神经网络八股功能扩展
第四讲 功能扩展
本讲目标;神经网络八股功能扩展
具体分为:
(1) 自制数据集,解决本领域应用
(2) 数据增强,扩充数据集
(3) 断点续训,存取模型
(4)参数提取,把参数存入文本
(5)acc/loss可视化,查看训练效果
(6) 应用程序,给图识物
一、回顾:
1、tf.keras搭建神经网络八股–六步法
(1) import – 导入所需的各种库和包
(2) x_train, y_train – 导入数据集、自制数据集、数据增强
(3) model = tf.keras.models.sequential()
/class MyModel(Model) model = MyModel–定义模型
(4) model.compile–配置模型
(5) model.fit – 训练模型、断点续训
(6) model.summary – 参数提取、acc/loss可视化,前向推理实现应用
2、 代码 mnist_train_baseline.py:
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()
Epoch 1/5
1875/1875 [==============================] - 2s 800us/step - loss: 0.2627 - sparse_categorical_accuracy: 0.9256 - val_loss: 0.1473 - val_sparse_categorical_accuracy: 0.9564
Epoch 2/5
1875/1875 [==============================] - 1s 722us/step - loss: 0.1136 - sparse_categorical_accuracy: 0.9661 - val_loss: 0.0986 - val_sparse_categorical_accuracy: 0.9708
Epoch 3/5
1875/1875 [==============================] - 1s 720us/step - loss: 0.0782 - sparse_categorical_accuracy: 0.9768 - val_loss: 0.0876 - val_sparse_categorical_accuracy: 0.9740
Epoch 4/5
1875/1875 [==============================] - 1s 718us/step - loss: 0.0593 - sparse_categorical_accuracy: 0.9812 - val_loss: 0.0802 - val_sparse_categorical_accuracy: 0.9755
Epoch 5/5
1875/1875 [==============================] - 1s 722us/step - loss: 0.0441 - sparse_categorical_accuracy: 0.9862 - val_loss: 0.0892 - val_sparse_categorical_accuracy: 0.9743
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
flatten (Flatten) (None, 784) 0
dense (Dense) (None, 128) 100480
dense_1 (Dense) (None, 10) 1290
=================================================================
Total params: 101,770
Trainable params: 101,770
Non-trainable params: 0
_________________________________________________________________
二、 本讲用 tf.keras 完善功能模块
1、 自制数据集,应对特定应用
1.1、 观察数据集数据结构,配成特征标签对
mnist_image_label 文件夹:
四个文件分别对应为训练集图片、训练集标签、测试集图片、测试集标签
图片文件夹:
标签文件:
import tensorflow as tf
from PIL import Image
import numpy as np
import os
path = 'E:\BaiduNetdiskDownload\中国大学MOOCTF笔记2.1共享给所有学习者\class4\MNIST_FC\mnist_image_label\\'
train_path = path + 'mnist_train_jpg_60000/'
train_txt = path + 'mnist_train_jpg_60000.txt'
x_train_savepath = path + 'mnist_x_train.npy'
y_train_savepath = path + 'mnist_y_train.npy'
test_path = path + 'mnist_test_jpg_10000/'
test_txt = path + 'mnist_test_jpg_10000.txt'
x_test_savepath = path + 'mnist_x_test.npy'
y_test_savepath = path + 'mnist_y_test.npy'
# def generateds(图片路径, 标签文件)
def generateds(path, txt):
f = open(txt, 'r')
contents = f.readlines()
f.close()
x,y_ = [], []
for content in contents:
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 / 255.0 # 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) #变成np.array格式
y_ = np.array(y_) #变成np.array格式
y_ = y_.astype(np.int64) #变为64整形
return x, y_ # 返回输入特征x,返回标签y_
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)
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()
-------------Load Datasets-----------------
Epoch 1/5
1875/1875 [==============================] - 2s 805us/step - loss: 0.2624 - sparse_categorical_accuracy: 0.9248 - val_loss: 0.1434 - val_sparse_categorical_accuracy: 0.9582
Epoch 2/5
1875/1875 [==============================] - 1s 751us/step - loss: 0.1173 - sparse_categorical_accuracy: 0.9645 - val_loss: 0.1044 - val_sparse_categorical_accuracy: 0.9687
Epoch 3/5
1875/1875 [==============================] - 1s 735us/step - loss: 0.0810 - sparse_categorical_accuracy: 0.9759 - val_loss: 0.0811 - val_sparse_categorical_accuracy: 0.9738
Epoch 4/5
1875/1875 [==============================] - 1s 755us/step - loss: 0.0604 - sparse_categorical_accuracy: 0.9812 - val_loss: 0.0761 - val_sparse_categorical_accuracy: 0.9756
Epoch 5/5
1875/1875 [==============================] - 1s 733us/step - loss: 0.0474 - sparse_categorical_accuracy: 0.9851 - val_loss: 0.0724 - val_sparse_categorical_accuracy: 0.9762
Model: "sequential_1"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
flatten_1 (Flatten) (None, 784) 0
dense_2 (Dense) (None, 128) 100480
dense_3 (Dense) (None, 10) 1290
=================================================================
Total params: 101,770
Trainable params: 101,770
Non-trainable params: 0
_________________________________________________________________
2、 数据增强,增大数据量
2.1、 数据增强(增大数据量)
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 = 所有数据将乘以相同的值
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]
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%]
)
2.2、 数据增强可视化 (代码 show_augmented _images.py)
# 代码 mnist_train_ex2.py:
import tensorflow as tf
import matplotlib.pyplot as plt
from tensorflow.keras.preprocessing.image import ImageDataGenerator
import numpy as np
mnist = tf.keras.datasets.mnist
(x_train, y_train), (x_test, y_test) = mnist.load_data()
x_train = x_train.reshape(x_train.shape[0], 28, 28, 1)
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)
print("xtrain",x_train.shape)
x_train_subset1 = np.squeeze(x_train[:12])
print("xtrain_subset1",x_train_subset1.shape)
print("xtrain",x_train.shape)
x_train_subset2 = x_train[:12] # 一次显示12张图片
print("xtrain_subset2",x_train_subset2.shape)
fig = plt.figure(figsize=(20,2))
plt.set_cmap('gray')
#显示原始图片
for i in range(0, len(x_train_subset1)):
ax = fig.add_subplot(1,12,i+1)
ax.set_xticklabels([])
ax.set_yticklabels([])
ax.imshow(x_train_subset1[i])
fig.suptitle('Subset of Original Training Images', fontsize=20)
plt.show()
# 显示增强后的图片
fig = plt.figure(figsize=(20,2))
for x_batch in image_gen_train.flow(x_train_subset2, batch_size=12, shuffle=False):
for i in range(0,12):
ax = fig.add_subplot(1, 12, i+1)
ax.set_xticklabels([])
ax.set_yticklabels([])
ax.imshow(np.squeeze(x_batch[i]))
fig.suptitle('Augmented Images', fontsize=20)
plt.show()
break
xtrain (60000, 28, 28, 1)
xtrain_subset1 (12, 28, 28)
xtrain (60000, 28, 28, 1)
xtrain_subset2 (12, 28, 28, 1)
代码 mnist_train_ex2.py:
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()
Epoch 1/5
1872/1875 [============================>.] - ETA: 0s - loss: 1.4291 - sparse_categorical_accuracy: 0.5355WARNING:tensorflow:Model was constructed with shape (None, None, None, None) for input KerasTensor(type_spec=TensorSpec(shape=(None, None, None, None), dtype=tf.float32, name='flatten_input'), name='flatten_input', description="created by layer 'flatten_input'"), but it was called on an input with incompatible shape (None, 28, 28).
1875/1875 [==============================] - 13s 7ms/step - loss: 1.4285 - sparse_categorical_accuracy: 0.5358 - val_loss: 0.4884 - val_sparse_categorical_accuracy: 0.8748
Epoch 2/5
1875/1875 [==============================] - 12s 7ms/step - loss: 0.9640 - sparse_categorical_accuracy: 0.7053 - val_loss: 0.3590 - val_sparse_categorical_accuracy: 0.8994
Epoch 3/5
1875/1875 [==============================] - 13s 7ms/step - loss: 0.8529 - sparse_categorical_accuracy: 0.7433 - val_loss: 0.2719 - val_sparse_categorical_accuracy: 0.9287
Epoch 4/5
1875/1875 [==============================] - 13s 7ms/step - loss: 0.7817 - sparse_categorical_accuracy: 0.7643 - val_loss: 0.2645 - val_sparse_categorical_accuracy: 0.9284
Epoch 5/5
1875/1875 [==============================] - 13s 7ms/step - loss: 0.7268 - sparse_categorical_accuracy: 0.7808 - val_loss: 0.2317 - val_sparse_categorical_accuracy: 0.9351
Model: "sequential"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
flatten (Flatten) (None, None) 0
dense (Dense) (None, 128) 100480
dense_1 (Dense) (None, 10) 1290
=================================================================
Total params: 101,770
Trainable params: 101,770
Non-trainable params: 0
_________________________________________________________________
3、断点续训,存取模型
3.1 读取模型
load_weights(路径文件名)
例如:
checkpoint_save_path = './cheakpoint/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。
代码p18_mnist_train_ex2.py:
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 = './cheakpoint/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()
Epoch 1/5
1875/1875 [==============================] - 2s 798us/step - loss: 0.2644 - sparse_categorical_accuracy: 0.9248 - val_loss: 0.1471 - val_sparse_categorical_accuracy: 0.9566
Epoch 2/5
1875/1875 [==============================] - 2s 829us/step - loss: 0.1175 - sparse_categorical_accuracy: 0.9650 - val_loss: 0.1162 - val_sparse_categorical_accuracy: 0.9636
Epoch 3/5
1875/1875 [==============================] - 1s 786us/step - loss: 0.0801 - sparse_categorical_accuracy: 0.9757 - val_loss: 0.0957 - val_sparse_categorical_accuracy: 0.9696
Epoch 4/5
1875/1875 [==============================] - 2s 972us/step - loss: 0.0615 - sparse_categorical_accuracy: 0.9812 - val_loss: 0.0805 - val_sparse_categorical_accuracy: 0.9748
Epoch 5/5
1875/1875 [==============================] - 2s 905us/step - loss: 0.0472 - sparse_categorical_accuracy: 0.9854 - val_loss: 0.0737 - val_sparse_categorical_accuracy: 0.9781
Model: "sequential_2"
_________________________________________________________________
Layer (type) Output Shape Param #
=================================================================
flatten_2 (Flatten) (None, 784) 0
dense_4 (Dense) (None, 128) 100480
dense_5 (Dense) (None, 10) 1290
=================================================================
Total params: 101,770
Trainable params: 101,770
Non-trainable params: 0
_________________________________________________________________
4、参数提取,写至文本
4.1、提取可训练参数
model.trainable_variables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
)
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]
注:precision=np.inf打印完整小数位;threshold=np.nan打印全部数组元素。
代码p19_mnist_train_ex4.py:
import tensorflow as tf
import os
import numpy as np
#设置显示全部内容,inf表示无穷大
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']
)
path = 'E:\BaiduNetdiskDownload\中国大学MOOCTF笔记2.1共享给所有学习者\class4\MNIST_FC\mnist_image_label\\'
checkpoint_save_path = 'cheakpoint/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()
#打印模型参数,存入文本
print(model.trainable_variables)
file = open(path+ 'weight.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loss
val_loss:测试集loss
sparse_categorical_accuracy:训练集准确率
val_sparse_categorical_accuracy:测试集准确率
代码p20_mnist_trian_ex5.py
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']
)
path = 'E:\BaiduNetdiskDownload\中国大学MOOCTF笔记2.1共享给所有学习者\class4\MNIST_FC\mnist_image_label\\'
checkpoint_save_path = 'cheakpoint/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(path + 'weight.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()
# 显示训练集和验证集的acc和loss曲线
# history中读取所需数据
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、应用程序,给图识物
6.1、给图识物
手写十个数,正确率90%以上合格
6.2、前向传播执行应用
predict(输入数据, batch_size=整数) 返回前向传播计算结果
注:predict参数详解。(1)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);(2)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;(3)verbose: 日志显示模式,0或1;(4)steps: 声明预测结束之前的总步数(批次样本),默认值 None;(5)返回:预测的 Numpy 数组(或数组列表)。
模型预测简单三步:
import tensorflow as tf
import os
import numpy as np
from PIL import Image
model_save_path = './checkpoint/mnist.ckpt'
model = tf.keras.models.Sequential([
tf.keras.layers.Flatten(input_shape=(28,28)),
tf.keras.layers.Dense(128, activation='relu'),
tf.keras.layers.Dense(10, activation = 'softmax')
])
# 加载模型
model.load_weights(model_save_path)
# 预测图片数量
preNum = int(input("input the number of test pictures: "))
for i in range(preNum):
# 预测图片打开路径
image_path = input("the path of test picture: ")
img = Image.open(image_path)
# 调整尺寸和类型
img = img.resize((28,28), Image.ANTIALIAS)
img_arr = np.array(img.convert('L'))
# 二值化
for i in range(28):
for j in range(28):
if img_arr[i][j] < 200:
img_arr[i][j] = 255
else:
img_arr[i][j] = 0
img_arr = img_arr / 255.0
x_predict = img_arr[tf.newaxis,...]
# 预测
result = model.predict(x_predict)
pred = tf.argmax(result, axis = 1)
print('\n')
tf.print(pred)