参考:
《TensorFlow技术解析与实战》
http://blog.csdn.net/qq_16949707/article/details/53126707?locationNum=3&fps=1
我们使用第一个Tensorflow程序类似于HelloWord,用来测试安装环境,同时也可以使同学们直观的感受一下Tensorflow的魅力:我们构造一个满足一元二次函数 y = ax2 + b 的原始数据,然后构建一个最简单的神经网络,仅包含一个输入层,一个隐藏层和一个输出层,通过Tensorflow将将隐藏层和输出层的weitghts和biases的值学习出来,看看随着训练次数的增加,损失值是不是不断减少。
# -*- coding:utf-8 -*-
import tensorflow as tf
import numpy as np
#构造满足一元二次方程的函数
#构建300个点,分布在-1~+1区间,直接采用np生成等差数列,并且将一维转换成300*1的二维
x_data = np.linspace(-1, 1, 300)[:, np.newaxis]
#print(x_data)
#加入一些噪声,均值为0,方差为0.05
noise = np.random.normal(0, 0.05, x_data.shape)
#print(noise)
#y = 2*x^2 - 0.5
y_data = 2 * np.square(x_data) - 0.5 + noise
#定义x和y的占位符来作为将要输入神经网络的变量
xs = tf.placeholder(tf.float32, [None, 1])
ys = tf.placeholder(tf.float32, [None, 1])
#定义添加层的函数
def add_layer(inputs, in_size, out_size, activation_function=None):
#构架权重
weights = tf.Variable(tf.random_normal([in_size, out_size]))
#构建偏置
biases = tf.Variable(tf.zeros([1, out_size]) + 0.1)
#矩阵乘积
Wx_plus_b = tf.matmul(inputs, weights) + biases
if activation_function is None:
outputs = Wx_plus_b
else:
outputs = activation_function(Wx_plus_b)
return outputs
#构造隐藏层
h1 = add_layer(xs, 1, 20, activation_function=tf.nn.relu)
#构造输出层
prediction = add_layer(h1, 20, 1, activation_function=None)
#计算预测值和真实值之间的误差
loss = tf.reduce_mean(tf.reduce_sum(tf.square(ys - prediction), reduction_indices=[1]))
#梯度下降
train_step = tf.train.GradientDescentOptimizer(0.05).minimize(loss)
#训练并输出
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
#训练10000次
for i in range(1000000):
sess.run(train_step, feed_dict={xs: x_data, ys: y_data})
if i % 20000 == 0:
print(sess.run(loss, feed_dict={xs: x_data, ys: y_data}))
print("Finish!")
代码的功能解释都在Python得 注释中。
###################################################################################拟合未知曲线:
import numpy as np
x = np.random.rand(100).astype(np.float32)
y = 3*x*x+1
import matplotlib.pyplot as plt
plt.plot(x,y,'ro')
plt.legend()
plt.show()
import tensorflow as tf
w1 = tf.Variable(tf.random_uniform([1],-1.0,1.0))
w2 = tf.Variable(tf.random_uniform([1],-1.0,1.0))
w3 = tf.Variable(tf.random_uniform([1],-1.0,1.0))
b = tf.Variable(tf.zeros([1]))
y_ = w1*x+w2*x*x+w3*x*x*x+b
loss = tf.reduce_mean(tf.square(y - y_))
optimizer = tf.train.GradientDescentOptimizer(0.5)
train = optimizer.minimize(loss)
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
for step in xrange(100):
sess.run(train)
print step, sess.run(w1),sess.run(w2),sess.run(w3),sess.run(b)
plt.plot(x,y,'ro')
y__ = sess.run(w1)*x+sess.run(w2)*x*x+sess.run(w3)*x*x*x+sess.run(b)
plt.plot(x,y__)
# plt.plot(x,0.37588856*x+2.09157467*x*x+0.59096539*x*x*x+0.9672721)
plt.legend()
plt.show()