拟合正弦函数

神经网络模拟思路：
1 构建数据集 ：通过np.linspace(a,b,c)来构建一个从a到b中取c个数。再用np.reshape(a,(b,c))改变输入数据x的矩阵形状，a表示的是哪个数据被改变，(b,c)是将其改变成这样形状的矩阵

2 创建神经网络模型参数部分 ：首先定好参数，再将神经元中的权值和偏重全部定义好

3 进行训练 ： 调用生成函数，建立占位符（一般在train函数上生成占位符），调用函数梯度方法修正函数。

mport tensorflow as tf
import numpy as np
import matplotlib.pyplot as plt

def gen_data():
    x = np.linspace(0,2*np.pi,100)
    x = np.reshape(x,(len(x),1))
    y = np.sin(x)
    return x,y

# 创建神经网络模型
def inference(x):
    w1 = tf.Variable(tf.truncated_normal(shape=[INPUT_NODE,HIDDEN_NODE],stddev=0.1,dtype=tf.float32),name='w1')
    b1 = tf.Variable(tf.constant(0,dtype=tf.float32,shape=[HIDDEN_NODE]))
    a1 = tf.nn.sigmoid(tf.matmul(x,w1))
    w2 = tf.Variable(tf.truncated_normal(shape=[HIDDEN_NODE, OUTPUT_NODE], stddev=0.1, dtype=tf.float32), name='w2')
    b2 = tf.Variable(tf.constant(0, dtype=tf.float32, shape=[OUTPUT_NODE]))
    y_ = tf.matmul(a1,w2) + b2
    return y_


def train():
    train_x,train_y = gen_data()
    x = tf.placeholder(dtype=tf.float32, shape=[None, INPUT_NODE], name='input_x')
    y_ = tf.placeholder(dtype=tf.float32, shape=[None, OUTPUT_NODE], name='input_y')
    y = inference(x)
    loss = tf.reduce_mean(tf.square(y_ - y))
    train_step = tf.train.GradientDescentOptimizer(LEARNING_RATE).minimize(loss)
    np.random.seed(200)
    shuffle_index = np.random.permutation(train_x.shape[0])
    shuffled_X = train_x[shuffle_index]
    shuffled_y = train_y[shuffle_index]

    fig = plt.figure()
    ax = fig.add_subplot(1,1,1)
    ax.plot(train_x,train_y,lw=5,c='r')
    plt.ion()
    plt.show()
    with tf.Session() as sess:
        sess.run(tf.global_variables_initializer())
        for i in range(50000):
            feed_dic = {x:shuffled_X,y_:shuffled_y}
            _,l = sess.run([train_step,loss],feed_dict=feed_dic)
            if(i+1)%50 == 0:
                print("### loss on train: ", l)
                try:
                    ax.lines.remove(lines[0])
                except Exception:
                    pass
                y_pre = sess.run(y, feed_dict={x: train_x})
                lines = ax.plot(train_x, y_pre, c='black')
                plt.pause(0.1)



if __name__ == '__main__':
    # x = np.linspace(1,50)
    # x = np.reshape(x,(5,5,2))
    # print(x)

    INPUT_NODE = 1
    OUTPUT_NODE = 1
    HIDDEN_NODE = 50
    LEARNING_RATE = 0.1
    # a = np.random.permutation(x)
    # print(a)
    # print(x)
    train()

这里需要注意的是：
1 np.random.permutataion的用法：这个可以打乱数组的顺序，.shape[0]的意思是数据的index，意思是将index打乱。
2 占位符一般写在train()函数里面