用sklearn和tensorflow做boston房价的回归计算的比较(2)--卷积神经网路CNN

既然sklearn已经足够简单高效，为啥要用卷积神经网络(cnn)呢，江湖传言它有两个大优势：

1、sklearn需要人工进行特征优选，cnn会进行自动优选特征

2、随着训练数据的增多，sklearn的准确性就没啥大变化了，cnn则是越来越准，没有瓶颈

说实在的就boston房价这个数据也就506行，13个特征(列)，对cnn来说实在太少了，没个10万行数据，都看不出它的优势；

另外cnn虽然不用人工特征优选，但是搭建它的拓扑结构实在是个难搞的事，最让人炸裂的是tensorflow的结构，真是让人费解，关于它的结构网上很多介绍，我就不说了，但是用cnn做回归计算的文章非常罕见，请点赞！上代码

本文链接：http://blog.csdn.net/baixiaozhe/article/details/54409966

#参考http://blog.csdn.net/jerry81333/article/details/52979206 周莫烦的系列视频教程，跪地推荐
import numpy as np
from sklearn import preprocessing
import tensorflow as tf
from sklearn.datasets import load_boston
from sklearn.model_selection import train_test_split
#波士顿房价数据
boston=load_boston()
x=boston.data
y=boston.target
x_3=x[:,3:6]
x=np.column_stack([x,x_3])#随意给x增加了3列，x变为16列，可以reshape为4*4矩阵了 没啥用，就是凑个正方形

print('##################################################################')

# 随机挑选
train_x_disorder, test_x_disorder, train_y_disorder, test_y_disorder = train_test_split(x, y,
train_size=0.8, random_state=33)
#数据标准化
ss_x = preprocessing.StandardScaler()
train_x_disorder = ss_x.fit_transform(train_x_disorder)
test_x_disorder = ss_x.transform(test_x_disorder)

ss_y = preprocessing.StandardScaler()
train_y_disorder = ss_y.fit_transform(train_y_disorder.reshape(-1, 1))
test_y_disorder=ss_y.transform(test_y_disorder.reshape(-1, 1))

#准确率计算
# def compute_accuracy(v_xs, v_ys):
#     global prediction
#     y_pre = sess.run(prediction, feed_dict={xs: v_xs, keep_prob: 1})
#     correct_prediction = tf.equal(tf.argmax(y_pre,1), tf.argmax(v_ys,1))
#     accuracy = tf.reduce_mean(tf.cast(correct_prediction, tf.float32))
#     result = sess.run(accuracy, feed_dict={xs: v_xs, ys: v_ys, keep_prob: 1})
#     return result

#变厚矩阵
def weight_variable(shape):
initial = tf.truncated_normal(shape, stddev=0.1)
return tf.Variable(initial)
#偏置
def bias_variable(shape):
initial = tf.constant(0.1, shape=shape)
return tf.Variable(initial)
#卷积处理 变厚过程
def conv2d(x, W):
# stride [1, x_movement, y_movement, 1] x_movement、y_movement就是步长
# Must have strides[0] = strides[3] = 1 padding='SAME'表示卷积后长宽不变
return tf.nn.conv2d(x, W, strides=[1, 1, 1, 1], padding='SAME')
#pool 长宽缩小一倍
def max_pool_2x2(x):
# stride [1, x_movement, y_movement, 1]
return tf.nn.max_pool(x, ksize=[1,2,2,1], strides=[1,2,2,1], padding='SAME')

# define placeholder for inputs to network
xs = tf.placeholder(tf.float32, [None, 16]) #原始数据的维度：16
ys = tf.placeholder(tf.float32, [None, 1])#输出数据为维度：1

keep_prob = tf.placeholder(tf.float32)#dropout的比例

x_image = tf.reshape(xs, [-1, 4, 4, 1])#原始数据16变成二维图片4*4
## conv1 layer ##第一卷积层
W_conv1 = weight_variable([2,2, 1,32]) # patch 2x2, in size 1, out size 32,每个像素变成32个像素，就是变厚的过程
b_conv1 = bias_variable([32])
h_conv1 = tf.nn.relu(conv2d(x_image, W_conv1) + b_conv1) # output size 2x2x32，长宽不变，高度为32的三维图像
#h_pool1 = max_pool_2x2(h_conv1)     # output size 2x2x32 长宽缩小一倍

## conv2 layer ##第二卷积层
W_conv2 = weight_variable([2,2, 32, 64]) # patch 2x2, in size 32, out size 64
b_conv2 = bias_variable([64])
h_conv2 = tf.nn.relu(conv2d(h_conv1, W_conv2) + b_conv2) #输入第一层的处理结果 输出shape 4*4*64

## fc1 layer ##  full connection 全连接层
W_fc1 = weight_variable([4*4*64, 512])#4x4 ，高度为64的三维图片，然后把它拉成512长的一维数组
b_fc1 = bias_variable([512])

h_pool2_flat = tf.reshape(h_conv2, [-1, 4*4*64])#把4*4，高度为64的三维图片拉成一维数组 降维处理
h_fc1 = tf.nn.relu(tf.matmul(h_pool2_flat, W_fc1) + b_fc1)
h_fc1_drop = tf.nn.dropout(h_fc1, keep_prob)#把数组中扔掉比例为keep_prob的元素
## fc2 layer ## full connection
W_fc2 = weight_variable([512, 1])#512长的一维数组压缩为长度为1的数组
b_fc2 = bias_variable([1])#偏置
#最后的计算结果
prediction =  tf.matmul(h_fc1_drop, W_fc2) + b_fc2
#prediction = tf.nn.relu(tf.matmul(h_fc1_drop, W_fc2) + b_fc2)
# 计算 predition与y 差距 所用方法很简单就是用 suare()平方,sum()求和,mean()平均值
cross_entropy = tf.reduce_mean(tf.reduce_sum(tf.square(ys - prediction), reduction_indices=[1]))
# 0.01学习效率,minimize(loss)减小loss误差
train_step = tf.train.AdamOptimizer(0.01).minimize(cross_entropy)

sess = tf.Session()
# important step
# tf.initialize_all_variables() no long valid from
# 2017-03-02 if using tensorflow >= 0.12
sess.run(tf.global_variables_initializer())
#训练500次
for i in range(200):
sess.run(train_step, feed_dict={xs: train_x_disorder, ys: train_y_disorder, keep_prob: 0.7})
print(i,'误差=',sess.run(cross_entropy, feed_dict={xs: train_x_disorder, ys: train_y_disorder, keep_prob: 1.0}))  # 输出loss值

# 可视化
prediction_value = sess.run(prediction, feed_dict={xs: test_x_disorder, ys: test_y_disorder, keep_prob: 1.0})
###画图###########################################################################
import matplotlib.pyplot as plt
fig = plt.figure(figsize=(20, 3))  # dpi参数指定绘图对象的分辨率，即每英寸多少个像素，缺省值为80
axes = fig.add_subplot(1, 1, 1)
line1,=axes.plot(range(len(prediction_value)), prediction_value, 'b--',label='cnn',linewidth=2)
#line2,=axes.plot(range(len(gbr_pridict)), gbr_pridict, 'r--',label='优选参数')
line3,=axes.plot(range(len(test_y_disorder)), test_y_disorder, 'g',label='实际')

axes.grid()
fig.tight_layout()
#plt.legend(handles=[line1, line2,line3])
plt.legend(handles=[line1,  line3])
plt.title('卷积神经网络')
plt.show()

结果是这样的：



上文中只训练了200次，其实正常来说都是1000次起的，无奈手里只有小mac mini，显卡是N卡的同学可以用tensorflow的gpu版跑跑试试