#tensorflow个人学习笔记(一)
#模块化神经网络的基本结构
###一、生成数据集
opt4_8_generateds.py
#coding:utf-8
#模块化神经网络的结构分为三部分,第一部分生成模拟数据集,在实际应用中,应该是整理数据集
#生成模拟数据集
import numpy as np #引入np模块
import matplotlib.pyplot as plt #引入plt模块
seed = 2 #设定随机种子
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(x0,x1)作为输入数据集
X = rdm.randn(300,2)
#从X这个300行2列的数据集中取出一行,判断两坐标平方和小于2则给Y赋值为1,其余为0
#最为输入数据集的标签(正确答案)
Y_ = [int(x0*x0 + x1*x1 < 2) for (x0,x1) in X] #Y_是一个列表,元素是300个1或0
#遍历Y中的每一个元素,1赋值'red',其余赋值'blue',这样可视化显示时,人可以肉眼直观区分
Y_c= [['red' if y else 'blue'] for y in Y_] #Y_c是一个,列表,其中的元素是['blue']或['red']
#对数据集X和标签Y进行形状整理,第一个元素为行,第二个元素表示多少列,显然X为两列,Y为一列
#-1表示未知
X = np.vstack(X ).reshape(-1,2) #先纵向堆放,然后在行数未知的情况下,把X整理成2列
Y_ = np.vstack(Y_).reshape(-1,1) #先纵向堆放,然后在行数未知的情况下,把Y_整理出1列
return X,Y_,Y_c #X是300行2列的矩阵,Y_是一个1、0列表,Y_c是['blue'],['red']列表
#print X
#print Y_
#print Y_c
#用plt.scatter画出数据集X各行第0列元素和第1列元素的点即各行的(x0,x1),用各行Y_c对应的值表示颜色(c是colcour)
#plt.scatter(X[:,0],X[:,1],c=np.squeeze(Y_c))
#plt.show()
#vstack 是为了沿数数值方向将矩阵堆叠,列数必须一致
#reshape是在不改变数值的前提下改变矩阵的形状 -1表示未知、缺省
###二、定义前向传播
opt4_8_forward.py
#coding:utf-8
#定义神经网络输入、参数和输出,定义前向传播过程
import tensorflow as tf #引入tf模块
def get_weight(shape,regularizer): #定义权重参数
w = tf.Variable(tf.random_normal(shape),dtype=tf.float32)
tf.add_to_collection('losses',tf.contrib.layers.l2_regularizer(regularizer)(w))
return w
def get_bias(shape):
b = tf.Variable(tf.constant(0.01, shape=shape))
return b
def forward(x,regularizer): #这个函数构建了神经网络结构
w1 = get_weight([2,11],regularizer)
b1 = get_bias([11])
y1 = tf.nn.relu(tf.matmul(x,w1)+b1) #激活函数,使矩阵的每一行的小于0的元素置0
w2 = get_weight([11,1],regularizer)
b2 = get_bias([1])
y = tf.matmul(y1,w2) + b2 #输出层不激活
return y
###三、定义反向传播过程
opt4_8_backward.py
#coding:utf-8
#引入各个模块
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
import opt4_8_generateds
import opt4_8_forward
#定义超参数 超参数是指学习之前定义的参数,而不是学习时运算得到的参数
STEPS = 40000 #训练40000轮
BATCH_SIZE = 30 #每次给神经网络喂入30个数据
LEARNING_RATE_BASE = 0.001
LEARNING_RATE_DECAY= 0.999
REGULARIZER = 0.01
def backward():
x = tf.placeholder(tf.float32, shape=(None,2))
y_= tf.placeholder(tf.float32, shape=(None,1))
X,Y_,Y_c = opt4_8_generateds.generateds()
y = opt4_8_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,使用均方误差
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)
with tf.Session() as sess:
init_op = tf.global_variables_initializer() #初始化所有变量
sess.run(init_op)
for i in range(STEPS): #使用for循环迭代STEPS轮
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: #每2k轮打印一次loss值
loss_v = sess.run(loss_total, feed_dict={x:X,y_:Y_})
print("After %d steps, loss is : %f" %(i, loss_v))
xx,yy = np.mgrid[-3:3:.01, -3:3:.01] #在-3与+3之间,以0.01为间距,生成网格坐标点,组成xx,yy坐标集
grid = np.c_[xx.ravel(), yy.ravel()]
probs= sess.run(y, feed_dict={x:grid}) #把xx,yy坐标集命名为grid,并喂入神经网络
probs= probs.reshape(xx.shape) #整理形状,命名为probs
plt.scatter(X[:,0], X[:,1], c=np.squeeze(Y_c)) #画出离散点
plt.contour(xx, yy, probs, levels = [.5]) #画出probs为0.5的曲线
plt.show() #展示图表
if __name__ == '__main__' :
backward()
###输出结果
After 0 steps, loss is : 15.128757
After 2000 steps, loss is : 0.263676
After 4000 steps, loss is : 0.181108
After 6000 steps, loss is : 0.145825
After 8000 steps, loss is : 0.116128
After 10000 steps, loss is : 0.099375
After 12000 steps, loss is : 0.095520
After 14000 steps, loss is : 0.095004
After 16000 steps, loss is : 0.093543
After 18000 steps, loss is : 0.092141
After 20000 steps, loss is : 0.091536
After 22000 steps, loss is : 0.091219
After 24000 steps, loss is : 0.091075
After 26000 steps, loss is : 0.090997
After 28000 steps, loss is : 0.090936
After 30000 steps, loss is : 0.090890
After 32000 steps, loss is : 0.090851
After 34000 steps, loss is : 0.090831
After 36000 steps, loss is : 0.090816
After 38000 steps, loss is : 0.090807
/home/lixiang/.local/lib/python2.7/site-packages/numpy/ma/core.py:6447: MaskedArrayFutureWarning: In the future the default for ma.maximum.reduce will be axis=0, not the current None, to match np.maximum.reduce. Explicitly pass 0 or None to silence this warning.
return self.reduce(a)
/home/lixiang/.local/lib/python2.7/site-packages/numpy/ma/core.py:6447: MaskedArrayFutureWarning: In the future the default for ma.minimum.reduce will be axis=0, not the current None, to match np.minimum.reduce. Explicitly pass 0 or None to silence this warning.
return self.reduce(a)