 61ms/step - loss: 1.1818 - accuracy: 0.6118 - val_loss: 1.5965 - val_accuracy: 0.4917
Epoch 40/50
90/90 [==============================] - 7s 69ms/step - loss: 1.2117 - accuracy: 0.5868 - val_loss: 2.2473 - val_accuracy: 0.3306
Epoch 41/50
90/90 [==============================] - 7s 66ms/step - loss: 1.1878 - accuracy: 0.6014 - val_loss: 1.6355 - val_accuracy: 0.4944
Epoch 42/50
90/90 [==============================] - 7s 60ms/step - loss: 1.1425 - accuracy: 0.6104 - val_loss: 2.2436 - val_accuracy: 0.3472
Epoch 43/50
90/90 [==============================] - 7s 65ms/step - loss: 1.1815 - accuracy: 0.5979 - val_loss: 1.7965 - val_accuracy: 0.4667
Epoch 44/50
90/90 [==============================] - 7s 63ms/step - loss: 1.1452 - accuracy: 0.6104 - val_loss: 1.8330 - val_accuracy: 0.4500
Epoch 45/50
90/90 [==============================] - 7s 63ms/step - loss: 1.1664 - accuracy: 0.5972 - val_loss: 1.8414 - val_accuracy: 0.4306
Epoch 46/50
90/90 [==============================] - 7s 59ms/step - loss: 1.1557 - accuracy: 0.6153 - val_loss: 2.9897 - val_accuracy: 0.3333
Epoch 47/50
90/90 [==============================] - 7s 72ms/step - loss: 1.1534 - accuracy: 0.6167 - val_loss: 1.9379 - val_accuracy: 0.4222
Epoch 48/50
90/90 [==============================] - 7s 61ms/step - loss: 1.1549 - accuracy: 0.6229 - val_loss: 1.6648 - val_accuracy: 0.4667
Epoch 49/50
90/90 [==============================] - 7s 59ms/step - loss: 1.1369 - accuracy: 0.6319 - val_loss: 1.7084 - val_accuracy: 0.4833
Epoch 50/50
90/90 [==============================] - 7s 60ms/step - loss: 1.1110 - accuracy: 0.6313 - val_loss: 1.7481 - val_accuracy: 0.4639
Epoch 1/50
90/90 [==============================] - 8s 60ms/step - loss: 3.1046 - accuracy: 0.0993 - val_loss: 2.8330 - val_accuracy: 0.0444
Epoch 2/50
90/90 [==============================] - 7s 61ms/step - loss: 2.7295 - accuracy: 0.1694 - val_loss: 2.6644 - val_accuracy: 0.1694
Epoch 3/50
90/90 [==============================] - 7s 62ms/step - loss: 2.5302 - accuracy: 0.2174 - val_loss: 2.5356 - val_accuracy: 0.1944
Epoch 4/50
90/90 [==============================] - 7s 62ms/step - loss: 2.3906 - accuracy: 0.2333 - val_loss: 2.3604 - val_accuracy: 0.2333
Epoch 5/50
90/90 [==============================] - 7s 68ms/step - loss: 2.2756 - accuracy: 0.2632 - val_loss: 2.3378 - val_accuracy: 0.2167
Epoch 6/50
90/90 [==============================] - 7s 60ms/step - loss: 2.1989 - accuracy: 0.2861 - val_loss: 2.2364 - val_accuracy: 0.2639
Epoch 7/50
90/90 [==============================] - 7s 63ms/step - loss: 2.1255 - accuracy: 0.3187 - val_loss: 2.1736 - val_accuracy: 0.3111
Epoch 8/50
90/90 [==============================] - 7s 64ms/step - loss: 2.1089 - accuracy: 0.3076 - val_loss: 1.9729 - val_accuracy: 0.3583
Epoch 9/50
90/90 [==============================] - 7s 69ms/step - loss: 2.0349 - accuracy: 0.3271 - val_loss: 1.9062 - val_accuracy: 0.3833
Epoch 10/50
90/90 [==============================] - 7s 66ms/step - loss: 1.9601 - accuracy: 0.3479 - val_loss: 1.8394 - val_accuracy: 0.4056
Epoch 11/50
90/90 [==============================] - 7s 65ms/step - loss: 1.9124 - accuracy: 0.3646 - val_loss: 1.8802 - val_accuracy: 0.3833
Epoch 12/50
90/90 [==============================] - 7s 66ms/step - loss: 1.8987 - accuracy: 0.3757 - val_loss: 1.8779 - val_accuracy: 0.4083
Epoch 13/50
90/90 [==============================] - 7s 61ms/step - loss: 1.8571 - accuracy: 0.3993 - val_loss: 1.7853 - val_accuracy: 0.4000
Epoch 14/50
90/90 [==============================] - 7s 63ms/step - loss: 1.7928 - accuracy: 0.4174 - val_loss: 1.8882 - val_accuracy: 0.4000
Epoch 15/50
90/90 [==============================] - 7s 63ms/step - loss: 1.7943 - accuracy: 0.4167 - val_loss: 1.8541 - val_accuracy: 0.3917
Epoch 16/50
90/90 [==============================] - 7s 59ms/step - loss: 1.7808 - accuracy: 0.4062 - val_loss: 1.8767 - val_accuracy: 0.3722
Epoch 17/50
90/90 [==============================] - 7s 71ms/step - loss: 1.7786 - accuracy: 0.4174 - val_loss: 1.7747 - val_accuracy: 0.3944
Epoch 18/50
90/90 [==============================] - 7s 67ms/step - loss: 1.7188 - accuracy: 0.4264 - val_loss: 1.7444 - val_accuracy: 0.4306
Epoch 19/50
90/90 [==============================] - 7s 63ms/step - loss: 1.7334 - accuracy: 0.4264 - val_loss: 1.9065 - val_accuracy: 0.3806
Epoch 20/50
90/90 [==============================] - 7s 67ms/step - loss: 1.6817 - accuracy: 0.4583 - val_loss: 1.7351 - val_accuracy: 0.4028
Epoch 21/50
90/90 [==============================] - 7s 68ms/step - loss: 1.6734 - accuracy: 0.4493 - val_loss: 1.7435 - val_accuracy: 0.4250
Epoch 22/50
90/90 [==============================] - 7s 67ms/step - loss: 1.6751 - accuracy: 0.4313 - val_loss: 1.8199 - val_accuracy: 0.3667
Epoch 23/50
90/90 [==============================] - 7s 59ms/step - loss: 1.6424 - accuracy: 0.4431 - val_loss: 1.7815 - val_accuracy: 0.3833
Epoch 24/50
90/90 [==============================] - 7s 64ms/step - loss: 1.6310 - accuracy: 0.4611 - val_loss: 2.1978 - val_accuracy: 0.3194
Epoch 25/50
90/90 [==============================] - 7s 58ms/step - loss: 1.6379 - accuracy: 0.4500 - val_loss: 1.7472 - val_accuracy: 0.4111
Epoch 26/50
90/90 [==============================] - 7s 66ms/step - loss: 1.5982 - accuracy: 0.4549 - val_loss: 1.7060 - val_accuracy: 0.4250
Epoch 27/50
90/90 [==============================] - 7s 64ms/step - loss: 1.6151 - accuracy: 0.4479 - val_loss: 1.7618 - val_accuracy: 0.3833
Epoch 28/50
90/90 [==============================] - 7s 67ms/step - loss: 1.5639 - accuracy: 0.4667 - val_loss: 1.7085 - val_accuracy: 0.4278
Epoch 29/50
90/90 [==============================] - 7s 68ms/step - loss: 1.5825 - accuracy: 0.4660 - val_loss: 1.7090 - val_accuracy: 0.4417
Epoch 30/50
90/90 [==============================] - 7s 59ms/step - loss: 1.5632 - accuracy: 0.4660 - val_loss: 1.6777 - val_accuracy: 0.4278
Epoch 31/50
90/90 [==============================] - 7s 59ms/step - loss: 1.5528 - accuracy: 0.4840 - val_loss: 1.7472 - val_accuracy: 0.4250
Epoch 32/50
90/90 [==============================] - 7s 59ms/step - loss: 1.5206 - accuracy: 0.4826 - val_loss: 1.6585 - val_accuracy: 0.4361
Epoch 33/50
90/90 [==============================] - 7s 64ms/step - loss: 1.5087 - accuracy: 0.5069 - val_loss: 1.8085 - val_accuracy: 0.3861
Epoch 34/50
90/90 [==============================] - 7s 62ms/step - loss: 1.4967 - accuracy: 0.4917 - val_loss: 1.7083 - val_accuracy: 0.4222
Epoch 35/50
90/90 [==============================] - 7s 64ms/step - loss: 1.5370 - accuracy: 0.4736 - val_loss: 1.8167 - val_accuracy: 0.3806
Epoch 36/50
90/90 [==============================] - 7s 63ms/step - loss: 1.5184 - accuracy: 0.4951 - val_loss: 1.7889 - val_accuracy: 0.4139
Epoch 37/50
90/90 [==============================] - 7s 61ms/step - loss: 1.4898 - accuracy: 0.5069 - val_loss: 1.7182 - val_accuracy: 0.4111
Epoch 38/50
90/90 [==============================] - 7s 59ms/step - loss: 1.4586 - accuracy: 0.5139 - val_loss: 1.7317 - val_accuracy: 0.4083
Epoch 39/50
90/90 [==============================] - 7s 66ms/step - loss: 1.4991 - accuracy: 0.4986 - val_loss: 1.8129 - val_accuracy: 0.3917
Epoch 40/50
90/90 [==============================] - 7s 62ms/step - loss: 1.4643 - accuracy: 0.5146 - val_loss: 1.7422 - val_accuracy: 0.4194
Epoch 41/50
90/90 [==============================] - 7s 61ms/step - loss: 1.4585 - accuracy: 0.5132 - val_loss: 1.6986 - val_accuracy: 0.4389
Epoch 42/50
90/90 [==============================] - 7s 63ms/step - loss: 1.4478 - accuracy: 0.5174 - val_loss: 1.6919 - val_accuracy: 0.4556
Epoch 43/50
90/90 [==============================] - 7s 58ms/step - loss: 1.4544 - accuracy: 0.5104 - val_loss: 1.6879 - val_accuracy: 0.4472
Epoch 44/50
90/90 [==============================] - 7s 58ms/step - loss: 1.4157 - accuracy: 0.5146 - val_loss: 1.6645 - val_accuracy: 0.4278
Epoch 45/50
90/90 [==============================] - 7s 63ms/step - loss: 1.4005 - accuracy: 0.5292 - val_loss: 1.7376 - val_accuracy: 0.4556
Epoch 46/50
90/90 [==============================] - 7s 67ms/step - loss: 1.4008 - accuracy: 0.5250 - val_loss: 1.7319 - val_accuracy: 0.4222
Epoch 47/50
90/90 [==============================] - 7s 59ms/step - loss: 1.4031 - accuracy: 0.5271 - val_loss: 1.7014 - val_accuracy: 0.4389
Epoch 48/50
90/90 [==============================] - 7s 67ms/step - loss: 1.3762 - accuracy: 0.5347 - val_loss: 1.7294 - val_accuracy: 0.4222
Epoch 49/50
90/90 [==============================] - 7s 60ms/step - loss: 1.4056 - accuracy: 0.5188 - val_loss: 1.7143 - val_accuracy: 0.4417
Epoch 50/50
90/90 [==============================] - 7s 64ms/step - loss: 1.4376 - accuracy: 0.5312 - val_loss: 1.7685 - val_accuracy: 0.4417

4、结果检验

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()