权重优化 正则化

W增加权重

l1   权重求和

l2 权重平方求和

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

x = tf.placeholder(tf.float32, shape=(None, 2))
y_ = tf.placeholder(tf.float32, shape=(None, 1))

w1 = get_weight([2,11], 0.01) 
b1 = get_bias([11])
y1 = tf.nn.relu(tf.matmul(x, w1)+b1)

w2 = get_weight([11,1], 0.01)
b2 = get_bias([1])
y = tf.matmul(y1, w2)+b2

loss_mse = tf.reduce_mean(tf.square(y-y_))
loss_total = loss_mse + tf.add_n(tf.get_collection('losses'))

   loss_v = sess.run(loss_total, feed_dict={x:X,y_:Y_})

#coding:utf-8
#0导入模块 ，生成模拟数据集
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
BATCH_SIZE = 30
seed = 2
#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列的矩阵中取出一行,判断如果两个坐标的平方和小于2，给Y赋值1，其余赋值0
#作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]
#遍历Y中的每个元素，1赋值'red'其余赋值'blue'，这样可视化显示时人可以直观区分
Y_c = [['red' if y else 'blue'] for y in Y_]
#对数据集X和标签Y进行shape整理，第一个元素为-1表示，随第二个参数计算得到，第二个元素表示多少列，把X整理为n行2列，把Y整理为n行1列
X = np.vstack(X).reshape(-1,2)
Y_ = np.vstack(Y_).reshape(-1,1)
print X
print Y_
print Y_c
#用plt.scatter画出数据集X各行中第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color的缩写）
plt.scatter(X[:,0], X[:,1], c=np.squeeze(Y_c))
plt.show()

#定义神经网络的输入、参数和输出，定义前向传播过程
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

def get_bias(shape): 
    b = tf.Variable(tf.constant(0.01, shape=shape))
    return b
 
x = tf.placeholder(tf.float32, shape=(None, 2))
y_ = tf.placeholder(tf.float32, shape=(None, 1))

w1 = get_weight([2,11], 0.01) 
b1 = get_bias([11])
y1 = tf.nn.relu(tf.matmul(x, w1)+b1)

w2 = get_weight([11,1], 0.01)
b2 = get_bias([1])
y = tf.matmul(y1, w2)+b2

#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：不含正则化
train_step = tf.train.AdamOptimizer(0.0001).minimize(loss_mse)

with tf.Session() as sess:
 init_op = tf.global_variables_initializer()
 sess.run(init_op)
 STEPS = 40000
 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_mse_v = sess.run(loss_mse, feed_dict={x:X, y_:Y_})
   print("After %d steps, loss is: %f" %(i, loss_mse_v))
    #xx在-3到3之间以步长为0.01，yy在-3到3之间以步长0.01,生成二维网格坐标点
 xx, yy = np.mgrid[-3:3:.01, -3:3:.01]
 #将xx , yy拉直，并合并成一个2列的矩阵，得到一个网格坐标点的集合
 grid = np.c_[xx.ravel(), yy.ravel()]
 #将网格坐标点喂入神经网络 ，probs为输出
 probs = sess.run(y, feed_dict={x:grid})
 #probs的shape调整成xx的样子
 probs = probs.reshape(xx.shape)
 print "w1:\n",sess.run(w1)
 print "b1:\n",sess.run(b1)
 print "w2:\n",sess.run(w2) 
 print "b2:\n",sess.run(b2)

plt.scatter(X[:,0], X[:,1], c=np.squeeze(Y_c))
plt.contour(xx, yy, probs, levels=[.5])
plt.show()

#定义反向传播方法：包含正则化
train_step = tf.train.AdamOptimizer(0.0001).minimize(loss_total)

with tf.Session() as sess:
 init_op = tf.global_variables_initializer()
 sess.run(init_op)
 STEPS = 40000
 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 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)
 print "w1:\n",sess.run(w1)
 print "b1:\n",sess.run(b1)
 print "w2:\n",sess.run(w2)
 print "b2:\n",sess.run(b2)

plt.scatter(X[:,0], X[:,1], c=np.squeeze(Y_c))
plt.contour(xx, yy, probs, levels=[.5])
plt.show()