0. Baseline
tf.keras搭建神经网络八股:六步法,实践鸢尾花分类、MNIST手写数字识别、Fashion-mnist数据集
- 第一步:import所需库
- 第二步:导入数据集。拓展:自制数据集、数据增强
- 第三步:逐层搭建神经网络
- 第四步:配置训练方法
- 第五步:执行训练过程。拓展:断点续训
- 第六步:打印网络结构。拓展:参数提取,acc/loss可视化,前向推理实现与应用
#------
import tensorflow as tf
#------
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
#------
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
#------
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['sparse_categorical_accuracy'])
#------
model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1)
#------
model.summary()
1. 自制数据集,解决本领域应用
1.1观察数据集数据结构,配成特征标签对
四个文件从上到下分别为:测试集图片文件、训练集图片文件、测试集标签、训练集标签
图片文件夹、标签文件的内容分别为:
1.2 定义函数generateds(图片路径, 标签文件)生成数据集
def generateds(path, txt):
f = open(txt, 'r') # 以只读形式打开txt文件
contents = f.readlines() # 读取文件中所有行
f.close() # 关闭txt文件
x, y_ = [], [] # 建立空列表
for content in contents: # 逐行取出
value = content.split() # 以空格分开,图片路径为value[0] , 标签为value[1] , 存入列表
img_path = path + value[0] # 拼出图片路径和文件名
img = Image.open(img_path) # 读入图片
img = np.array(img.convert('L')) # 图片变为8位宽灰度值的np.array格式
img = img / 255. # 数据归一化 (实现预处理)
x.append(img) # 归一化后的数据,贴到列表x
y_.append(value[1]) # 标签贴到列表y_
print('loading : ' + content) # 打印状态提示
x = np.array(x) # 变为np.array格式
y_ = np.array(y_) # 变为np.array格式
y_ = y_.astype(np.int64) # 变为64位整型
return x, y_ # 返回输入特征x,返回标签y_
convert()
是图像实例对象的一个方法,接受一个 mode 参数,用以指定一种色彩模式。
mode参数 | 色彩模式 |
---|---|
‘1’ | 1位像素,黑白,每字节一个像素存储 |
‘L’ | 8位像素,黑白 |
‘P’ | 8位像素,使用调色板映射到任何其他模式 |
‘RGB’ | 3 × 8 3\times8 3×8位像素,真彩色 |
‘RGBA’ | 4 × 8 4\times8 4×8位像素,带透明度掩模的真彩色 |
‘CMYK’ | 4 × 8 4\times8 4×8位像素,分色 |
‘YCbCr’ | 3 × 8 3\times8 3×8位像素,彩色视频格式 |
‘I’ | 32位有符号整数像素 |
‘F’ | 32位浮点像素 |
1.3 完整代码
#---导入库---
import tensorflow as tf
from PIL import Image
import numpy as np
import os
train_path = './mnist_image_label/mnist_train_jpg_60000/'
train_txt = './mnist_image_label/mnist_train_jpg_60000.txt'
x_train_savepath = './mnist_image_label/mnist_x_train.npy'
y_train_savepath = './mnist_image_label/mnist_y_train.npy'
test_path = './mnist_image_label/mnist_test_jpg_10000/'
test_txt = './mnist_image_label/mnist_test_jpg_10000.txt'
x_test_savepath = './mnist_image_label/mnist_x_test.npy'
y_test_savepath = './mnist_image_label/mnist_y_test.npy'
#---定义生成数据集的函数---
def generateds(path, txt):
f = open(txt, 'r') # 以只读形式打开txt文件
contents = f.readlines() # 读取文件中所有行
f.close() # 关闭txt文件
x, y_ = [], [] # 建立空列表
for content in contents: # 逐行取出
value = content.split() # 以空格分开,图片路径为value[0] , 标签为value[1] , 存入列表
img_path = path + value[0] # 拼出图片路径和文件名
img = Image.open(img_path) # 读入图片
img = np.array(img.convert('L')) # 图片变为8位宽灰度值的np.array格式
img = img / 255. # 数据归一化 (实现预处理)
x.append(img) # 归一化后的数据,贴到列表x
y_.append(value[1]) # 标签贴到列表y_
print('loading : ' + content) # 打印状态提示
x = np.array(x) # 变为np.array格式
y_ = np.array(y_) # 变为np.array格式
y_ = y_.astype(np.int64) # 变为64位整型
return x, y_ # 返回输入特征x,返回标签y_
#---如果文件已经存在则直接载入数据集,若不存在则先生成数据集再保存---
if os.path.exists(x_train_savepath) and os.path.exists(y_train_savepath) and os.path.exists(
x_test_savepath) and os.path.exists(y_test_savepath):
print('-------------Load Datasets-----------------')
x_train_save = np.load(x_train_savepath)
y_train = np.load(y_train_savepath)
x_test_save = np.load(x_test_savepath)
y_test = np.load(y_test_savepath)
x_train = np.reshape(x_train_save, (len(x_train_save), 28, 28))
x_test = np.reshape(x_test_save, (len(x_test_save), 28, 28))
else:
print('-------------Generate Datasets-----------------')
x_train, y_train = generateds(train_path, train_txt)
x_test, y_test = generateds(test_path, test_txt)
print('-------------Save Datasets-----------------')
x_train_save = np.reshape(x_train, (len(x_train), -1))
x_test_save = np.reshape(x_test, (len(x_test), -1))
np.save(x_train_savepath, x_train_save)
np.save(y_train_savepath, y_train)
np.save(x_test_savepath, x_test_save)
np.save(y_test_savepath, y_test)
#---搭建网络结构---
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
#---配置训练方法--
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['sparse_categorical_accuracy'])
#---执行训练过程---
model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1)
#---打印网络结构---
model.summary()
2. 数据增强,扩充数据集
数据增强可以帮助扩展数据集,对图像的增强就是对图像的简单形变,用来应对因拍照角度不同引起的图片变形。
image_gen_train=tf.keras.preprocessing.image.ImageDataGenerator(增强方法)
image_gen_train.fit(x_train)
常用增强方法:
- 缩放系数:
rescale
=所有数据将乘以提供的值 - 随机旋转:
rotation_range
=随机旋转角度数范围 - 宽度偏移:
width_shift_range
=随机宽度偏移量 - 高度偏移:
height_shift_range
=随机高度偏移量 - 水平翻转:
horizontal_flip
=是否水平随机翻转 - 随机缩放:
zoom_range
=随机缩放的范围 [1-n,1+n]
举例来说,
image_gen_train = ImageDataGenerator(
rescale=1./255, #原像素值0~255归至0~1
rotation_range=45, #随机45度旋转
width_shift_range=.15, #随机宽度偏移 [-0.15,0.15)
height_shift_range=.15, #随机高度偏移 [-0.15,0.15)
horizontal_flip=True, #随机水平翻转
zoom_range=0.5) #随机缩放到 [1-50%,1+50%]
image_gen_train.fit(x_train)
2.1 完整代码
#------
import tensorflow as tf
from tensorflow.keras.preprocessing.image import ImageDataGenerator
#------
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1) # 给数据增加一个维度,从(60000, 28, 28)reshape为(60000, 28, 28, 1)
image_gen_train = ImageDataGenerator(
rescale=1. / 1., # 如为图像,分母为255时,可归至0~1
rotation_range=45, # 随机45度旋转
width_shift_range=.15, # 宽度偏移
height_shift_range=.15, # 高度偏移
horizontal_flip=False, # 水平翻转
zoom_range=0.5 # 将图像随机缩放阈量50%
)
image_gen_train.fit(x_train)
#------
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
#------
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['sparse_categorical_accuracy'])
#------
model.fit(image_gen_train.flow(x_train, y_train, batch_size=32), epochs=5, validation_data=(x_test, y_test),
validation_freq=1)
#------
model.summary()
注:
- 数据增强函数的输入要求是4维,通过reshape调整:
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
model.fit(x_train,y_train,batch_size=32,……)
变为model.fit(image_gen_train.flow(x_train, y_train,batch_size=32), ……)
3. 断点续训,存取模型
3.1 读取模型
- 生成ckpt文件时会同步生成索引表,通过判断是否有索引表就可判断是否保存过模型参数了;
- 如果有索引表,就可以调用load_weights()函数读取参数。
checkpoint_save_path="./checkpoint/mnist.ckpt"
if os.path.exists(checkpoint_save_path+'.index'):
print('----------load the model----------')
model.load_weights(checkpoint_save_path)
3.2 保存模型
- 借助tensorflow给出的回调函数,直接保存参数和网络.
tf.keras.callbacks.ModelCheckpoint(
filepath=路径文件名,
save_weights_only=True,
monitor='val_loss', # val_loss or loss
save_best_only=True)
history = model.fit(x_train, y_train, batch_size=32, epochs=5,
validation_data=(x_test, y_test), validation_freq=1,
callbacks=[cp_callback])
注:monitor
配合save_best_only
可以保存最优模型,包括:训练损失最小模型、测试损失最小模型、训练准确率最高模型、测试准确率最高模型等。
3.3 完整代码
#----------
import tensorflow as tf
import os
#----------
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
#----------
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
#----------
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['sparse_categorical_accuracy'])
#----------
checkpoint_save_path = "./checkpoint/mnist.ckpt"
if os.path.exists(checkpoint_save_path + '.index'):
print('-------------load the model-----------------')
model.load_weights(checkpoint_save_path)
#----------
cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_save_path,
save_weights_only=True,
save_best_only=True)
history = model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1,
callbacks=[cp_callback])
#----------
model.summary()
4. 参数提取,把参数存入文本
4.1 提取可训练参数
model.trainable_variables
:返回模型中可训练的参数
4.2 设置print输出格式
np.set_printoptions(precision=小数点后按四舍五入保留几位,threshold=数组元素数量少于或等于门槛值,打印全部元素;否则打印门槛值+1个元素,中间用省略号补充)
precision=np.inf
打印完整小数位;threshold=np.nan
打印全部数组元素。
示例:
>>> np.set_printoptions(precision=5)
>>> print(np.array([1.123456789]))
[1.12346]
>>> np.set_printoptions(threshold=5)
>>> print(np.arange(10))
[0 1 2 … , 7 8 9]
4.3 完整代码
#----------
import tensorflow as tf
import os
import numpy as np
np.set_printoptions(threshold=np.inf)
#----------
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
#----------
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
#----------
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['sparse_categorical_accuracy'])
#----------
checkpoint_save_path = "./checkpoint/mnist.ckpt"
if os.path.exists(checkpoint_save_path + '.index'):
print('-------------load the model-----------------')
model.load_weights(checkpoint_save_path)
cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_save_path,
save_weights_only=True,
save_best_only=True)
history = model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1,
callbacks=[cp_callback])
#----------
model.summary()
print(model.trainable_variables)
file = open('./weights.txt', 'w')
for v in model.trainable_variables:
file.write(str(v.name) + '\n')
file.write(str(v.shape) + '\n')
file.write(str(v.numpy()) + '\n')
file.close()
5.acc/loss可视化,查看训练效果
5.1 acc曲线与loss曲线
history=model.fit(训练集数据, 训练集标签, batch_size=, epochs=,
validation_split=用作测试数据的比例,validation_data=测试集,
validation_freq=测试频率)
history:
loss
:训练集lossval_loss
:测试集losssparse_categorical_accuracy
:训练集准确率val_sparse_categorical_accuracy
:测试集准确率
5.2 完整代码
import tensorflow as tf
import os
import numpy as np
from matplotlib import pyplot as plt
np.set_printoptions(threshold=np.inf)
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train, x_test = x_train / 255.0, x_test / 255.0
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')
])
model.compile(optimizer='adam',
loss=tf.keras.losses.SparseCategoricalCrossentropy(from_logits=False),
metrics=['sparse_categorical_accuracy'])
checkpoint_save_path = "./checkpoint/mnist.ckpt"
if os.path.exists(checkpoint_save_path + '.index'):
print('-------------load the model-----------------')
model.load_weights(checkpoint_save_path)
cp_callback = tf.keras.callbacks.ModelCheckpoint(filepath=checkpoint_save_path,
save_weights_only=True,
save_best_only=True)
history = model.fit(x_train, y_train, batch_size=32, epochs=5, validation_data=(x_test, y_test), validation_freq=1,
callbacks=[cp_callback])
model.summary()
print(model.trainable_variables)
file = open('./weights.txt', 'w')
for v in model.trainable_variables:
file.write(str(v.name) + '\n')
file.write(str(v.shape) + '\n')
file.write(str(v.numpy()) + '\n')
file.close()
########### show #############
# 显示训练集和验证集的acc和loss曲线
acc = history.history['sparse_categorical_accuracy']
val_acc = history.history['val_sparse_categorical_accuracy']
loss = history.history['loss']
val_loss = history.history['val_loss']
plt.subplot(1, 2, 1)
plt.plot(acc, label='Training Accuracy')
plt.plot(val_acc, label='Validation Accuracy')
plt.title('Training and Validation Accuracy')
plt.legend()
plt.subplot(1, 2, 2)
plt.plot(loss, label='Training Loss')
plt.plot(val_loss, label='Validation Loss')
plt.title('Training and Validation Loss')
plt.legend()
plt.show()
6. 前向传播执行应用程序
predict(输入数据, batch_size=整数)
:返回前向传播计算结果
- x:输入数据,Numpy 数组(或者 Numpy 数组的列表,如果模型有多个输出);* batch_size:整数,由于GPU的特性,batch_size最好选用8,16,32,64……,如果未指定,默认为32;
- verbose: 日志显示模式,0或1;
- steps: 声明预测结束之前的总步数(批次样本),默认值 None;
- 返回:预测的 Numpy 数组(或数组列表)。
6.1 完整代码1
from PIL import Image
import numpy as np
import tensorflow as tf
model_save_path = './checkpoint/mnist.ckpt'
#复现模型(前向传播)
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')])
#加载参数
model.load_weights(model_save_path)
preNum = int(input("input the number of test pictures:"))
#读入待识别的图片
for i in range(preNum):
image_path = input("the path of test picture:")
img = Image.open(image_path)
img = img.resize((28, 28), Image.ANTIALIAS)
img_arr = np.array(img.convert('L'))
#预处理:白底黑字图片(测试集)转换为黑底白字图片(训练集)
img_arr = 255 - img_arr
img_arr = img_arr / 255.0
print("img_arr:",img_arr.shape)
x_predict = img_arr[tf.newaxis, ...]
print("x_predict:",x_predict.shape)
result = model.predict(x_predict)#预测结果
pred = tf.argmax(result, axis=1)
print('\n')
tf.print(pred)
注:
1、输出结果pred是张量,需要用tf.print,print打印出来是tf.Tensor([1], shape=(1,), dtype=int64);
2、去掉二值化,出现无法收敛问题,需要对数据集进行归一化。
6.2 完整代码2
图片预处理:将输入图片转换为只有白色和黑色的高对比图片:保留图片有效信息的同时,滤去了噪声。
from PIL import Image
import numpy as np
import tensorflow as tf
model_save_path = './checkpoint/mnist.ckpt'
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation='softmax')])
model.load_weights(model_save_path)
preNum = int(input("input the number of test pictures:"))
for i in range(preNum):
image_path = input("the path of test picture:")
img = Image.open(image_path)
img = img.resize((28, 28), Image.ANTIALIAS)
img_arr = np.array(img.convert('L'))
for i in range(28):
for j in range(28):
#输入图片转换为只有白色和黑色的高对比图片:保留图片有效信息的同时,滤去了噪声
if img_arr[i][j] < 200:
img_arr[i][j] = 255
else:
img_arr[i][j] = 0
img_arr = img_arr / 255.0#归一化。维度为(28,28)
x_predict = img_arr[tf.newaxis, ...]#维度为(1,28,28),由于训练网络是按batch送入数据,所以需要转换
result = model.predict(x_predict)
pred = tf.argmax(result, axis=1)
print('\n')
tf.print(pred)