[吃药深度学习随笔] 正则化

过拟合与正则化

正则化计算方法

#Tensorflow代码
loss(w) = tf.contrib.layers.l1_regularizer(REGULARIZER)(w)

#Tensorflow代码
loss(w) = tf.contrib.layers.l2_regularizer(REGULARIZER)(w)

tf.add_to_collection('losses', tf.contrib.layers.l2_regularizer(regularizer)(w)
loss = cem + tf.add_n(tf.get_collection('losses'))

matplotlib可视化工具模块使用方法：

例子：

#coding:utf-8
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

BATCH_SIZE = 30
SEED = 2333
#基于Seed参数随机数
rdm = np.random.RandomState(SEED)
#随机数返回300行2列的矩阵 表示300组坐标点（x,y)
X = rdm.randn(300,2)
#从X中取一行，若其x^2+y^2<2 则Y=1 反之Y=0
#作为标签的正确答案
Y_ = [int(x0*x0 + x1*x1 <2) for(x0,x1) in X]
#遍历Y中元素，1赋值红色，0赋值蓝色
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0列元素和1列元素
#用各行Y_c的值表示颜色
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_biase(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_biase([11])
y1 = tf.nn.relu(tf.matmul(x, w1)+b1)

w2 = get_weight([11,1], 0.01)
b2 = get_biase([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'))


#定义反向传播函数: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_mse_v = sess.run(loss_mse, feed_dict={x:X, y_:Y_})
            print("经过%d步后，loss：%f" %(i, loss_mse_v))
    #xx在-3到3之间以步长为0.01, yy在-3到3之间以步长为0.01，生成二维网格坐标点
    xx, yy = np.mgrid[-3:3:.01, -3:3:.01]
    #讲xx,yy拉直，并合并成一个2列矩阵，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调整为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()