神经网络 (动态拟合函数)
前言
我用的是 tensorflow1.14 + spyder + python 3.5
因为这里涉及到了session,对于tf 2.0版本的并不支持
可以重新创建一个基于tf 1.14的虚拟环境
成果展示
最后成像是动态拟合展示,初始函数在神经网络中经过多次迭代,最终完成拟合。
流程
首先,导入三个模块
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
神经网络添加层
设置好偏重Weights, 偏差biases,构建Wx_plus_b存储没有被激活的值。等价 y = Wx + b
激活函数
tf.random_normal()函数用于从“服从指定正态分布的序列”中随机取出指定个数的值。
tf.random_normal(shape, mean=0.0, stddev=1.0, dtype=tf.float32, seed=None, name=None)
tf.zeros( shape, dtype=tf.float32, name=None) 构建 零张量
def add_layer(inputs, in_size,out_size, activation_function=None):
Weights = tf.Variable(tf.random_normal([in_size, out_size]))
biases = tf.Variable(tf.zeros([1, out_size]) + 0.1)
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
产生数据集 ,基本的二次函数 y = x^2+0.5
设定一些噪声点
x_data = np.linspace(-1, 1, 300)[:, np.newaxis]
noise = np.random.normal(0, 0.05, x_data.shape)
y_data = np.square(x_data) - 0.5 + noise
输入层 只有一个x_data,有一个data,只有一个属性,输入只有一个神经网络
构建隐藏层,假设隐藏层有十个神经元(1, 10)activation——function可以选择 sigmoid, relu, tanh等等。
定义一个输出层prediction(10, 1)输出是y_data,只有一个。
loss是判断估计和实际的差值
训练方式采用梯度下降法,步长为0.1。通常学习效率都是小于1的,夺取尝试。
l1 = add_layer(xs, 1, 10, activation_function = tf.nn.sigmoid)
prediction = add_layer(l1, 10, 1, activation_function = None)
loss = tf.reduce_mean(tf.reduce_sum(tf.square(ys - prediction),
reduction_indices=[1]))
train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)
初始变量,对Seeion对话中的事件进行执行( tf1 固定套路)
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
数据可视化,对当前的的图像进行处理
fig = plt.figure()
ax = fig.add_subplot(1,1,1)
ax.scatter(x_data, y_data)
plt.ion()
开始的placeholder == 占位符,后赋值。 和 feed_dict绑定
feed.dict == {字典}
这个模块进行了设定好计算展示频率的运行。
for i in range(1000):
sess.run(train_step, feed_dict = {xs: x_data, ys: y_data})
if i % 10 == 0:
try:
ax.lines.remove(lines[0])
except Exception:
pass
#print(sess.run(loss, feed_dict = {xs: x_data, ys: y_data}))
prediction_value = sess.run(prediction, feed_dict = {xs: x_data})
lines = ax.plot(x_data, prediction_value, 'r-', lw = 5)
plt.pause(0.1)
整体代码如下
import tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt
# 添加层
def add_layer(inputs, in_size,out_size, activation_function=None):
Weights = tf.Variable(tf.random_normal([in_size, out_size]))
biases = tf.Variable(tf.zeros([1, out_size]) + 0.1)
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
x_data = np.linspace(-1, 1, 300)[:, np.newaxis]
noise = np.random.normal(0, 0.05, x_data.shape)
y_data = np.square(x_data) - 0.5 + noise
xs = tf.placeholder(tf.float32, [None, 1])
ys = tf.placeholder(tf.float32, [None, 1])
l1 = add_layer(xs, 1, 10, activation_function = tf.nn.sigmoid)
prediction = add_layer(l1, 10, 1, activation_function = None)
loss = tf.reduce_mean(tf.reduce_sum(tf.square(ys - prediction),
reduction_indices=[1]))
train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)
init = tf.initialize_all_variables()
sess = tf.Session()
sess.run(init)
# 图像可视化
fig = plt.figure()
ax = fig.add_subplot(1,1,1)
ax.scatter(x_data, y_data)
plt.ion()
for i in range(1000):
sess.run(train_step, feed_dict = {xs: x_data, ys: y_data})
if i % 10 == 0:
try:
ax.lines.remove(lines[0])
except Exception:
pass
#print(sess.run(loss, feed_dict = {xs: x_data, ys: y_data}))
prediction_value = sess.run(prediction, feed_dict = {xs: x_data})
lines = ax.plot(x_data, prediction_value, 'r-', lw = 5)
plt.pause(0.1)