卷积神经网络（猫狗分类）
概述
- 数据来源：kaggle数据
- 下载地址：从https://www.kaggle.com/c/dogs-vs-cats/data （需要注册，并下载，文件大小800M）
- 目标：根据图像识别猫狗分类
- 方法：卷积神经网络
数据集整理
代码语言：javascript复制# 创建新数据集
import os, shutil
# 原始数据
original_dataset_dir = '/home/sunqi/python_study/train'
# 用来保存小数据集，缩小运算
base_dir = '/home/sunqi/python_study/cats_and_dogs_small'
# 创建文件夹
# os.mkdir(base_dir)
# 创建训练文件夹
train_dir = os.path.join(base_dir, 'train')
# os.mkdir(train_dir)
# 创建验证文件夹
validation_dir = os.path.join(base_dir, 'validation')
# os.mkdir(validation_dir)
# 创建测试文件夹
test_dir = os.path.join(base_dir, 'test')
# os.mkdir(test_dir)
# 放训练猫图
train_cats_dir = os.path.join(train_dir, 'cats')
# os.mkdir(train_cats_dir)
# 放训练狗图
train_dogs_dir = os.path.join(train_dir, 'dogs')
# os.mkdir(train_dogs_dir)
# 放验证猫图
validation_cats_dir = os.path.join(validation_dir, 'cats')
# os.mkdir(validation_cats_dir)
# 放验证狗图
validation_dogs_dir = os.path.join(validation_dir, 'dogs')
# os.mkdir(validation_dogs_dir)
# 放测试猫图
test_cats_dir = os.path.join(test_dir, 'cats')
# os.mkdir(test_cats_dir)
# 放测试狗图
test_dogs_dir = os.path.join(test_dir, 'dogs')
# os.mkdir(test_dogs_dir)# 将图片放入建立好的文件夹
# 前1000张图，文件名为cat.1-999,所以通过循环进行
fnames = ['cat.{}.jpg'.format(i) for i in range(1000)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(train_cats_dir, fname)
shutil.copyfile(src, dst)
fnames = ['cat.{}.jpg'.format(i) for i in range(1000, 1500)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(validation_cats_dir, fname)
shutil.copyfile(src, dst)
fnames = ['cat.{}.jpg'.format(i) for i in range(1500, 2000)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(test_cats_dir, fname)
shutil.copyfile(src, dst)
fnames = ['dog.{}.jpg'.format(i) for i in range(1000)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(train_dogs_dir, fname)
shutil.copyfile(src, dst)
fnames = ['dog.{}.jpg'.format(i) for i in range(1000, 1500)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(validation_dogs_dir, fname)
shutil.copyfile(src, dst)
fnames = ['dog.{}.jpg'.format(i) for i in range(1500, 2000)]
for fname in fnames:
src = os.path.join(original_dataset_dir, fname)
dst = os.path.join(test_dogs_dir, fname)
shutil.copyfile(src, dst)# 构建模型
# 这里和之前的内容一样
from keras import layers
from keras import models
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))Using TensorFlow backend.from keras import optimizers
model.compile(loss='binary_crossentropy',
optimizer=optimizers.RMSprop(lr=1e-4),
metrics=['acc'])数据预处理
- 读取图像文件
- 将JPEG 文件解码为RGB 像素网格
- 将这些像素网格转换为浮点数张量
- 将像素值（0-255 范围内）缩放到0-1区间
from keras.preprocessing.image import ImageDataGenerator
train_datagen = ImageDataGenerator(rescale=1./255)#图像缩放
test_datagen = ImageDataGenerator(rescale=1./255)
# 这个是一个python生成器，里面的东西不知道是什么，不断读取的意思
train_generator = train_datagen.flow_from_directory(
train_dir,# 目标目录
target_size=(150, 150),# 将图像大小调整为150*150,对应模型的输入张量
batch_size=20,
class_mode='binary')#标签
validation_generator = test_datagen.flow_from_directory(
validation_dir,
target_size=(150, 150),
batch_size=20,
class_mode='binary')Found 2000 images belonging to 2 classes.
Found 1000 images belonging to 2 classes.# 利用批量生成器拟合模型
history = model.fit_generator(
train_generator,# python生成器，相当于循环
steps_per_epoch=100,#抽100个批量
epochs=30,#30次迭代
validation_data=validation_generator,
validation_steps=50)
# 保存模型
model.save('cats_and_dogs_small_1.h5')Epoch 1/30
100/100 [==============================] - 13s 130ms/step - loss: 0.6921 - acc: 0.5240 - val_loss: 0.6474 - val_acc: 0.5540
Epoch 2/30
100/100 [==============================] - 12s 120ms/step - loss: 0.6591 - acc: 0.6095 - val_loss: 0.6260 - val_acc: 0.5480
Epoch 3/30
100/100 [==============================] - 12s 119ms/step - loss: 0.6157 - acc: 0.6675 - val_loss: 0.6492 - val_acc: 0.6670
Epoch 4/30
100/100 [==============================] - 12s 120ms/step - loss: 0.5689 - acc: 0.7135 - val_loss: 0.5767 - val_acc: 0.6150
Epoch 5/30
100/100 [==============================] - 12s 119ms/step - loss: 0.5439 - acc: 0.7225 - val_loss: 0.7602 - val_acc: 0.6490
Epoch 6/30
100/100 [==============================] - 12s 119ms/step - loss: 0.5148 - acc: 0.7300 - val_loss: 0.7002 - val_acc: 0.6870
Epoch 7/30
100/100 [==============================] - 12s 120ms/step - loss: 0.4917 - acc: 0.7730 - val_loss: 0.6359 - val_acc: 0.6550
Epoch 8/30
100/100 [==============================] - 12s 120ms/step - loss: 0.4717 - acc: 0.7750 - val_loss: 0.8081 - val_acc: 0.7080
Epoch 9/30
100/100 [==============================] - 12s 120ms/step - loss: 0.4472 - acc: 0.7875 - val_loss: 0.5322 - val_acc: 0.7040
Epoch 10/30
100/100 [==============================] - 12s 120ms/step - loss: 0.4158 - acc: 0.8170 - val_loss: 0.3662 - val_acc: 0.7460
Epoch 11/30
100/100 [==============================] - 12s 119ms/step - loss: 0.4017 - acc: 0.8155 - val_loss: 0.4341 - val_acc: 0.7200
Epoch 12/30
100/100 [==============================] - 12s 120ms/step - loss: 0.3803 - acc: 0.8270 - val_loss: 0.4769 - val_acc: 0.7250
Epoch 13/30
100/100 [==============================] - 12s 121ms/step - loss: 0.3574 - acc: 0.8425 - val_loss: 0.6853 - val_acc: 0.7240
Epoch 14/30
100/100 [==============================] - 12s 120ms/step - loss: 0.3360 - acc: 0.8565 - val_loss: 0.2616 - val_acc: 0.7040
Epoch 15/30
100/100 [==============================] - 12s 119ms/step - loss: 0.3012 - acc: 0.8735 - val_loss: 0.5142 - val_acc: 0.7460
Epoch 16/30
100/100 [==============================] - 12s 119ms/step - loss: 0.2834 - acc: 0.8880 - val_loss: 0.6345 - val_acc: 0.7280
Epoch 17/30
100/100 [==============================] - 12s 119ms/step - loss: 0.2663 - acc: 0.8955 - val_loss: 0.4337 - val_acc: 0.7220
Epoch 18/30
100/100 [==============================] - 12s 119ms/step - loss: 0.2482 - acc: 0.9065 - val_loss: 0.3528 - val_acc: 0.7420
Epoch 19/30
100/100 [==============================] - 12s 120ms/step - loss: 0.2251 - acc: 0.9120 - val_loss: 0.4797 - val_acc: 0.7270
Epoch 20/30
100/100 [==============================] - 12s 118ms/step - loss: 0.2171 - acc: 0.9125 - val_loss: 0.3961 - val_acc: 0.7180
Epoch 21/30
100/100 [==============================] - 12s 119ms/step - loss: 0.1905 - acc: 0.9365 - val_loss: 1.1006 - val_acc: 0.7300
Epoch 22/30
100/100 [==============================] - 12s 118ms/step - loss: 0.1747 - acc: 0.9400 - val_loss: 0.5780 - val_acc: 0.7140
Epoch 23/30
100/100 [==============================] - 12s 120ms/step - loss: 0.1476 - acc: 0.9495 - val_loss: 0.6581 - val_acc: 0.7220
Epoch 24/30
100/100 [==============================] - 12s 120ms/step - loss: 0.1385 - acc: 0.9540 - val_loss: 0.3878 - val_acc: 0.7340
Epoch 25/30
100/100 [==============================] - 12s 119ms/step - loss: 0.1168 - acc: 0.9625 - val_loss: 0.4306 - val_acc: 0.7410
Epoch 26/30
100/100 [==============================] - 12s 119ms/step - loss: 0.1064 - acc: 0.9675 - val_loss: 0.5825 - val_acc: 0.7260
Epoch 27/30
100/100 [==============================] - 12s 119ms/step - loss: 0.1011 - acc: 0.9720 - val_loss: 1.1769 - val_acc: 0.7090
Epoch 28/30
100/100 [==============================] - 12s 119ms/step - loss: 0.0864 - acc: 0.9765 - val_loss: 0.9043 - val_acc: 0.7370
Epoch 29/30
100/100 [==============================] - 12s 118ms/step - loss: 0.0761 - acc: 0.9790 - val_loss: 0.6539 - val_acc: 0.7320
Epoch 30/30
100/100 [==============================] - 12s 118ms/step - loss: 0.0597 - acc: 0.9855 - val_loss: 0.4191 - val_acc: 0.7160# 绘制进度曲线
import matplotlib.pyplot as plt
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs = range(1, len(acc) 1)
plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.figure()
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
plt.show()
在第5轮之后就出现了训练和验证的差异，存在过拟合问题
数据增强
数据增强是从现有的训练样本中生成更多的训练数据，方法是利用多种能够生成可信图像的随机变换来增加样本，比如对图片进行角度变换，平移等方法 目的是为了防止模型的过拟合
代码语言：javascript复制# 对图片的数据增强
datagen = ImageDataGenerator(
rotation_range=40,#角度
width_shift_range=0.2,#宽度移动
height_shift_range=0.2,#高度移动
shear_range=0.2,#随机切换角度
zoom_range=0.2,#随机放大
horizontal_flip=True,#随机将一半图像水平翻
fill_mode='nearest')#填充方法# 查看几个数据增强后的结果
from keras.preprocessing import image# 导入图像处理模块
fnames = [os.path.join(train_cats_dir, fname) for
fname in os.listdir(train_cats_dir)]
img_path = fnames[3]#找一张图看看
img = image.load_img(img_path, target_size=(150, 150))#读取图片调整大小
# 转换为数组，并调整形状
x = image.img_to_array(img)
x = x.reshape((1,) x.shape)
# 进行随机变换
i = 0
for batch in datagen.flow(x, batch_size=1):
plt.figure(i)
imgplot = plt.imshow(image.array_to_img(batch[0]))
i = 1
if i % 4 == 0:#目的是生成限制生成四张图片
break
plt.show()
生成的四张图片和原始图片内容一致，但是角度方向等存在差异，又可以看做不一致
Dropout减少过拟合
前面的数据增强可以减少一部分过拟合，但是不足以完全消除过拟合，因此可以添加一个Dropout层再次减少过拟合 Dropout层进行的操作是舍弃一些值较小的特征根
代码语言：javascript复制## 重新训练一个模型
model = models.Sequential()
model.add(layers.Conv2D(32, (3, 3), activation='relu',
input_shape=(150, 150, 3)))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(64, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Conv2D(128, (3, 3), activation='relu'))
model.add(layers.MaxPooling2D((2, 2)))
model.add(layers.Flatten())
model.add(layers.Dropout(0.5))
model.add(layers.Dense(512, activation='relu'))
model.add(layers.Dense(1, activation='sigmoid'))
model.compile(loss='binary_crossentropy',
optimizer=optimizers.RMSprop(lr=1e-4),
metrics=['acc'])# 使用数据增强训练
train_datagen = ImageDataGenerator(
rescale=1./255,
rotation_range=40,
width_shift_range=0.2,
height_shift_range=0.2,
shear_range=0.2,
zoom_range=0.2,
horizontal_flip=True,)
test_datagen = ImageDataGenerator(rescale=1./255)# 测试数据不能使用数据增强
train_generator = train_datagen.flow_from_directory(
train_dir,
target_size=(150, 150),
batch_size=32,
class_mode='binary')
validation_generator = test_datagen.flow_from_directory(
validation_dir,
target_size=(150, 150),
batch_size=32,
class_mode='binary')
history = model.fit_generator(
train_generator,
steps_per_epoch=100,
epochs=100,
validation_data=validation_generator,
validation_steps=50)Found 2000 images belonging to 2 classes.
Found 1000 images belonging to 2 classes.
Epoch 1/100
100/100 [==============================] - 27s 271ms/step - loss: 0.6917 - acc: 0.5234 - val_loss: 0.6728 - val_acc: 0.5019
Epoch 2/100
100/100 [==============================] - 25s 252ms/step - loss: 0.6838 - acc: 0.5726 - val_loss: 0.7077 - val_acc: 0.6198
Epoch 3/100
100/100 [==============================] - 25s 253ms/step - loss: 0.6066 - acc: 0.6809 - val_loss: 0.6399 - val_acc: 0.6961
Epoch 8/100
100/100 [==============================] - 26s 255ms/step - loss: 0.5883 - acc: 0.6888 - val_loss: 0.7738 - val_acc: 0.6991
Epoch 9/100
100/100 [==============================] - 25s 254ms/step - loss: 0.5854 - acc: 0.6793 - val_loss: 0.5877 - val_acc: 0.6720
Epoch 10/100
100/100 [==============================] - 25s 253ms/step - loss: 0.5666 - acc: 0.7105 - val_loss: 0.5192 - val_acc: 0.7030
Epoch 11/100
100/100 [==============================] - 26s 263ms/step - loss: 0.5717 - acc: 0.6932 - val_loss: 0.6657 - val_acc: 0.6334
Epoch 12/100
100/100 [==============================] - 25s 251ms/step - loss: 0.5644 - acc: 0.7143 - val_loss: 0.4830 - val_acc: 0.7367
Epoch 13/100
100/100 [==============================] - 26s 255ms/step - loss: 0.5572 - acc: 0.7051 - val_loss: 0.5695 - val_acc: 0.7210
Epoch 14/100
100/100 [==============================] - 25s 251ms/step - loss: 0.5526 - acc: 0.7128 - val_loss: 0.4823 - val_acc: 0.7506
Epoch 15/100
100/100 [==============================] - 25s 253ms/step - loss: 0.5372 - acc: 0.7289 - val_loss: 0.6230 - val_acc: 0.6914
Epoch 16/100
100/100 [==============================] - 26s 258ms/step - loss: 0.5427 - acc: 0.7162 - val_loss: 0.5232 - val_acc: 0.7494
Epoch 17/100
100/100 [==============================] - 25s 250ms/step - loss: 0.5378 - acc: 0.7384 - val_loss: 0.5064 - val_acc: 0.7494
Epoch 18/100
100/100 [==============================] - 27s 270ms/step - loss: 0.5448 - acc: 0.7159 - val_loss: 0.4783 - val_acc: 0.7468
Epoch 19/100
100/100 [==============================] - 25s 254ms/step - loss: 0.5202 - acc: 0.7427 - val_loss: 0.6163 - val_acc: 0.7360
Epoch 20/100
100/100 [==============================] - 26s 257ms/step - loss: 0.5259 - acc: 0.7306 - val_loss: 0.4490 - val_acc: 0.7571
Epoch 21/100
100/100 [==============================] - 25s 254ms/step - loss: 0.5214 - acc: 0.7380 - val_loss: 0.5362 - val_acc: 0.7627
Epoch 22/100
100/100 [==============================] - 26s 258ms/step - loss: 0.5174 - acc: 0.7481 - val_loss: 0.4221 - val_acc: 0.7345
Epoch 23/100
100/100 [==============================] - 27s 268ms/step - loss: 0.5056 - acc: 0.7443 - val_loss: 0.6451 - val_acc: 0.7538
Epoch 24/100
100/100 [==============================] - 26s 258ms/step - loss: 0.5038 - acc: 0.7541 - val_loss: 0.6212 - val_acc: 0.7352
Epoch 25/100
100/100 [==============================] - 25s 252ms/step - loss: 0.5005 - acc: 0.7491 - val_loss: 0.5299 - val_acc: 0.7442
Epoch 26/100
100/100 [==============================] - 25s 254ms/step - loss: 0.5027 - acc: 0.7591 - val_loss: 0.6129 - val_acc: 0.7278
Epoch 27/100
100/100 [==============================] - 25s 255ms/step - loss: 0.4866 - acc: 0.7626 - val_loss: 0.4164 - val_acc: 0.7861
Epoch 28/100
100/100 [==============================] - 26s 264ms/step - loss: 0.4867 - acc: 0.7563 - val_loss: 0.5636 - val_acc: 0.7456
Epoch 29/100
100/100 [==============================] - 26s 258ms/step - loss: 0.4985 - acc: 0.7535 - val_loss: 0.4185 - val_acc: 0.7693
Epoch 30/100
100/100 [==============================] - 25s 251ms/step - loss: 0.4884 - acc: 0.7582 - val_loss: 0.5540 - val_acc: 0.7253
Epoch 31/100
100/100 [==============================] - 26s 256ms/step - loss: 0.4805 - acc: 0.7708 - val_loss: 0.3387 - val_acc: 0.7700
Epoch 32/100
100/100 [==============================] - 25s 251ms/step - loss: 0.4758 - acc: 0.7638 - val_loss: 0.3540 - val_acc: 0.7751
Epoch 33/100
100/100 [==============================] - 26s 262ms/step - loss: 0.4763 - acc: 0.7751 - val_loss: 0.4014 - val_acc: 0.7538
Epoch 34/100
100/100 [==============================] - 25s 252ms/step - loss: 0.4736 - acc: 0.7758 - val_loss: 0.3543 - val_acc: 0.7622
Epoch 35/100
100/100 [==============================] - 27s 269ms/step - loss: 0.4729 - acc: 0.7758 - val_loss: 0.3109 - val_acc: 0.7963
Epoch 36/100
100/100 [==============================] - 25s 254ms/step - loss: 0.4740 - acc: 0.7713 - val_loss: 0.6062 - val_acc: 0.7790
Epoch 37/100
100/100 [==============================] - 25s 252ms/step - loss: 0.4802 - acc: 0.7666 - val_loss: 0.4107 - val_acc: 0.7976
Epoch 38/100
100/100 [==============================] - 26s 259ms/step - loss: 0.4459 - acc: 0.7847 - val_loss: 0.4499 - val_acc: 0.7938
Epoch 39/100
100/100 [==============================] - 25s 252ms/step - loss: 0.4638 - acc: 0.7795 - val_loss: 0.3357 - val_acc: 0.7881
Epoch 40/100
100/100 [==============================] - 27s 273ms/step - loss: 0.4673 - acc: 0.7732 - val_loss: 0.3718 - val_acc: 0.7764
Epoch 41/100
100/100 [==============================] - 25s 253ms/step - loss: 0.4564 - acc: 0.7759 - val_loss: 0.4136 - val_acc: 0.7932
Epoch 42/100
100/100 [==============================] - 25s 252ms/step - loss: 0.4441 - acc: 0.7921 - val_loss: 0.5463 - val_acc: 0.7944
Epoch 43/100
100/100 [==============================] - 25s 254ms/step - loss: 0.4525 - acc: 0.7828 - val_loss: 0.7509 - val_acc: 0.7358
Epoch 44/100
100/100 [==============================] - 25s 253ms/step - loss: 0.4503 - acc: 0.7787 - val_loss: 0.2707 - val_acc: 0.8027
Epoch 45/100
100/100 [==============================] - 27s 266ms/step - loss: 0.4425 - acc: 0.7990 - val_loss: 0.4280 - val_acc: 0.7816
Epoch 46/100
100/100 [==============================] - 25s 250ms/step - loss: 0.4389 - acc: 0.7958 - val_loss: 0.2661 - val_acc: 0.7931
Epoch 47/100
100/100 [==============================] - 26s 255ms/step - loss: 0.4325 - acc: 0.7965 - val_loss: 0.4029 - val_acc: 0.8241
Epoch 48/100
100/100 [==============================] - 26s 255ms/step - loss: 0.4368 - acc: 0.7910 - val_loss: 0.3177 - val_acc: 0.7906
Epoch 49/100
100/100 [==============================] - 25s 252ms/step - loss: 0.4303 - acc: 0.7999 - val_loss: 0.5532 - val_acc: 0.8014
Epoch 50/100
100/100 [==============================] - 26s 262ms/step - loss: 0.4190 - acc: 0.8056 - val_loss: 0.2984 - val_acc: 0.7970
Epoch 51/100
100/100 [==============================] - 25s 250ms/step - loss: 0.4372 - acc: 0.7958 - val_loss: 0.4816 - val_acc: 0.7792
Epoch 52/100
100/100 [==============================] - 27s 271ms/step - loss: 0.4146 - acc: 0.8084 - val_loss: 0.7304 - val_acc: 0.8112
Epoch 53/100
100/100 [==============================] - 25s 254ms/step - loss: 0.4406 - acc: 0.7932 - val_loss: 0.5950 - val_acc: 0.8141
Epoch 54/100
100/100 [==============================] - 26s 257ms/step - loss: 0.4172 - acc: 0.8071 - val_loss: 0.4298 - val_acc: 0.7790
Epoch 55/100
100/100 [==============================] - 25s 252ms/step - loss: 0.4205 - acc: 0.8097 - val_loss: 0.3395 - val_acc: 0.8173
Epoch 56/100
100/100 [==============================] - 26s 255ms/step - loss: 0.4145 - acc: 0.8090 - val_loss: 0.5249 - val_acc: 0.7680
Epoch 57/100
100/100 [==============================] - 26s 262ms/step - loss: 0.3981 - acc: 0.8185 - val_loss: 0.3922 - val_acc: 0.7938
Epoch 58/100
100/100 [==============================] - 25s 248ms/step - loss: 0.4123 - acc: 0.8084 - val_loss: 0.3180 - val_acc: 0.8033
Epoch 59/100
100/100 [==============================] - 25s 254ms/step - loss: 0.4123 - acc: 0.8049 - val_loss: 0.3938 - val_acc: 0.8144
Epoch 60/100
100/100 [==============================] - 25s 254ms/step - loss: 0.3896 - acc: 0.8200 - val_loss: 0.4451 - val_acc: 0.8115
Epoch 61/100
100/100 [==============================] - 26s 256ms/step - loss: 0.3918 - acc: 0.8247 - val_loss: 0.4170 - val_acc: 0.8196
Epoch 62/100
100/100 [==============================] - 26s 260ms/step - loss: 0.4121 - acc: 0.8188 - val_loss: 0.4389 - val_acc: 0.8103
Epoch 63/100
100/100 [==============================] - 25s 255ms/step - loss: 0.4033 - acc: 0.8163 - val_loss: 0.5156 - val_acc: 0.7977
Epoch 64/100
100/100 [==============================] - 25s 252ms/step - loss: 0.3946 - acc: 0.8163 - val_loss: 0.2393 - val_acc: 0.8099
Epoch 65/100
100/100 [==============================] - 25s 249ms/step - loss: 0.3801 - acc: 0.8283 - val_loss: 0.3290 - val_acc: 0.8223
Epoch 66/100
100/100 [==============================] - 26s 256ms/step - loss: 0.3953 - acc: 0.8150 - val_loss: 0.4010 - val_acc: 0.8235
Epoch 67/100
100/100 [==============================] - 25s 253ms/step - loss: 0.3879 - acc: 0.8274 - val_loss: 0.7320 - val_acc: 0.7392
Epoch 68/100
100/100 [==============================] - 25s 252ms/step - loss: 0.3910 - acc: 0.8116 - val_loss: 0.2840 - val_acc: 0.8035
Epoch 69/100
100/100 [==============================] - 26s 263ms/step - loss: 0.3904 - acc: 0.8241 - val_loss: 0.2904 - val_acc: 0.8147
Epoch 70/100
100/100 [==============================] - 26s 255ms/step - loss: 0.3844 - acc: 0.8239 - val_loss: 0.2826 - val_acc: 0.8009
Epoch 71/100
100/100 [==============================] - 25s 251ms/step - loss: 0.3807 - acc: 0.8307 - val_loss: 0.4430 - val_acc: 0.7735
Epoch 72/100
100/100 [==============================] - 26s 260ms/step - loss: 0.3808 - acc: 0.8198 - val_loss: 0.2435 - val_acc: 0.8164
Epoch 73/100
100/100 [==============================] - 25s 252ms/step - loss: 0.3695 - acc: 0.8355 - val_loss: 0.2928 - val_acc: 0.8396
Epoch 74/100
100/100 [==============================] - 26s 262ms/step - loss: 0.3845 - acc: 0.8266 - val_loss: 0.3940 - val_acc: 0.8287
Epoch 75/100
100/100 [==============================] - 25s 249ms/step - loss: 0.3816 - acc: 0.8204 - val_loss: 0.3728 - val_acc: 0.7345
Epoch 76/100
100/100 [==============================] - 25s 247ms/step - loss: 0.3707 - acc: 0.8365 - val_loss: 0.5501 - val_acc: 0.7900
Epoch 77/100
100/100 [==============================] - 26s 257ms/step - loss: 0.3659 - acc: 0.8370 - val_loss: 0.7177 - val_acc: 0.8041
Epoch 78/100
100/100 [==============================] - 25s 253ms/step - loss: 0.3660 - acc: 0.8364 - val_loss: 0.3636 - val_acc: 0.8211
Epoch 79/100
100/100 [==============================] - 26s 263ms/step - loss: 0.3589 - acc: 0.8441 - val_loss: 0.7369 - val_acc: 0.8196
Epoch 80/100
100/100 [==============================] - 25s 249ms/step - loss: 0.3549 - acc: 0.8403 - val_loss: 0.4898 - val_acc: 0.8376
Epoch 81/100
100/100 [==============================] - 27s 268ms/step - loss: 0.3559 - acc: 0.8362 - val_loss: 0.5360 - val_acc: 0.8223
Epoch 82/100
100/100 [==============================] - 25s 252ms/step - loss: 0.3660 - acc: 0.8342 - val_loss: 0.2845 - val_acc: 0.8003
Epoch 83/100
100/100 [==============================] - 25s 253ms/step - loss: 0.3560 - acc: 0.8381 - val_loss: 0.4417 - val_acc: 0.8319
Epoch 84/100
100/100 [==============================] - 26s 260ms/step - loss: 0.3612 - acc: 0.8492 - val_loss: 0.5339 - val_acc: 0.7642
Epoch 85/100
100/100 [==============================] - 25s 253ms/step - loss: 0.3589 - acc: 0.8295 - val_loss: 0.4682 - val_acc: 0.8096
Epoch 86/100
100/100 [==============================] - 27s 270ms/step - loss: 0.3407 - acc: 0.8458 - val_loss: 0.2344 - val_acc: 0.8318
Epoch 87/100
100/100 [==============================] - 25s 252ms/step - loss: 0.3416 - acc: 0.8520 - val_loss: 0.5362 - val_acc: 0.7868
Epoch 88/100
100/100 [==============================] - 26s 257ms/step - loss: 0.3525 - acc: 0.8445 - val_loss: 0.2842 - val_acc: 0.8305
Epoch 89/100
100/100 [==============================] - 25s 252ms/step - loss: 0.3437 - acc: 0.8523 - val_loss: 0.3767 - val_acc: 0.8454
Epoch 90/100
100/100 [==============================] - 25s 249ms/step - loss: 0.3410 - acc: 0.8551 - val_loss: 0.4359 - val_acc: 0.8376
Epoch 91/100
100/100 [==============================] - 27s 265ms/step - loss: 0.3308 - acc: 0.8614 - val_loss: 0.4693 - val_acc: 0.8222
Epoch 92/100
100/100 [==============================] - 26s 256ms/step - loss: 0.3391 - acc: 0.8499 - val_loss: 0.2185 - val_acc: 0.8528
Epoch 93/100
100/100 [==============================] - 25s 252ms/step - loss: 0.3470 - acc: 0.8463 - val_loss: 0.4285 - val_acc: 0.7957
Epoch 94/100
100/100 [==============================] - 25s 252ms/step - loss: 0.3271 - acc: 0.8565 - val_loss: 0.7049 - val_acc: 0.8357
Epoch 95/100
100/100 [==============================] - 26s 258ms/step - loss: 0.3227 - acc: 0.8590 - val_loss: 0.6066 - val_acc: 0.7951
Epoch 96/100
100/100 [==============================] - 26s 260ms/step - loss: 0.3323 - acc: 0.8561 - val_loss: 0.6961 - val_acc: 0.8048
Epoch 97/100
100/100 [==============================] - 25s 251ms/step - loss: 0.3219 - acc: 0.8621 - val_loss: 0.5641 - val_acc: 0.8376
Epoch 98/100
100/100 [==============================] - 26s 265ms/step - loss: 0.3180 - acc: 0.8653 - val_loss: 0.5996 - val_acc: 0.8228
Epoch 99/100
100/100 [==============================] - 25s 251ms/step - loss: 0.3349 - acc: 0.8580 - val_loss: 0.7057 - val_acc: 0.8115
Epoch 100/100
100/100 [==============================] - 25s 251ms/step - loss: 0.3390 - acc: 0.8474 - val_loss: 0.4669 - val_acc: 0.8286进行了100次迭代，废了好长的时间，可以看出模型的精确度在不断的增加
代码语言：javascript复制model.save('cats_and_dogs_small_2.h5')# 保存训练好的模型# 再次绘制曲线
import matplotlib.pyplot as plt
acc = history.history['acc']
val_acc = history.history['val_acc']
loss = history.history['loss']
val_loss = history.history['val_loss']
epochs = range(1, len(acc) 1)
plt.plot(epochs, acc, 'bo', label='Training acc')
plt.plot(epochs, val_acc, 'b', label='Validation acc')
plt.title('Training and validation accuracy')
plt.legend()
plt.figure()
plt.plot(epochs, loss, 'bo', label='Training loss')
plt.plot(epochs, val_loss, 'b', label='Validation loss')
plt.title('Training and validation loss')
plt.legend()
plt.show()
在精确度的图中，训练集和验证机的模型精确度都是增加的，最终的精确度为acc: 0.8474 val_acc: 0.8286， 相比较没有进行正则化的模型精确度acc: 0.9855 - val_acc: 0.7160，虽然在训练集上的精确度高于未经正则化，但是在验证集上是显著高于未经正则化的模型的
结束语
随着学习的增加，数据的运算不断的增加，要是有块GPU就好了
love&peace
