# -*- coding:utf-8 -*-
import tensorflow as tf

import numpy as np

import matplotlib.pyplot as plt

#训练数据

x_data = np.linspace(-6.0,6.0,30)[:,np.newaxis]

y_data = np.power(x_data,3) + 0.7

#验证数据

t_data = np.linspace(-20.0,20.0,40)[:,np.newaxis]

ty_data = np.power(t_data,3) + 0.7

#占位符

x = tf.placeholder(tf.float32,[None,1])

y = tf.placeholder(tf.float32,[None,1])

#network

#--layer one--

l_w_1 = tf.Variable(tf.random_normal([1,10]))

l_b_1 = tf.Variable(tf.zeros([1,10]))

l_fcn_1 = tf.matmul(x, l_w_1) + l_b_1

relu_1  = tf.nn.relu(l_fcn_1)

#---layer two----

l_w_2 = tf.Variable(tf.random_normal([10,20]))

l_b_2 = tf.Variable(tf.zeros([1,20]))

l_fcn_2 = tf.matmul(relu_1, l_w_2) + l_b_2

relu_2  = tf.nn.relu(l_fcn_2)

#---output---

l_w_3 = tf.Variable(tf.random_normal([20,1]))

l_b_3 = tf.Variable(tf.zeros([1,1]))

l_fcn_3 = tf.matmul(relu_2, l_w_3) + l_b_3

#relu_3  = tf.tanh(l_fcn_3)

# init

init = tf.global_variables_initializer()

#定义 loss func

loss = tf.reduce_mean(tf.square(y-l_fcn_3))

learn_rate =0.001

train_step = tf.train.GradientDescentOptimizer(learn_rate).minimize(loss)

with tf.Session() as sess:

    sess.run(init);

    for epoch in range(20):

        for step in range(5000):

            sess.run(train_step,feed_dict={x:x_data,y:y_data})

    y_pred = sess.run(l_fcn_3,feed_dict={x:t_data})

    print sess.run(l_fcn_3,feed_dict={x:[[10.]]})

    plt.figure()

    plt.scatter(t_data,ty_data)

    plt.plot(t_data,y_pred,'r-')

    plt.show()