TensorFlow神经网络：模块化的神经网络八股
1、前向传播：
- 搭建从输入到输出的网络结构
forward.py:
# 定义前向传播过程
def forward(x, regularizer):
w =
b =
y =
return y
# 给w赋初值，并把w的正则化损失加到总损失中
def get_weight(shape, regularizer):
w = tf.Variable()
tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w))
return w
def get_bias(shape)
b = tf.Variable()
return b
2、反向传播
- 训练网络，优化网络参数，提高模型准确性
backward.py：
# 定义反向传播
def backward():
# 对数据集x和标准答案y_占位
x = tf.placeholder()
y_ = tf.placeholder(
)
# 利用forward模块复现前向传播网络的结构，计算得到y
y = forward.forward(x, REGULARIZER)
# 定义轮数计数器
global_step = tf.Variable(0, trainable = False)
# 定义损失函数
loss =
'''
# 均方误差
loss = tf.reduce_mean(tf.square(y - y_))
# 交叉熵
ce = tf.nn.sparse_softmax_cross_entropy_with_logits(logits = y, lables = tf.argmax(y_, 1))
loss = tf.reduce_mean(ce)
'''
# 在训练网络模型时
# 常常将1正则化、2指数衰减学习率、3滑动平均这三个方法作为优化模型的方法
'''
# 使用正则化时的损失函数
loss = loss(y, y_) + tf.add_n(tf.get_collection('losses'))
# 使用指数衰减的学习率时，加上：
learning_rate = tf.train.exponential_decay(
LEARNING_RATE_BASE,
global_step,
数据集总样本数/BATCH_SIZE,
LEARNING_RATE_DECAY,
staircase = True)
'''
# 上面的损失函数和学习率选好之后，定义反向传播过程使用梯度下降
train_step = tf.train.GradientDescentOptimizer(learning_rate).minimize(loss, global_step = global_step)
# 如果使用滑动平均时，加上：
'''
ema = tf.train.ExponentialMovingAverage(MOVING_AVERAGE_DECAY, global_step)
ema_op = ema.apply(tf.trainable_variables())
with tf.control_dependencies([train_step, ema_op]):
train_op = tf.no_op(name = 'train')
'''
# 训练过程
with tf.Session() as sess:
# 初始化所有参数
init_op = tf.global_variables_initializer()
sess.run(init_op)
# 循环迭代
for i in range(STEPS):
# 每轮调用sess.run执行训练过程train_step
sess.run(train_step, feed_dict = {x: , y_: })
# 每运行一定轮数，打印出当前的loss信息
if i % 轮数==0
print
3、判断主文件
# 判断python运行文件是否为主文件，如果是，则执行
if __name__ == '__main__':
backward()
4、实例模块化展示
- 加入指数衰减学习率–优化效率
- 加入正则化–提高泛化性能
- 模块化设计
①generateds.py
# modelNN_generateds.py
# 数据导入模块，生成模拟数据集
# coding: utf-8
import numpy as np
import matplotlib.pyplot as plt
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2' #hide warnings
seed = 2
def generateds():
# 基于seed产生随机数
rdm = np.random.RandomState(seed)
# 随机数返回300行2列的矩阵，表示300组坐标点，作为输入数据集
X = rdm.randn(300, 2)
# 手工标注数据分类
Y_ = [int(x0*x0 + x1*x1 < 2)for (x0, x1) in X]
# Y_为1，标记红色，否则蓝色
Y_c = [['red' if y else 'blue'] for y in Y_]
# 对数据集和标签进行reshape， X整理为n行2列，Y为n行1列，第一个元素-1表示n行
X = np.vstack(X).reshape(-1, 2)
Y_ = np.vstack(Y_).reshape(-1, 1)
return X, Y_, Y_c
print("X:\n")
print(X)
print("Y_:\n")
print(Y_)
print("Y_c:\n")
print(Y_c)
② forward.py
# modelNN_generateds.py
# 前向传播模块
# 定义神经网络的输入、参数和输出，定义前向传播过程
# coding: utf-8
import tensorflow as tf
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2' #hide warnings
# 给w赋初值，并把w的正则化损失加到总损失中
def get_weight(shape, regularizer):
w = tf.Variable(tf.random_normal(shape), dtype = tf.float32)
tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w))
return w
# 给b赋初值
def get_bias(shape):
b = tf.Variable(tf.constant(0.01, shape = shape))
return b
def forward(x, regularizer):
w1 = get_weight([2, 11], regularizer)
b1 = get_bias([11])
y1 = tf.nn.relu(tf.matmul(x, w1) + b1)
w2 = get_weight([11, 1], regularizer)
b2 = get_bias([1])
y = tf.matmul(y1, w2) + b2 #输出层不通过激活函数
return y
③ backward.py
# modelNN_generateds.py
# 反向传播模块
# 定义神经网络的反向传播过程
# coding: utf-8
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
import modelNN_generateds
import modelNN_forward
import os
os.environ['TF_CPP_MIN_LOG_LEVEL']='2' #hide warnings
# 定义超参数
STEPS = 40000 #训练轮数
BATCH_SIZE = 30
LEARNING_RATE_BASE = 0.001 #初始学习率
LEARNING_RATE_DECAY = 0.999 # 学习率衰减率
REGULARIZER = 0.01 # 正则化参数
def backward():
# placeholder占位
x = tf.placeholder(tf.float32, shape = (None, 2))
y_ = tf.placeholder(tf.float32, shape = (None, 1))
# 生成数据集
X, Y_, Y_c = modelNN_generateds.generateds()
# 前向传播推测输出y
y = modelNN_forward.forward(x, REGULARIZER)
# 定义global_step
global_step = tf.Variable(0, trainable = False)
# 定义指数衰减学习率
learning_rate = tf.train.exponential_decay(
LEARNING_RATE_BASE,
global_step,
300/BATCH_SIZE,
LEARNING_RATE_DECAY,
staircase = True)
# 定义损失函数
loss_mse = tf.reduce_mean(tf.square(y - y_))
loss_total = loss_mse + tf.add_n(tf.get_collection('losses'))
# 定义反向传播方法：包含正则化
train_step = tf.train.AdamOptimizer(learning_rate).minimize(loss_total)
# 定义训练过程
with tf.Session() as sess:
init_op = tf.global_variables_initializer()
sess.run(init_op)
for i in range(STEPS):
start = (i * BATCH_SIZE) % 300
end = start + BATCH_SIZE
sess.run(train_step, feed_dict = {x: X[start:end], y_:Y_[start:end]})
if i % 2000==0:
loss_v = sess.run(loss_total, feed_dict = {x: X, y_: Y_})
print("after %d steps, loss for total is %f" %(i, loss_v))
xx, yy = np.mgrid[-3:3:.01, -3:3:.01]
grid = np.c_[xx.ravel(), yy.ravel()]
probs = sess.run(y, feed_dict = {x: grid})
probs = probs.reshape(xx.shape)
# 可视化
plt.scatter(X[:, 0], X[:, 1], c = np.squeeze(Y_c))
# 给probs值为0.5的所有点(xx, yy)上色
plt.contour(xx, yy, probs, levels = [.5])
plt.show()
# 判断python运行文件是否为主文件，如果是，则执行
if __name__ == '__main__':
backward()
