python的matplotlib模块实现tensorflow结果可视化。
'''
# tensorflow输出结果可视化(三层全连接神经网络)
# python3.6.1/tensorflow1.2.1
'''
# coding=utf-8
# 导入模块
import tensorflow as tf
import numpy as np#
import matplotlib.pyplot as plt # python数据图形化工具
# 添加神经网络层的函数
def add_layer(inputs, in_size, out_size, activation_function=None):
# 权值
Weights = tf.Variable(tf.random_normal([in_size, out_size]))
# 偏置项
biases = tf.Variable(tf.zeros([1, out_size]) + 0.1)
# 加权和
Wx_plus_b = tf.matmul(inputs, Weights) + biases
# 激活函数
if activation_function is None:
outputs = Wx_plus_b
else:
outputs = activation_function(Wx_plus_b)
return outputs
#=== 生成真实数据
# 输入数据
x_data = np.linspace(-1, 1, 300)[:, np.newaxis]# 300个从-1~1之间等差的行向量
# 噪点
noise = np.random.normal(0, 0.05, x_data.shape)
# 输出数据
y_data = np.square(x_data) - 0.5 + noise
#plt.scatter(x_data, y_data)
#plt.show()
#=== 生成占位符
xs = tf.placeholder(tf.float32, [None, 1])
ys = tf.placeholder(tf.float32, [None, 1])
#=== add hidden layer 隐含层
l1 = add_layer(xs, 1, 10, activation_function=tf.nn.relu)
#=== add output layer 输出层
prediction = add_layer(l1, 10, 1, activation_function=None)
#=== 损失函数和梯度下降
loss = tf.reduce_mean(tf.reduce_sum(tf.square(ys-prediction), reduction_indices=[1]))
train_step = tf.train.GradientDescentOptimizer(0.1).minimize(loss)
#=== 生成会话和初始化变量
init = tf.global_variables_initializer()
sess = tf.Session()
sess.run(init)
#tensorflow >= 0.12 的版本不支持
# tf.initialize_all_variables()初始化向量}
#=== plot真是数据
fig = plt.figure()# 生成图片框
ax = fig.add_subplot(1,1,1)# 图片框分割成一块,编号为111
ax.scatter(x_data, y_data)# 以点的形式显示出真实数据
#plt.ion()# 连续的打印
#plt.show()# 输出图片
#=== 训练神经挖网络
for i in range(1000):
sess.run(train_step, feed_dict={xs: x_data, ys: y_data})
if i % 50 == 0:# 每50步输出一次
try:
ax.lines.remove(lines[0])# 删除lines中第一条数据
except Exception:
pass
prediction_value =
sess.run(prediction, feed_dict={xs: x_data})
lines =
ax.plot(x_data, prediction_value, 'r-', lw=5)
# 以线的形式输出预测数据,线的颜色为红色,线宽为5